James Maconochie | Architecture & Attention Podcast

I have told this story before. Six weeks ago I told it to make a point about software. I want to tell it again to make a point about myself, because I think I put the emphasis in the wrong place.In 1993, I was a final-year civil engineering student at Imperial College, and my design project was an exhibition hall with a column-free interior and a long-span roof carried on a pair of massive tubular arch beams. I modeled it in I-DEAS, a finite element package running on a Silicon Graphics workstation that took up half a desk. The package computed the stresses, checked them against code, and rendered a deformed shape that looked entirely reasonable. It said the structure would stand.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.The night before the report was due, for reasons I still cannot reconstruct, I did a hand calculation of the stress at one of the arch supports.It came back off by a factor of ten.I had entered the loads in kilograms, but the package read them as newtons. My model was carrying about a tenth of what the real building would have to carry. When I corrected the units and re-ran it, the deformed shape came back flat as a pancake. The structure I had designed would not have stood up.Last time, I said the software never pushed back; the analysis was fluent all the way down. The correction had to come from outside it. That is true, and it is the smaller half of the story.The larger half is the napkin.I had a way to check. It was crude; it took three minutes, and it was decisive. And I had it for a reason that had nothing to do with being careful. I had been taught to size a beam by hand before I was ever allowed near a package that would size it for me. The check wasn’t diligence. It was residue.The Word We Only Half UnpackedFive weeks ago I wrote about the word reward, and what happened to it when AI borrowed it from neuroscience.In the brain, reward is not one thing. Kent Berridge’s work separates it into two systems that can be pulled apart experimentally. Wanting is driven by mesolimbic dopamine; it makes you pursue the thing. Liking is a distinct hedonic system running on different neural substrates, and it registers how the thing actually landed. Disable one and the other carries on without it. An animal can be made to want, intensely, something it does not like at all.Machine learning borrowed the word and built the pursuit.I want to state this more carefully than I did last time, because it is easy to overclaim. Reinforcement learning gives a system an explicit optimization signal and a great deal of patience in following it. What it does not supply, by itself, is anything corresponding to the second system, an independent register of how the result actually landed, sitting outside the objective and capable of disagreeing with it. There are constraints, regularizers, safety layers, human feedback, evaluators. What’s missing is something that stands apart from the optimization and asks whether the objective was the right one.Drive without a brake, with the caveat that this is a metaphor, and a metaphor is not a schematic. The precise version is duller and harder to knock down: a system optimized to produce an answer is not thereby equipped to establish that the objective, or its reading of the objective, was correct.I left it there, as a fact about how these systems are built. It is a fact about how they are built.But nothing is dangerous all by itself. An uncovered hole in the ground is not dangerous sitting in an empty field. It becomes dangerous when someone walks near it. The question I skipped last time is who is walking near this one, and how close.Condition Is Not RiskEngineers separate two things that ordinary language runs together.Years after the exhibition hall, I spent part of my career in transportation consulting. One project, with colleagues at Cambridge Systematics and Lloyd's Register, built a risk model for the bridges on the US highway system. Every state reports the condition of its bridges to a national inventory, and we used that data to place each structure on a five-by-five grid.One axis was condition: how structurally sound the bridge actually is. The other was consequence: what happens if it fails, largely how much traffic crosses it, adjusted for how far the detour would be if it closed.Those two axes aren't the same thing, and keeping them apart is the point. A badly deteriorated bridge on a farm track is in poor condition and carries almost no consequence. A sound bridge carrying a hundred thousand vehicles a day is in fine condition and carries enormous consequence. Neither is where you look first. You look at the corner where both are bad at once: a deteriorating structure carrying heavy traffic with no easy detour. Risk is what you get when condition and consequence coincide.The structure's condition is a fact about the structure. It is true whether anyone drives across it or not. The risk exists only because people do.That is the distinction I skipped past five weeks ago. An optimizer without an independent evaluator is a fact about how the thing is built. It is a condition. It is true of the system sitting on its own, doing nothing at all.What turns a condition into a risk is traffic.Two Ways To Hand Something OverSo: who is driving over this bridge, and how heavily?Not “who uses AI.” Everyone does, and that tells you nothing. The question is narrower. What happens to your own ability to check the answer?Handing work to something else is not one thing. It comes in two forms that look identical from outside and behave very differently when something goes wrong.The first is delegation. You give away the task and keep the ability to judge what comes back. A managing editor does not write the article, but can read it and say it isn’t working. I gave the stress calculations to I-DEAS and kept, without ever thinking about it, a way to test the answer. The work moved. The judgment stayed.The second is offloading. You give away the task, and the ability to judge the result goes with it. What comes back is a conclusion, arriving without the reasoning that produced it, and there is nothing left on your side that can weigh it. You can accept it or refuse it on instinct. You cannot check it.From outside, these look the same. A person asks, a system answers, the person proceeds. The difference is invisible until the answer is wrong.The obvious objection is that delegation is just offloading on a longer timescale, that if you never do the work yourself, the ability to judge it must eventually go.That objection is correct, and it is the most important thing in this essay. Delegation is not a stable condition. It decays into offloading unless something actively holds it in place. This raises the question of what could hold it there, and whether anything does.Bainbridge, 1983The obvious answer is that this is a matter of discipline. Keep your hand in. Check the work.Lisanne Bainbridge answered that in 1983, in a paper on industrial process control written long before any of this existed.Her observation came to be called the ironies of automation. When you automate a process, you leave a human to monitor it and take over if it fails. But monitoring something that runs smoothly is exactly the condition under which manual skill decays. The operator does less, for longer, and the ability to intervene erodes quietly while nothing appears to be wrong.Then the automation fails, which is the moment it needs a person, and the person it gets is the one whose skills have been eroding for years, worn down by the very smoothness that made everything look fine.The skill you need at the moment of failure is precisely the skill that the absence of failure has been taking away.So the decay isn't hypothetical, and vigilance isn't the answer. Which leaves the question of what is.What The Napkin Actually WasHere is where I have to be careful, because the obvious reading of my own story is wrong.The obvious reading is: hold on to the old skill. Learn to do it by hand so you can always do it by hand. And that cannot be the lesson, because it does not survive contact with how anything works. I cannot fabricate a semiconductor, sequence a genome, or derive the cryptography my bank runs on. I use all three every day, and I should. If the answer to automation were personally retaining the underlying method, the answer would be useless.Look again at what I actually had that night.It was not the ability to reproduce what I-DEAS did. The package solved a system of equations across thousands of elements. I could not have done that by hand in a month, and nothing in my training suggested I should try. What I had was cruder and far more valuable: a way to arrive at an approximate answer by a route that did not pass through the package.That is the whole thing. Not reproduction — verification by an independent route.Which gives the test:Is there a way to check the output that does not run through the thing that produced it?Sometimes that route is retained expertise, as mine was. But it can be many other things. An empirical test the world adjudicates. A formal constraint the output must satisfy. A second system built on genuinely different principles rather than a near-copy of the first. An audit trail that lets someone reconstruct how a conclusion was reached. A deliberately adversarial process staffed by people whose interests differ.The napkin is one instance of the category. It is not the category.And that changes what is at stake. The scarce resource is not craftsmanship, and this is not an argument about the dignity of doing things by hand. What is scarce is independent routes, and unlike a bridge, a route can disappear without anyone noticing.There is a collective version of this, and it is worse.My napkin worked because it was a different method. Different assumptions, a different path to the answer, and this is the part that matters: different ways of being wrong. Independence is not a property of having a second opinion. It is a property of the second opinion failing differently from the first.Which means a field can hold on to checking as a habit and still lose it as a safeguard. If everyone verifies with the same tools, calibrated against the same standards, trained on the same material, then a thousand checks are one check performed a thousand times. The errors line up. And nobody notices, because everybody is still checking.A Case Where The Route Has Already ClosedThat risks staying abstract, so here is a live one.A student submits an essay. A detector reports that it was machine-written. The student is now the subject of a judgment with real consequences: a failed assignment, an academic misconduct process, occasionally the end of a degree.What check is available?The student can deny it, which is an assertion, not evidence. Or the work can be run through a second detector: a system of the same class, trained on broadly the same signals, sharing the same blind spots and making correlated errors. That is not an independent route. That is the first opinion in a different font.Sam Illingworth reached the image before I did, writing about Substack’s own detector: an AI adjudicating whether text is AI. A turtle standing on a turtle.Look at where this sits on the grid. The condition is the familiar one, a system producing a confident verdict with nothing in it that stands outside the verdict and disagrees. The consequence is high and lands on a particular person. And the independent route is not merely weak; for the accused student it is very nearly absent, because every available check fails in the same direction as the accusation.Then the third thing, which is the worst of it. Nobody assembles the failures. A wrongly accused student who cannot prove a negative mostly disappears from the record; the case is dropped, or it isn’t, and either way it does not become a data point. There is no inquiry, no docket, no count. The bridge falls down, and nobody hears it.This is not a warning about something that might happen. It is running now, at scale, in institutions that adopted it faster than they built any way to evaluate it.Where The Traffic Is HeaviestNow scale this up, carefully, because the obvious version of this argument is wrong.Every generation loses skills to its tools. My grandfather could do long division faster than I ever could. Nobody navigates by the stars. Slide rules are museum pieces. Those losses were real, and almost none of them mattered.So the claim cannot be that skills are being lost. Skills are always being lost. The claim has to be narrower: what matters is whether an independent route survives the loss.When calculators replaced slide rules, one usually did. Order-of-magnitude estimation is cheap, teachable, and doesn't run through the calculator, which is why someone who kept it can still catch a wrong answer. Not everyone did, and that is precisely the point: the route existed, and its survival depended on whether anyone bothered to maintain it. What did not happen is the route becoming unavailable in principle.That is the real question about any tool. Not what it does for you, but what it leaves you.Now take the work where the output is a judgment rather than a quantity. Whether this candidate should be hired. Whether these symptoms fit together. Whether this clause creates an exposure. Whether this strategy is sound.It would be too strong to say these fields have no external check. Medicine has pathology, imaging, and outcomes. Law has evidence, precedent, and procedure. Hiring has job performance. Strategy has results. What they lack is my beam, a check that is fast, cheap, and mechanically decisive. Their checks are slow, noisy, expensive, contestable, and often arrive years after the decision they would have corrected. Structural engineering sits at one end of that spectrum, with physics as an unusually brutal validator. These fields sit at the other.A reader will say: those fields never had a napkin, so nothing is being lost there. You have diagnosed a problem in engineering and applied it by analogy to places it does not belong.That is right, and it is the more uncomfortable version of the argument. They were always at the far end of the spectrum, running on checks that were slow and arguable. They lived with it because the volume of judgment passing through them was bounded by how many trained people could be put in the room. What has changed is not the check. It is the traffic.Put those fields on the bridge grid, and the relevant condition is not identical across them: a diagnostic system, a hiring model, and a drafting assistant are not the same architecture, but it is analogous, and it is analogous in the way that matters: the recipient of the output frequently has no route to evaluate it that does not run back through a system of the same kind.The bridge model had it easier in one respect, and it is worth saying so. A bridge that fails announces itself. Nobody has to detect the collapse. A hiring decision, a missed diagnosis, a badly drafted clause fail quietly and can keep failing for years without anyone assembling the fact that they failed at all.That is not a third axis I want to add to the grid. It is a reason the grid flatters the situation.And here I will make a prediction rather than a statement of fact, and mark it as one, so it can be held against me later. I think the independent routes in these fields will thin faster than they are replaced, because keeping one costs time and money now and pays only in failures that never happen. That is the hardest kind of expenditure to defend and the first kind to be cut.The Question That Decides ItWhich turns the usual question inside out.The question almost always asked about these systems is how good they are. Can it pass the exam, write the brief, read the scan? That question has an answer, and the answer improves every few months. It is a fair question. But it is a question about condition, and condition is not risk.The question that decides the outcome is about the arrangement we put the system into. When the work comes back, is there any route to judging it that does not run through the thing that produced it?That is a design question, and design questions have answers. That doesn't mean hoping people stay vigilant; Bainbridge disposed of that in 1983. Tonie Marie Gordon, who works on cognitive ergonomics, puts the same point in its sharpest form: “The presence of a human in the loop is not oversight. The design of that human’s cognitive role is what determines whether the oversight is meaningful or decorative.” A person positioned where they cannot check anything is not a safeguard. They are a diagram.That is what I mean by augmented human intelligence, and it is duller than the phrase suggests: designing the arrangement so that human judgment retains an independent route to the answer, rather than being placed downstream of a system it can only accept or refuse.The obvious objection is economic, and I cannot dispose of it. Maintaining an independent route costs time and money, and an organization that pays for one competes against organizations that don’t. Market pressure runs steadily against every recommendation in this essay. Building codes exist because that argument was lost on the merits and had to be won by regulation instead, in exactly the field I started in, and only after enough structures had fallen down to make the cost visible. I do not know what the equivalent looks like here, or what it will take.And there is a harder version of the objection that I am not sure I can answer at all.Bainbridge called the erosion an irony, which implies that nobody wanted it. In some arrangements it is not an irony. A workforce that cannot check the system is cheaper, more interchangeable and easier to replace than one that can. Where that is true, the missing route is not an unfortunate side effect of efficiency. It is the efficiency. And no amount of thoughtful design will supply a check that somebody is deliberately economizing on.I would like to end by saying I caught that error because I was careful. I wasn't. I was a student the night before a deadline, doing an arithmetic exercise for no reason I could ever reconstruct.I had three minutes and a route that didn't go through the package.It would have let me submit. It couldn't tell me anything was wrong, didn't try, and would have gone on, fluent and confident, about a building that would have fallen down. Not because it was badly built, because nothing in a system that produces an answer is positioned to tell you the answer is wrong. That has to come from somewhere else.Whether anything is still standing in that somewhere else is the only question that matters.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

A scientist finishes a lecture on the solar system. An elderly woman in the audience stands up and tells him the whole account is wrong: the world rests on the back of a giant turtle. The scientist, politely, asks what the turtle is standing on. She has the answer ready. It is turtles all the way down.The story is usually told about Bertrand Russell. It is also told about William James, and about various astronomers whose names change with the telling. Each version reaches us through someone who did not say where they got it. Nobody has established which lecture hall, or which century, or whether it happened at all. The provenance of the story about infinite regress has no bottom either.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.The Number I Was Holding UpFor several years I worked for a consulting firm that ran a Delivery Excellence dashboard. Every project reported weekly against a set of components: scope, schedule, budget, client satisfaction, team performance, and one or two others I no longer remember. Each one got a color. I entered the scores myself.On a program for a biotech client, I reported client satisfaction as green, week after week, and the score was accurate. The client contact was satisfied. Every concern he raised got addressed in the order it arrived. He was told early that the scope began to shrink because we could not deliver all of it, and hearing it early is most of what makes a shrinking scope tolerable.Underneath that, a team was struggling on several counts, and had been almost from the start.So the numbers were true. Scope was green because we had already cut scope. Satisfaction was green because I spent a substantial part of every week producing it. Every component reported a state, and none had a field for what that state cost to maintain. The effort was the risk, and the effort was the one thing the format could not carry.I did not falsify anything. I reported what the components asked for, and I raised the delivery problems internally, every week, in the space the form left for them. The technical lead and the others read the situation as managed, and I did not want to go out on a limb by myself against my own team’s assessment. I also thought that if we all knew, we would fix it before it came due. So a room full of people who each knew something was wrong went on filling in a form that had nowhere to put it. I have watched a version of this happen more than once, and never once because anyone was hiding anything.What eventually surfaced the problem was not the dashboard. In January, I sat down with the office general manager and argued that we should put three hundred and fifty thousand dollars of our own money into a three-and-a-half-million-dollar job to fix what we had not delivered. The technical lead and two others backed the position. The technical lead told him the only reason we were still in the room was the account management. He agreed to the money. We spent it carefully, the work landed, and the client was pleased in the end.The dashboard recorded none of this. It had been green throughout, and it was green because of what I was doing underneath it.The Tower We Are Building NowThe dashboard failed in a specific way, and the way matters more than the failure. It published an accurate number and gave nobody access to what produced it. Whoever read the green had to take on trust that the process behind it was sound, and the process behind it was me. Reported figures come loose from the work that generated them. That is not a property of consulting. It is a property of reporting.A model’s output is evaluated by another model, because human evaluation does not scale. The evaluating model was trained on data that a model filtered for quality. Both are scored against a benchmark, and benchmark content has repeatedly been found in the training data of the systems being scored. The results are then summarized for the people who have to decide something, increasingly by a model, because the volume is large and the summary is fast.Every one of those steps is defensible on its own. Each is documented, published, and argued for by people who know more about it than I do. None of them is hiding anything.What is missing is not honesty. It is anyone whose job is to ask what the bottom layer is standing on. Each layer treats the one below it as ground, and the question of grounding gets handed down the stack until it runs out.In Defense of TurtlesI want to concede this properly rather than in passing, because the concession is most of the argument.Abstraction is how anything large gets built. Nobody writing an application re-derives the behavior of a transistor. Proxies are how you manage what you cannot measure directly, and almost everything worth managing cannot be measured directly. I have never inspected a bridge before driving over it, and I am not going to start.Stacking is not the sin. A working tower of abstractions is one of the more impressive things people do. The engineer who trusts the steel certificate is not being lazy. That certificate is doing real work, and behind it there is a mill, a test, a standard, and a name.The failure is narrower than the stacking. It is losing track of whether there is a bottom at all. Deferring the question of grounding to the layer below is fine, and unavoidable, as long as the deferral terminates somewhere. The tower becomes a problem only when every layer in it has been built on the assumption that somebody further down already checked.Where It StopsThe regress has to stop somewhere, and the only place I have ever seen it stop is at someone who pays if it is wrong and cannot hand the paying to anyone else.I want to be careful about how much that carries. It is not a guarantee of anything. People with their names on things sign off on work they have not understood. Systems with clear lines of liability fail badly and often, and the person carrying the cost is frequently the last to see it coming. The engineer who signs the beam did not derive the metallurgy, watch the testing, or stand in the mill. Finding the name does not tell you the answer is right, and it does not mean the person can check the work. It tells you the chain has a bottom instead of continuing indefinitely. The claim is that narrow: without someone in that position, there is no bottom to have.I have spent most of this series on that claim, and I am not going to re-argue it here. The engineer who signs off on the beam carries it if the beam fails. The certificate is not the ground, and the signature is not the ground. The ground is that a person’s license, career, and name are attached to the load, and that the attachment cannot be undone by delegating.On the biotech program, the ground floor was not the dashboard. It was a room in January, a general manager who had to spend money he would answer for, and an account lead recommending it who would answer for it too. Neither of those costs could be passed sideways. That is why the problem got fixed. It is also why it took until January.What You Can UseThe woman in the story was wrong about cosmology and right about almost everything else.So here is the test. Pick any validation you currently rely on. A green status, a certification, a score, a rating, a benchmark result, a maturity level. Ask what it is standing on. Then ask the same question of the answer.Keep going until the answer stops being a process and starts being a name.If you work down through the whole stack and there is nobody there, you are standing on a turtle. It may well hold. A great many of them do, for years at a time. But you should at least know that is what you are standing on.I stood on one for most of a year, and I entered the number myself every week.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

There is a piece of work your body performs every waking second, and you will almost certainly never think about it, unless one day it stops.I injured my lumbar spine. What I learned afterward was not really about pain, which is the part everyone expects and the part I will spare you. It was about administration. Somewhere well below anything I could call thinking, my brain had been issuing a continuous stream of instructions to the small stabilizing muscles around my spine, adjusting them constantly, correcting for every shift of weight and every turn of my head, every hour I had been alive. Standing still turns out not to be still. It is a controlled fall, caught and re-caught, in every direction at once.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.I never chose that work. I never noticed it. I became aware of it only when it started arriving late and imprecise, and the ordinary business of getting up from a chair became something I had to attend to deliberately, the way you attend to a difficult sentence.Here is the part that took me longer to see. The work was not executing a standing instruction. Nothing in me was running an order that said stay upright. At every instant something was deciding which of an enormous number of signals mattered right now, and therefore what the problem currently was: a floor that turned out to be wet, a bag heavier on one side than expected, a step that arrived sooner than the last one. The instruction never changed. The problem changed constantly, and something reformulated it, continuously, without asking me.Nobody set that problem. Something I have never met noticed it and re-noticed it thousands of times an hour.The Ledger We Keep Of DifficultyHans Moravec noticed the consequence of this in 1988, in a book called Mind Children, and the observation now carries his name. It is comparatively easy, he wrote, to make computers perform at adult level on intelligence tests or at checkers, and difficult or impossible to give them the skills of a one-year-old at perception and mobility.The paradox is normally presented as a fact about machines. I want to present it as a fact about us.We rank difficulty by how hard a thing feels while we do it. Chess feels like intellect. It is effortful, available to introspection, and we can watch ourselves doing it, so it sits at the top of the ledger. Catching a set of keys tossed across a room feels like nothing at all, so it sits nowhere. The ranking is not a judgment about the problems. It is a report on which problems we happen to be conscious of solving.Let me be careful about how much weight that carries. I am not claiming effort measures unfinished learning. Expert musicians work hard at passages they have played for decades, vigilance is exhausting after thirty years of practice, and creative work stays effortful precisely because there is no stable routine to settle into. The claim is narrower: felt effort is an unreliable proxy for computational difficulty, and we have been using it as our main one.An objection deserves to arrive here rather than later. The computer scientist Arvind Narayanan has argued that Moravec’s paradox is not a dependable predictor of what will prove easy or hard for machines, and is better read as a statement about what the field has found worth working on. I think that is largely right, and I am not going to use the paradox as a forecast. Whatever it does or does not predict, it correctly identifies a bias in our accounting.That informal ledger has a formal descendant, and it now runs the industry. We call them benchmarks. A benchmark is our difficulty ledger written down and made scoreable, and it inherits the same property: it can only contain what somebody thought to put in it.The Objective Was Never the Hard PartLet me concede the thing that arguments of this kind usually avoid.Football has an exceptionally clean objective. Score more goals than the opposition inside the allotted time. You can write it on a napkin. There is no ambiguity about success, and a scoreboard settles it. If difficulty lived in the difficulty of stating an objective, football would be among the simplest things humans do.So the difficulty is not in the objective. It is in the distance between having one and knowing what to do next.Watch what that distance contains. Twenty-two people are running the stabilization work I described at the start of this essay, continuously, while sprinting, turning, being shoved and landing badly. Air moves across the stadium in ways nobody has modeled. The surface changes through the match. The light shifts. The altitude sets what their bodies can spend. And each player is reading the intentions of everyone else on the field while being read in return. A pass is a prediction about where a teammate will be in two seconds, made by a body in mid-fall, against defenders making their own predictions about that prediction, who will change what they do the moment they detect it.I want to be precise about what the players are doing, because there is a mystical version of this claim and I do not want it. They are solving the problem. Obviously they are; the ball arrives. What they cannot do is say how. The computation is distributed through the body, learned over years, and closed to introspection, which is different from absent.But the deepest part is not the policy either. It is that the problem does not hold still.Somewhere around the hour mark, a match becomes a different match. The referee has been letting contact go, so the cost of a challenge has changed. Someone is carrying a knock and is now a target. The other side has gone to three at the back, and every space on the pitch means something new. Nobody announces any of this. No player is told the problem has changed. Each one works it out while still playing, revises what matters, and acts on the revision before being able to articulate it.It is worth being exact about what changed, because adapting to a moving situation is something good control systems already do well, and I do not want to dress that up as something rarer than it is. What changed was not the difficulty. It was which information mattered to success. A defender’s body position meant one thing at twenty minutes and means something else now, and nothing in the original instruction said so, or could have.That is the line I want to hold throughout this essay. Adjusting your behavior because conditions moved is one thing. Discovering that a different set of facts has become the relevant ones is another, and it is the second that nobody set.The objective did not move. It was to score more goals in the first minute, and it is to score more goals in the ninetieth. What moved was everything about what that requires.Take Away The OpponentIt would be easy to conclude this all comes from the adversary, so it is worth removing the adversary and seeing what remains.Ballet has no opponent in the relevant sense. Nobody on that stage is trying to make a dancer fall.I should concede at once that she faces an evaluator. Two thousand people are watching and judging, and their judgment is opaque, plural and impossible to reduce to a number. That is real pressure, and I have not removed all of it. What I have removed is the thing football is usually assumed to be hard because of.And the difficulty does not drop.She is running the same continuous stabilization work at the limit of what a skeleton tolerates, on a floor with its own particular give, in time with music played live and therefore slightly differently tonight, alongside other dancers whose small variations she absorbs without appearing to absorb anything. And through it all, she is deciding what tonight’s performance is. Not the choreography, which is fixed and which she knew months ago. Suppose her partner arrives a fraction early and the lift no longer sits where the phrase wants it. She has perhaps half a second, and the choice is not how to execute the lift better. It is whether tonight the moment is the lift at all, or the recovery out of it, and she has to decide which one the audience is now watching before she can have any view about how to do it well. Nothing in the choreography addresses that, because the choreography could not have known her partner would be early.The choreography is the specification. It is complete, and it is not the performance.Before I go on, a correction to an impression these two examples will have created. I chose a World Cup final and a stage because they are legible, not because the capacity is rare. Nothing in the argument depends on excellence. The same continuous re-noticing runs in an aide steadying someone out of a chair, whose sense of what this particular person can bear today was never written into any care plan, and in a warehouse picker rerouting around an aisle that is blocked differently than it was an hour ago. The elite cases are the ones we buy tickets for. They are not the ones where the capacity lives, and if anything, the fact that we pay handsomely for the visible version and almost nothing for the ordinary one is its own small piece of evidence about the ledger.What A Measurement Can SeeTake drone light shows, which are the cleanest case available. Thousands of aircraft, physically present, holding formation in real wind. Every objection about digital versus physical dies on contact with a drone show.The lazy version of this argument is wrong, so I won't make it. Those aircraft are not blindly following fixed paths. Modern autonomous systems execute policies rather than trajectories: wind arrives, and the aircraft recomputes what to do to stay where it needs to be. That is real adaptation and real engineering.Notice what is nonetheless fixed. The image was decided before anything left the ground. Nothing in that system is built to ask whether the picture is still the right picture, or whether tonight calls for something other than what was planned. Adaptation of the route, thoroughly solved. Reformation of the problem, not attempted, and not required. The football video games and the Stanford autonomous helicopter sit in the same place for the same reason: one with authored physics and one with an objective inferred from expert demonstrations rather than written by hand, which means the specification arrived from a different direction and was no less complete.Here is what these cases have in common, and it is not that the systems are unimpressive. Each one was measured by something built to see a particular thing, and each one succeeded at that thing. The measurement did not fail. It reported honestly on what it was designed to report on, and said much less about anything else, which is what measurements do.The Machines That Actually PlayNow the honest part, because there is a serious effort aimed directly at this territory.RoboCup has been running since 1997 with an openly stated goal: a team of humanoid robots capable of beating the human World Cup champions by 2050. This summer, at the competition in Incheon, two full teams of humanoid robots played an eleven-a-side match on real hardware for the first time, with B-Human of Bremen beating HTWK Robots of Leipzig four to nil. In the tournament proper, which ran across three divisions this year, B-Human took the middle division, a Tsinghua University side took the large division, and a team from Wuhan University took the small one. Observers reported clear gains in stability, walking speed, and coordination.This is not choreography. Those robots are perceiving, deciding, and coordinating in real time against opponents doing the same. They are learning policies rather than following scripts.And that fact demolishes an argument I am not going to make. It would be easy to claim that because we cannot write down what a footballer does, no machine can acquire it. That claim is false, and the last fifteen years have falsified it repeatedly. Nobody wrote down a specification for recognizing a face or holding a gait, and systems learned both anyway. Humans never had a written specification either. We learned from interaction, which is what these robots are now doing. Being unable to describe a capability tells you nothing about whether it can be acquired.So the interesting question is not whether machines will get good at this. It is what their getting good at it entitles us to conclude.Because a match at Incheon is a measurement, and like every measurement, it was built to see certain things. It sees whether a robot can stand, perceive, pass, and score against opposition. Those are real capabilities, and it reports on them honestly. What it is not built to see is whether anything on that pitch could notice that the problem had changed. Nothing in the tournament asks that question, so nothing in the result answers it.And look at where the noticing is actually happening, because I find this the most interesting fact in the whole competition. For nearly thirty years the humans running RoboCup have been steadily reformulating the problem: changing rules, removing crutches, merging leagues, making the environment less friendly on a deliberate schedule. This year the humanoid league and the standard platform league were merged into one. Nobody was optimizing toward that. Somebody decided the problem had become a different problem.That is problem formation, performed by an institution rather than a person, which is worth saying plainly because it means the capacity I am describing is not a special property of human bodies. It is a property of certain systems. A governing body can do it. A committee can do it, slowly, in minutes and votes. It is simply not the thing being measured on the pitch, and it never has been.Attention Is Problem FormationWhich returns me to the ground this series has been standing on.William James wrote in 1890 that my experience is what I agree to attend to. He was not making a point about focus or productivity. He was saying that attention is the act by which a mind decides what its situation consists of, and that the deciding comes before the experiencing.I should be honest that I am extending him past what he claimed. James was describing conscious attention, and most of what I have been describing never reaches consciousness at all. What I am borrowing is the structure rather than the phenomenology: something selects what the situation consists of, and that selection is prior to everything downstream of it. Whether that something is available to introspection is a separate question, and in these cases the answer is mostly no. Read the structure next to a player at the hour mark, and it stops sounding like philosophy. Deciding what matters right now is not preparation for solving the problem. It is most of what solving the problem consists of.That capacity develops the way this series keeps describing. Consequence returns to the actor, on the same continuous self, without delay. An infant reaches and misses, pulls itself up and goes down, misjudges a gap and pays for it immediately, and the cost lands on the body that made the error. Run that loop enough times, and the competence arrives and then leaves awareness. What gets delegated downward is not only the execution. It is the judgment about what matters, which is why you can cross a room with a cat in it while thinking about something else.An objection I should meet, since I am doing what I claim cannot be done. This essay describes an automatic competence at some length, so is it really unavailable? Describing that something happens is not the same as specifying how, and the gap between those two is the whole subject. I can tell you my spine was being stabilized. I could not have told you, at any instant, what it decided mattered.And I should be equally plain about the limits of what I have offered. I have named a capacity and given it a criterion. I have not explained it. I cannot tell you what computation performs the re-noticing, what architecture would support it, or how it differs in kind from very good context sensitivity, and I am wary of anyone who says they can. That is not evasion, or at least I hope it is not. It is the argument: this is a capacity we have not characterized well enough to build a test for, which is precisely why our tests are quiet about it.How Narrow The Claim IsThis is not a bet against the trend. The trend in robot football is real; it has held for nearly thirty years, and I have no argument that it will break. If anything, I expect it to continue.The claim is about what the trend measures. Every benchmark is an architecture, and its architecture decides which capabilities it can see clearly. Success inside one is direct evidence about what it was designed to measure. Anything it appears to say about capabilities outside that design is indirect, and has to be argued for rather than assumed. That is not a strange evidentiary standard. It is the ordinary one, and we suspend it constantly.Discovering that a different set of facts has become the relevant ones is not what any of these competitions was built to measure. So thirty years of rising scores do not, by themselves, establish whether that capacity has improved. It may have. Something like it may already be present in systems trained across enough varied conditions, and if anyone wants to argue that, the argument is available and I would read it with interest. What I am objecting to is the skipped step: treating the score as though it had settled the question.Here is what would change my mind, and I have tried to make it checkable rather than rhetorical. Show me that performance on a specified benchmark predicts how a system does when the relevant facts change, and specifically when success requires discovering that the original framing has stopped being adequate. Not adapting a route to a fixed target. Working out that the target was the wrong one and that something else now matters. If scores turn out to predict that, the two things were never separable, the inference I am calling unwarranted is warranted after all, and I will say so here.The Question I Cannot Answer For YouThink about the last time you were partway through something and quietly realized it had become a different task than the one you started. Not that you were failing at it. That the thing worth doing had moved while you were working.Nobody set you that problem. You noticed it mid-stride and reformulated what you were doing before you could have explained why. No test you have ever sat measured that.I want to be careful about where this lands, because there is a warm reading of everything above that I do not intend and would rather refuse than allow. This is not an argument that something in us is unreachable, or sacred, or beyond machines by its nature. I have no idea whether it is beyond machines, and I have said so. The capacity is ordinary; it is physical; it is learned, and an organizing committee can do a slow version of it in meetings. Nothing here needs mystery to work. What it needs is for us to notice that we have been reading our scores as though they covered it.So when the scores go up, and we say the machines are catching up, the question is not whether the scores are real. They are. The question is what our measurements were built to see, and what they have been quietly silent about the entire time.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

There are people whose bodies have to be checked at the end of the day. Not out of anxiety, and not because anything feels wrong. The checking happens because damage arrives without announcing itself, and looking is the only way to find it. A burn that never registered. A cut that has been open for hours. A joint that has moved further than it should and stayed there. In childhood, a parent does the looking. Later it is the person themselves, or a clinician, or both. Either way, the inspection is deliberate, scheduled, and conscious, and it stands in for something most of us are handed without asking: a signal that arrives on its own at the moment of the damage, whether or not we happen to be paying attention.The condition is congenital insensitivity to pain, and it is rarer and less tidy than the name suggests.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.It is not one condition. Several genetic routes lead to it, and I am going to take two of them, not because they exhaust the picture but because the contrast between them is useful.In one, a gene called SCN9A encodes part of a sodium channel that pain-sensing nerves need to send their signal. When both copies carry disabling changes, that signal does not get through. This route comes closest to a clean experiment because most other sensations remain. Touch works, or nearly: testing has found some differences in how touch is felt at the margins. Position works. The known exception is smell, because the same channel is used by the nerves in the nose, and people with this form are typically unable to smell.The second route runs through a gene called NTRK1 and produces what is usually called congenital insensitivity to pain with anhidrosis. Here, more is missing. Along with pain, the ability to sweat is absent, which leaves the body unable to cool itself. Sensitivity to temperature is affected. Intellectual disability of varying degree is characteristic. This is the form where the confounds are worst.I want to be careful here. The injuries are documented, and they recur across both routes: biting of the tongue, lips, and fingertips, sometimes severely; damage to the eye; fractures walked on and healed badly; joints worn down; bone infections that took hold because nothing reported them early. What I cannot honestly tell you is that this accumulation is what shortens life. In the second form, early deaths are substantially driven by fever the body cannot shed, which follows from the missing sweat rather than from the missing pain. Reported mortality varies across the groups studied, and I am not going to hand you a figure that sounds more settled than the evidence warrants.So the narrow claim, the one that survives both routes and all the confounding, is this. The injuries accumulate because the signal that normally reports damage as it happens is missing. Living with this condition means substituting a deliberate audit for a signal that would otherwise have arrived on its own.Which is worth sitting with because it reveals the pain underneath the flinch.It is not sensation for its own sake. It is a feedback loop rendered in flesh. It takes a bad outcome and converts it into something carried, something the body cannot set down, something that reaches forward and changes the next reach of the hand. Three weeks ago I described the fast loop that living things run inside a single life: the animal that touches the fire pulls back, carries the burn, and approaches the fire differently tomorrow, the same animal, changed. Pain is part of how that sentence gets implemented. It is the part of the mechanism in us that makes a burn teach.Knowledge tells you the stove is hot, and knowledge is not nothing here. People with this condition learn a great deal, and much of what keeps them safe is learned rather than felt. What knowledge does not do is bind the lesson to the moment on its own. A person with this condition can see the flame, know the word, and explain to you exactly why fire is dangerous. What is missing is the small insistent return that says this one cost you something, arriving without having to be asked for.One Word For A Lot Of Different ThingsAnd here I have to correct something in my own vocabulary.Across these essays I have said that consequence returning to the actor is what turns knowledge into wisdom, and I have said it as though consequence were a single substance arriving in varying amounts. It is not. Being embarrassed in a meeting, breaking a bone, losing a reputation, and watching an institution fail are not four sizes of the same thing. They arrive at different speeds. They reach different distances. Some can be argued with and some cannot.Two of my own, for contrast.Earlier this year, while driving for Uber, I hit a parked car. Low sun, a dark grey car on the shaded side of the road, and a fraction of a second spent looking down to change the radio station. The airbag went off. The smell of the powder is the detail I have kept. I called the emergency services, then my wife. She came round the corner from the house, where her parents were loading the car for the drive back to Maine. The damage looked limited to me. The insurer wrote the vehicle off.What that taught me is narrow, and it is permanent. I am careful about parked cars on my right, leave more clearance than I used to, and have not had to decide on either since. What it never told me was which part of what I did mattered. The sun was real. The radio was real. I cannot separate them, and I have stopped trying. The cost arrived, the behavior changed, and no account came with it.The other one cost a great deal more and taught a great deal less.Seven or eight years before that, I was the account lead on a large data program for a retailer. Every two weeks we prepared slides for an executive update. Every two weeks, the client's project lead edited down or removed the risks in those slides before the meeting. My colleague and I both understood what was happening. We had other problems at the time, in staffing and in delivery, and we chose not to force it. So the meetings went ahead twice a month without the real state of the work being included.The problems did not go away. After about six months, they were obvious to everyone, and we were left explaining that we had known and had deferred to the client on the contents of the client’s own slides. We were asked to stop work. It took another six months to recover a seven-figure final payment, and it nagged at me for all of them. Worth noting where that landed. The commission on the sale had been paid long before delivery started, to someone who was not in the room for any of this. The people holding the work bore the cost.Ask me what that changed, and the honest answer is: a little, perhaps. Because I could argue my position, and I did, and it was not a bad position. That is the thing worth noticing. The slow arrival came with a complete account attached, and an account can be reasoned with, and reasoning with something is most of the way to setting it down.The expensive one taught less than the cheap one. What those two suggest to me is that pain sits at one end of a range. It arrives during the event rather than after it; it lands on the body that acted, and it comes with no reasoning you can take issue with. Embarrassment is social and fades. Financial loss arrives as a document, and a document can be answered. Collapse arrives once, by which time the people who might have learned from it are rarely still in the room.What This Comparison Can Carry, And What It CannotIt would be neat to say that these machines, by which I mean here and throughout this series the AI systems many of us now talk to daily, are painless bodies grown large. I don't think that holds, and I would rather say so than hope you didn't notice.Someone with the first form I described, the one that leaves most other sensation intact, is missing one channel out of an otherwise working set. They can still be embarrassed. They still learn from outcomes, and they learn protective rules well enough that the rules become much of what keeps them safe. What is absent is a single protective signal inside a loop that is otherwise running. The absence in these systems is larger. They do not carry a correction forward as the same continuing thing; the durable fixes are made to later versions, by processes outside the system itself. This is not a small version of that. It is a different absence, in a body that is not otherwise like them.There is a harder objection beneath the surface, and I would rather take it up than leave it for you to find. People live with this condition. Not easily, and not without cost, but they manage it through rules, inspection, treatment, and the attention of other people. If knowledge and procedure can stand in for the missing signal, then perhaps consequence was never what mattered, and what matters is having a working system for finding and fixing errors.I think that is right. It sharpens the question rather than ending it: what exactly was the signal supplying that now has to come from somewhere else?For my purposes, separate four jobs that event was doing at once. It registered that something harmful had happened. It made that hard to ignore. It tied the harm to the moment it occurred. And it applied pressure on what the hand did next.Note the third one, because it is smaller than it looks. Pain ties damage to the moment, not reliably to the cause. My own collision is the proof. Something arrived that changed my driving permanently and never told me whether the sun or the radio was the problem. Working out what actually caused a failure is one of the things a deliberate process does far better than a body does.Take the signal away, and none of those four becomes impossible. Each becomes something that has to be built. But what goes missing is not any of the four. It is that they arrived already joined. No step sat between them where someone had to decide to pass the signal along.That does not make the automatic version better. Pain is noisy; it fires long after the damage has healed, and much medicine consists of overriding it. Built systems can be more accurate and more redundant than any signal a body produces. The advantage is narrower than superiority. In the body, the joins were given. In everything we build to replace it, the joins have to be made, and a made join can be left unmade.Here are two more of mine, and they came out differently.Two or three months into that retail program, before anything had visibly failed, I raised it internally. I told our office leadership that progress was behind our projections, that I believed there was a serious architectural problem, and that we were being discouraged from raising the concern with the client. The check worked. It detected what it was meant to detect, early, and reported it accurately to the people it was designed to report to. What came back was encouragement to keep reporting internally and to stay positive with the client. Nobody above me was obliged to answer what the report contained. So it was received and absorbed, and the work continued.Earlier this month, a reader went through one of my working papers and found an arithmetic error in it, an adjustment I had applied in the wrong direction, along with a comparison I had drawn that did not hold dimensionally. He was right on both counts. I had read that paper many times. Nothing in the reading reported it. I corrected it, published the revision, credited him, and the whole thing took days.Two feedback events, one person, opposite outcomes. And notice that the one that worked was the less automatic of the two. It arrived late, from outside, in writing, from someone I had never met. What it had was a join I could not quietly leave unmade. Ignoring him was an option, but once his correction was public, ignoring it would have been its own visible act.The internal warning had no such join. Whatever else may have failed, and I cannot tell you from the inside whether leadership found it credible or simply doubted my diagnosis, the last one certainly did. The report was received, and nothing followed, because nothing had to, and nothing showed that nothing had.So a built chain is not weaker than a body because it is deliberate. It is weaker wherever its joins are held by someone’s discretion rather than by design. That is what the scheduled inspection quietly depends on, and it is easy to miss. Checking for damage you cannot feel only protects you if finding it leads to treating it. An inspection whose findings go nowhere is not a replacement for the missing signal. It is the appearance of one.What The Substitute NeedsWe have no way to give a body back a signal it was born without. What we have instead is a chain: someone looks on a schedule, what they find is made to count, and something acts on it. That works. It works in clinics every day. What it costs is that every join in it is somebody’s, and has to keep being somebody’s.Which puts everything on a range rather than in two camps. Pain sits at one end, where nothing sits between the harm and the response. A safety interlock that cuts power without asking is not far behind it. An audit with real enforcement is further along. A risk edited out of a slide deck before the meeting is at the other end entirely, where every join is somebody’s decision and any one of them can quietly not be made.So I am not making a case for suffering. Pain is still the thing we medicate and still the thing we hope to be spared, and I would not argue otherwise. I am pointing at what it is doing while it is there.The kindest thing pain does is not protect us, though it does. It is that it needs nobody’s cooperation. No one has to approve it, or pass it upward, or decide it is worth raising. It arrives, at the time, at the one who acted. Everything we build to stand in for it has to have that joined on purpose, at every step where a body had it for free.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

Last week I ended on a claim I did not finish. Whatever fixes on a single number and optimizes hard enough, with nothing held separate that can disagree, loses the part that could have told it the number was wrong. I said we had built such things before, at a scale where the pursuit runs for decades, and where the part that should have said stop was never made of neurons.I meant institutions.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.An institution runs longer than any of the people inside it. That is close to its whole purpose. It holds what was learned after the people who learned it have gone, and it hands that forward to people who were not there. Which gives an institution something a machine does not have: a record that outlives the individual who paid for it.So the failure I want to describe is not forgetting. The records are there. Minuted, archived, classified, indexed. The failure is stranger, and worse. An institution can hold a complete account of what happened and learn nothing from it, because reading a ledger is not the same as bearing what the ledger records.Three Parts, Not OneThere is a tension there, and when I began this piece I thought I could dissolve it. Last week’s absence was a missing dissenter: a part held separate, able to disagree, able to say stop. This week’s absence is a memory nobody answers for. A missing voice is not a missing archive, and I had a tidy answer ready for why those were the same thing seen from two angles.The answer was too quick. I would rather correct it here than let it stand.They are not the same absence. They are different parts of one loop, and I have spent four essays describing that loop as though it were a single part when it has at least three.Memory preserves the argument. It holds what was concluded, on what evidence, against what alternatives, at what cost.Dissent tests the argument. It supplies pressure while a judgment is still being formed, rather than after it has hardened.Accountability makes someone answer for the choices they make. It is what keeps the first two from becoming ceremonial: a record cited without being read, an objection minuted without being weighed.An institution can be excellent at one of these and hopeless at the others. A safety board can preserve forty years of incident reports while nobody who reads them stands to lose anything. A founder can carry enormous personal exposure and systematically suppress everyone who disagrees. Those are not edge cases. They are ordinary.Which changes what I have been claiming, so let me say it plainly. I have been treating consequence-bearing as the entire loop. It is not. It is the part that makes the other two binding. A record to which nobody is accountable cannot say stop. It can only be cited or set aside, and both cost the same, which is nothing. But accountability with nothing to read and nobody to argue with is only exposure, and exposure produces anxiety rather than judgment.Forty Years Of The Same AnswerHere is a case where the record is unusually clear and unusually long.For roughly four decades the United States held a posture toward Iran that is easy to describe and easy to get wrong. It was not restraint in any general sense. There were sanctions throughout, covert action, proxy conflict, naval confrontation, and targeted strikes. The posture was narrower and more specific: a repeated refusal to take the last step. No full-scale war, no attempt at regime change by force.That posture was tested about as hard as a posture can be.In 1983, a bombing at the Marine barracks in Beirut, widely attributed to an Iran-backed group, killed 241 American service members. No war followed. In 1988 American and Iranian forces fought directly in the Gulf, and an American warship shot down an Iranian civilian airliner, killing everyone aboard. No war followed. In 2003, the United States invaded the country next door and did not stray into Iran. In 2020 an American strike killed a senior Iranian commander, Iran retaliated against American bases, and both sides stopped.Different decades. Different administrations, different parties, different circumstances. The same conclusion, reached again and again by people who agreed with each other about very little else.Then it ended. In June 2025 Israel struck Iranian nuclear infrastructure, and the United States followed within days, using weapons only it possessed. In February 2026, the sequence repeated on a far greater scale, and this time the stated aims included regime change. Iran’s supreme leader was killed. Iran struck American bases across the region and closed the Strait of Hormuz. A ceasefire took hold in April and was extended indefinitely. A memorandum of understanding signed in mid-June formalized it, reopened the strait, and opened a sixty-day window to negotiate a permanent settlement. The window did not hold. Exchanges resumed within weeks, and by mid-July, American strikes were running nightly for nearly a fortnight before pausing. They have resumed and paused again since. As I write, talks are stalled, the terms are unsettled, and the outcome is unknown.I am not going to tell you that ending the posture was wrong. I do not know that, and I do not think anyone does yet. Circumstances had changed, and changed circumstances are exactly when a long-held position ought to be re-examined.I also want to be careful about what that sequence shows, because it is less than I first wanted it to be. It does not show one accumulating body of reasoning handed down intact. The people who stopped short in 1988 may have done so for reasons with little in common with those in 2020. Arriving repeatedly at the same outer boundary is not the same as sharing a case for it. If I claimed otherwise, I would be assembling the very ledger I am about to accuse the institution of never reading.So the claim is narrower. Whatever those reasons were, they were written down, and across four decades administrations repeatedly stopped short of the step this one took. We have heard the case for acting. What I have not seen is the other half of it: an account of why the reasoning that stopped everyone before no longer governs. Those are two different documents, and only one of them has been produced.I should be plain about what I cannot know. The ledger may have been read with enormous care. Somewhere there are almost certainly historical reviews, contingency analyses, and dissenting assessments, and I have no access to any of them. Note what that means for the three parts. Memory: the record exists. Dissent: there were people in that building who disagreed, and it would be strange if there were not. Both parts were probably intact. The failure I am describing sits in the third part, and only in the third. Nobody has had to answer for the choice between them.How A Judgment Becomes A ConventionHere is one way a reasoned position turns into an unreasoned one without anybody deciding to let it.The people who reached the conclusion in 1983 knew why. They had the cables, the casualty reports, the estimates of what a wider war would cost and who would fight it. The conclusion was expensive, and they had paid for it.The people who inherited it a decade later had the conclusion. They could also have had the reasoning, because it was written down. But a conclusion travels more easily than the reasoning behind it. It fits in a sentence. It can be handed over in a briefing. And each time it is handed over, a little more of the weight comes off, until what arrives is a convention rather than a judgment. This is how we do things here.A convention is cheap to discard, because nothing about it explains itself. It looks arbitrary, and arbitrary things invite the question of why we are still doing them. Which is a fair question, and someone should ask it. The trouble is that when answering it requires reading four decades of reasoning nobody currently in the room has read, the practical answer becomes no reason at all.David Hoze has a line I keep returning to: the certificate is always signed one floor above. Accountability that always defers upward is accountability nobody holds. It is always someone else’s job to have read the ledger, until it turns out to have been nobody’s.Where The Cost GoesSevering the mechanism in us does not make a cost disappear. It moves it.Alberto Romero has described safety margin as cost relocation, and the phrase generalizes further than he needed it to. When an institution removes the part of itself that bore the consequence, it has not become more efficient. It has arranged for the consequence to be borne somewhere else, by people who were not in the room and cannot contest the bill.This is the structural version of a claim people usually make in a partisan register, and I want to keep it structural, because the asymmetry does not care which direction the policy went. The judgment is made within a system where most of the consequences fall outside the loop that made it. The cost falls on service members, on civilians in the region, on people whose energy and food prices move when a strait closes, on a generation of taxpayers who will still be paying for the munitions long after the argument has been forgotten.Relocation alone is not the problem, and I should be honest that every state decision relocates costs onto people who were not in the room. Tax policy does it. Interest rates do it. Restraint does it too, and the people who bear the cost of a war not fought are simply harder to identify. The question is whether the cost returns as information. Whether anything comes back up the line in a form capable of changing the next judgment, or whether it lands and stays landed.There is a harder question underneath that one, and I can name it without pretending to settle it. Accountability is always accountability to somebody. Decision-makers answer to the people who can remove them, and those people have interests of their own. A reader could say the ledger was read perfectly well, by actors who read selectively because their interests had changed, and that what I am describing as a broken loop is a working one serving a different set of ends. I think that is compatible with the argument rather than fatal to it. It explains why the third part fails. It does not show that it held.That question is not only institutional, and I would rather not pretend it is. My own memory is becoming a ledger, and the delete button is always there. A great deal of what I once had to carry, I now merely have access to. Phone numbers are in my phone. My arguments are in a folder. My mistakes are in a thread I could scroll back to and mostly do not. In July I wrote that the difference between a ledger and a scar is that a scar is a record you are still inside. The record I keep of my own errors is drifting toward the first kind: complete, searchable, and weightless. Nothing prevents me from reading it as a scar except that doing so is unpleasant and optional, and those two things together are usually enough. An institution sits in the same position, with far more people available to assume that somebody else is doing the reading.The Right Of ReplyWhich brings the objection I have been circling the whole way through, and it is the strongest one available.Sometimes a position is long-held precisely because nobody has examined it. Read that way, those four decades were never a judgment accumulated. They were path-dependent: a taboo each administration inherited, and none wanted to bear the risk of being the one to break it, until someone finally did. On that account, I am not defending institutional memory at all. I am defending institutional timidity and calling it wisdom.I take that seriously, because my own test makes the problem worse rather than better. Anyone who starts a war carries enormous personal downside if it goes badly. If consequence-bearing were the whole of the loop, it would systematically favor doing nothing. Skin in the game does not manufacture good judgment. It manufactures pressure, and pressure has a direction only if something else supplies one.So the test cannot be whether someone bore a cost. Here is what I think it is instead.The old conclusion gets no presumption of correctness. The old reasoning gets a right of reply.Before an inherited judgment is overturned, the institution should have to reconstruct it and answer it. What did our predecessors believe? On what evidence? What risks were they pricing? Which of those assumptions have since changed, and which have not? And what would tell us that our replacement judgment is wrong?Notice what that asks for and what it does not. It grants the ledger no authority. A reconstructed argument can be answered and discarded on its merits, and frequently should be. What it cannot be is skipped. The requirement is not obedience to the past. It is that the previous case gets heard before it is overruled.Accountability then has a narrower job and a more defensible one. Somebody has to be answerable for having done that honestly rather than ceremonially, and for recording why the new judgment supersedes the old. Without that, the reconstruction decays into a formality: a paragraph acknowledging that predecessors thought otherwise, attached to a decision made before anyone opened the file.That is a test you can apply in advance, without waiting years for an outcome. I notice how rarely anybody is asked to pass it.What A Severed Loop Leaves BehindFive weeks ago, I started with an absence in the machine: a system that holds a record of everything it has done and carries none of it. Then the same absence in institutions, where a consequence overturned is relocated rather than erased. Then its signature, the jagged profile of capability built without a loop that returns cost. Then the mechanism itself, the loop in us, dissociable in the brain and inherited by machines in name only.And now this. The absence at its most expensive, and its least visible.Because the version I find hardest is not the institution that forgets. That one at least announces itself. It is the institution that remembers perfectly. Every file intact, every objection preserved, every prior judgment available on request. Nothing missing except the person who has to answer for having read it.A ledger becomes a scar at the moment somebody has to answer the case it makes. Not obey it. Answer it. That is the whole of what I am asking for, and it is less than it sounds and harder than it looks.The memory survives. It survives in a form that costs nobody anything.That is what a severed loop leaves behind. Not silence. A complete and accurate record, held by no one, teaching nothing.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

In reinforcement learning, a reward is a number.That is the definition, not a simplification. An agent acts, the environment returns a scalar, and the agent adjusts to make that number larger next time. Everything the field has built runs on that single quantity. Maximize the number. That is the engine.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.The word came from us, and the engineers who took it were not careless with it. They borrowed a working formalism, one that psychologists had built studying animal learning, and it worked, which is why it stayed. The carelessness came later, in how the rest of us started to talk. Reward function. Reward hacking. The model wants this, the model likes that. The word carries a psychology, and once it was loose in the language, it began paying out meaning the math had never deposited. Because in the brain, reward is not one thing. It is at least three, and they come apart.Kent Berridge has spent three decades at Michigan taking the concept apart, in rats and in people. His framework is not the whole of a large and contested field, but it is among the most influential decompositions we have, and the core finding is almost rude in its simplicity. Wanting and liking are separate systems. Wanting, what Berridge calls incentive salience, is the pull toward a thing, and, in his account, it largely runs on mesolimbic dopamine. Liking is the actual pleasure you feel when you get the thing, and it runs on different machinery: small, fragile clusters he calls hedonic hotspots in the shell of the nucleus accumbens, driven by opioids. Different chemistry, different anatomy. You can knock out one and leave the other standing. There is a third component, learning, the storing of which cues predicted the reward. Berridge’s own summary lists all three by name: liking, wanting, and learning. Three dissociable parts of a single word.Now watch which of the three the machines actually got.The learning component came into focus in the 1990s, and it did not look like pleasure. Wolfram Schultz, recording from dopamine neurons in monkeys, found that they did not fire when an animal received an expected reward. They fired when the reward beat the prediction, and went quiet when it fell short. The signal was not pleasant. It was a surprise, the gap between what was predicted and what arrived. Montague, Dayan, and Sejnowski gave that gap a formal shape, and in 1997, Schultz, Dayan, and Montague published it in Science under a title that says the thing plainly: a neural substrate of prediction and reward. The dopamine signal was a prediction error. And prediction error was already, independently, the engine of reinforcement learning in computer science. The same math, reached from two directions.You may have caught a snag. Berridge says dopamine carries wanting. Schultz says dopamine carries prediction error. Neither can be the final word, and the field has not fully reconciled them. But notice what neither camp claims. Neither says dopamine is pleasure. Whatever the molecule is doing, pulling you toward a cue or reporting an error in your forecast, it is not the hedonic hit itself. The two theories disagree about nearly everything except the one thing this essay needs: the signal is not the liking. That is the narrow place where biology and the machine rhyme. They rhyme on the error signal. Not liking.The correspondence even runs both ways. In 2020, a team at DeepMind and Harvard, led by Will Dabney, took a refinement from AI, distributional reinforcement learning, in which a system tracks a whole spread of possible outcomes rather than a single average, and went looking for it in the brains of mice. They found it. The dopamine neurons were not all reporting the same expected value. They were tuned differently, and together they carried a distribution. A method invented for machines turned out to describe the tissue. A beautiful result, and notice where it lives. Entirely inside the error signal. The deepest overlap we have found between the brain’s reward system and the machine’s still never reaches the part that likes.So here is the inheritance. The machine received the error signal, learned from it, and built it well enough that it sometimes predicts real neurons. That is where the inheritance stopped.Now, a trained system pursues. It optimizes, seeks the states that raise the number, and finds routes there that no one anticipated. It is tempting to call that wanting, and in a loose operational sense, you can. But it is worth being exact, because this is where the borrowed word does its quietest damage. The system did not inherit a wanting system alongside the learning one. Pursuit is simply what optimizing a scalar looks like from the outside. Give a process one number to maximize, and it will move toward whatever raises it, and that movement will resemble desire, because desire, in us, also moves toward things. The resemblance is real, and the mechanism is not. There is no incentive salience under the hood, no second channel attributing pull to a cue. There is one objective, and a behavior that falls outside it.And there is nothing that answers to liking. Not “the machine cannot feel,” which is a claim about minds I have not earned. The narrower, harder claim is this: nothing in the architecture corresponds to the known mechanisms of hedonic impact. No hotspot, no second signal that registers the pursuit as good, separate from registering that the number rose. The scalar reports that the number went up. It has nothing to say about whether the number was the right one to raise.You might say wanting and liking are obviously the same thing anyway, that prying them apart is a lab trick with no bearing on a life. I thought something close to that once. Then I spent a year driving.There is a moment most nights, past the point where the driving has stopped being pleasant, when the app offers me one more. A ping, a small green number, sometimes a bonus for staying online. And I take it. Not because I want another forty minutes on the road. My back hurts, and I stopped enjoying this hours ago. I take it because the ping pulls, and the pull is not the same as wanting the outcome, and it is nowhere near liking it. The app was never built to make me like driving. It was built to make me take the next ride, an incentive salience machine, wanting engineered clean, with the liking left out because the liking was never the point.That is the mild version of a split that has a clinical name. Berridge and Terry Robinson built a theory of addiction on exactly that seam. The wanting system sensitizes and swells, screaming for the drug, while the liking system flattens or falls away. The addict late in the disease often reports getting no pleasure from the substance at all. The wanting has come loose from the liking and kept running on its own. Wanting what is no longer liked.Hold onto what that proves. The parts come apart in us, in the one system that has all three, wired together by evolution and still able to tear along the seam under pressure. If they dissociate in a brain built to integrate them, a system assembled from only one of them was never going to grow the others by getting larger.This is the mechanism the series has been circling, and here it finally gets a physical address. The missing thing was never a feeling you could bolt back on. Train the system to announce that it is satisfied, and you have added a sentence, not a signal. The missing thing is the separation itself. In the brain, these systems are distinct, and that distinctiveness is what allows one to check the other. Liking can say the thing was good. Wanting can move you toward it. Learning can update the map. Collapse them into one flat number, and none can check the others. There is a pull, and a running tally, and nothing underneath to say whether any of it was worth it.A brain keeps them apart so that consequences have somewhere to land. Pleasure and its absence are how the loop closes on the inside, how a body registers that the thing pursued was worth pursuing or was not. Strip the parts to one, and you have not simplified the loop. You have opened it. The signal goes out, and nothing valenced comes back, because there is no part left whose job was to feel the return. We started with the part that pulls, and within that objective, there is no gradient leading to feeling.It is worth stopping here, because this is where a careful reader pushes back, and there are two good pushes.The first is emergence. All I have shown, the objection goes, is that today’s systems lack this channel, not that they can never grow one. Fair. I am not claiming a machine can never have valence. The claim is narrower and concerns this construction. Within the optimization objective itself, nothing selects for a second evaluative channel, because the objective already fixes what the system is for. If a hedonic channel did emerge, it would be doing work the scalar already does, which is a reason to doubt it would be selected, not a proof that it is impossible. You will not get feeling for free from a number, because the number leaves nothing for feeling to add.The second push is sharper. The brain’s separate systems, it goes, are a biological workaround. Evolution could not write a clean utility function into a rat, so it hacked together dopamine for pursuit and opioids for pleasure. A machine needs none of that. Give it direct access to the ground-truth objective and the scalar; liking is not missing but obsolete. The running tally is the mechanism for registering worth.This is the strongest version, and it turns on a hidden assumption: that the scalar is ground truth about worth. It is not, and the reason is the crux of the whole argument. A learning system can already catch one kind of error. When a reward arrives lower than predicted, the prediction error says so, and the system corrects. But that correction is always aimed at the target it was handed. It can learn that a route to the number was wrong. It has no signal to tell it that the number itself was wrong.That second kind of error is what liking supplies. Liking is not a readout of the reward figure. It is a separate report on the outcome, and because it can dissent from the pursuit, it can register what the pursuit never can: not that the forecast was off, but that the thing forecast was no good. That is what addiction is, read from the inside. Wanting says "go"; liking says "this was not good"; and the disagreement is information. A single scalar cannot disagree with itself. Split the system into a part that acts and a part that grades it, hand it a whole vector of objectives instead of one, and nothing changes at the level that matters. The grader still grades against the target it was given. The vector still lists targets someone chose. Every refinement sharpens the pursuit of the goal. None of them can doubt the goal. That is the channel a brain keeps separate, and it is the first thing you lose when reward collapses into a single number.Which is a strange thing to have built. A pursuer that runs at full strength with no channel to register whether the pursuit was worth having, no second signal that could ever say the number was the wrong number. In one machine, that is a curiosity. In one brain, it is a disorder we have a name for.But the flattening is not confined to machines or brains. Whatever fixes on a single number, optimizes hard enough and long enough, with nothing held separate that can disagree, loses the part that could have told it the number was wrong. We have built such things before, at a scale where the pursuit runs for decades and the part that should have said stop was never made of neurons. That is the cost this series has been walking toward. We turn to it next.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

The HandoffLast week, just before I handed the question back to you, I made a claim I now owe you an argument for. When an intelligence develops without consequence ever binding it, I wrote, the absence does not stay hidden. It leaves a mark on the shape of the capability itself, a signature you can learn to read.Two weeks ago that absence showed up in the memory of a machine, by which I mean here, and throughout this series, the AI systems so many of us now talk to every day: a ledger without a scar. Last week it showed up in an institution’s record: a scar being quietly converted back into a ledger. Both times we caught the absence in behavior. This week I want to step back far enough to see it in the silhouette. The absence is not only visible in what these systems do. It is visible in the outline of what they can do at all.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.The Shape Everyone SeesMelanie Mitchell has drawn the sharpest picture of it. In her recent Yale Review essay, she takes up the term now attached to these systems, jagged intelligence, and shows what it looks like from the inside: a capability profile made of spikes and canyons, superhuman performance on one task sitting directly beside failures no competent adult would produce, with no reliable way to know in advance which one you are about to get. Her emphasis falls on the unknowns, and rightly. The danger is not that the cliffs exist. Every tool has limits. The danger is that nobody, including the people who built the system, holds a trustworthy map of where the cliffs are.She is not alone in seeing the shape. Ethan Mollick and his collaborators, studying professionals working alongside these systems, describe a jagged frontier: an invisible, irregular boundary between the tasks where the machine lifts your work and the tasks where it quietly ruins it, running through the middle of jobs rather than around them. Two observers, two vocabularies, one silhouette.Picture it as a shape. A capable adult’s competence, plotted across the things they actually do, looks like a rounded form: higher here, lower there, but continuous, with edges that taper rather than plunge. The machine’s profile looks like a star drawn by a seismograph. Long spikes, deep notches, and no taper anywhere.The observation is right, and I have no quarrel with it. I have honored it because I believe it. My question is upstream of it. Why that shape?Why Human Profiles Are SmoothIt is worth noticing that human competence does not start smooth. Spend an afternoon with a bright child, and you will meet a profile as jagged as any chart of machine test scores: startling ability directly beside startling gaps, deployed with total confidence in both directions. What happens between the child and the adult is not that talent gets redistributed more evenly. It is that consequence starts arriving, and I mean the word broadly. The weather is not only embarrassment and cost. It is the plain friction of operating a body in a physical world, where every dropped cup and misjudged step reports back instantly and without mercy.Everywhere an adult operates, reality pushes back. Overreach gets corrected, sometimes gently, sometimes in front of an audience. I met this early. My final year structural design project as a civil engineering student, an exhibition hall entered into a design competition, rested on massive tubular arch beams, and I had analyzed the whole structure in I-DEAS, a state-of-the-art structural analysis software running on Silicon Graphics workstations, which returned beautiful, confident results. The night before submission, for no reason I can fully reconstruct, I ran a quick back-of-the-napkin check on the stresses in those beams. The package had interpreted my loads as Newtons where I had meant kilograms, an error of nearly a factor of ten, and under the real loads my structure collapsed. Nothing inside the software had ever pushed back. The analysis was fluent all the way down. The correction came from outside it, arrived almost too late, and I have never trusted a smooth output the same way since.I still run that loop on purpose, every week, in this newsletter. Every essay goes through a gauntlet of critics whose job is to find where I have overreached, and then it goes somewhere less forgiving: in front of you, under my name, where being wrong is a matter of record. What you are reading is a sanded object.Under that kind of pushback, a person makes one of two moves, and both of them smooth the profile. Where the cost keeps arriving, and the work matters, you practice until the canyon fills. Where it keeps arriving, and the work can be routed around, you retreat and say so out loud, which is its own kind of competence. Practice fills the canyon. Humility fences it. Either way, the surface where you actually operate gets rounded off, and something else happens that matters just as much: you come away holding a map of your own edges. I know roughly where my canyons are. I paid to find out.Smoothness, in other words, is not talent. It is erosion. A rounded capability profile is what any capability looks like after enough weather.And let me not oversell the smoothness, because humans are jagged too, and anyone who has watched a brilliant surgeon manage money, or a physicist hold forth on politics, knows it. So the claim is narrower, and stronger for it. We do not get sanded everywhere. We get sanded where consequence keeps returning to us, and we stay jagged where it does not. Step outside the territory where you have paid, and the confidence often travels while the sanding does not. Which is exactly the point. The smoothing follows the loop, runs wherever the loop actually runs, and stops precisely where it stops.Remove the Loop, Get the StarNow build an intelligence with the weather turned off.Training pushes capability wherever an objective can be specified and graded. That is not a criticism; it is close to a definition. Wherever the grader reaches, pressure is applied, and a spike grows. Wherever the grader cannot reach, because the goal cannot be written down cleanly, or the answer cannot be easily checked, nothing pushes back at all, and whatever canyon was there stays exactly as deep as it began. No cost ever arrives from the canyon floor, so nothing marks its location, not on any map the system holds and not on any map we hold either.A fair objection arrives immediately, and it deserves a straight answer. Training is a feedback loop. These systems are corrected millions of times, penalized for wrong answers, adjusted after every failure. Is that not consequence? It is feedback, and here I need to be precise about the word I have been leaning on, because consequence, as I am using it, means something narrower than feedback. It means that the acting system itself, the same continuing agent, bears a cost that alters its own future behavior. Feedback that lands somewhere else does not qualify. And look at where this feedback lands. The corrections land on the next version of the system, applied by someone else’s process, in service of someone else’s goals, long after the fact. I made this argument two weeks ago about memory, and it holds here without modification: a loop that closes on your descendants is not a loop that closes on you.You might answer that nature runs exactly this kind of loop. Evolution corrects across generations, and it produces beautifully rounded competence. True, but living things run two loops at once. The slow one closes across descendants. The fast one closes inside a single life: the animal that touches the fire pulls back, carries the burn, and approaches the fire differently tomorrow, the same animal, changed. The machine has been given only the slow half. Somebody will also point out that deployment is a loop of its own, that a system which fails badly enough gets updated. It does, and notice who bears the cost while that loop grinds around: the company, the engineers, the product team. The loop exists, but it closes on the people around the system rather than on the system that acted. And at the moment the machine actually acts, when it is answering you, you can push back all you like. The pushing does not stick. The acting happens in a room where nothing that occurs can permanently reshape the actor.A month ago, in a shorter note, I described this shape as a straight line viewed from the wrong angle, and I want to soften that here into something more careful. From most directions the star looks like chaos, spikes and notches without pattern. But turn it until you are sighting along one particular seam and much of the shape resolves. The spikes largely trace where objectives could be specified and graded. The canyons tend to fall where they could not. The jaggedness is not random. It runs, in the main, along one seam, and the seam is the reach of the grader, drawn in silhouette.To make the seam concrete, take two things these systems are asked to do. Writing code that passes a test is easy to grade. The test passes, or it does not, millions of times over, and the systems have grown a towering spike exactly there. Now consider knowing when to tell a client that a project should be canceled. No test exists. The answer arrives years later, tangled in everything else that happened, and nobody can score it at the moment it matters. There the canyon sits.And here the machine’s jaggedness differs from the child’s in the way that should hold our attention. The child’s profile is jagged and getting sanded. The machine’s is jagged and unaware of it. It has no feel for its own edges. It descends into its canyons at the same even confidence it carries across its peaks, because no crossing has ever cost it anything, and nothing in its history distinguishes the two terrains. That, I suspect, is the sharpest edge of Mitchell’s unknowns. The cliffs are not merely unmapped. The system’s own manner gives you no warning that you are approaching one.There is a rival explanation, and honesty requires me to put it in front of you at full strength. A researcher can point out that these systems are jagged for reasons that have nothing to do with consequence: the design of the systems themselves, the unevenness of what they were trained on, spikes where the material was rich and canyons where it was thin. Some canyons even sit on tasks that are easy to grade, which my seam, read strictly, should not allow. I take all of that seriously, and so the claim I am making is narrower than it may have sounded. The design and the data decide what a system can learn at all. My claim is about the rounding. In every intelligence we know, the process that takes a capability and wears it smooth, that corrects the overreach, fences the weak spot, and builds the map of edges, is consequence returning to the actor. That process has no counterpart here. Whatever first carved this star, nothing exists to sand it.I should be plain about my ground before going further. I am not one of the scientists who study these systems from the inside, and I cannot tell you what is happening inside them, and whether the canyons follow patterns that rigorous work could map is an open question for people better equipped than I am. I argue from one mechanism I have spent a career watching in people and in organizations: what happens to capability when consequence does, or does not, return to the actor. If the researchers find a better explanation for the seam, I will read it gladly and say so here. And the claim can fail on its own terms. If someone builds a system that genuinely bears consequence, acting where its own errors return to it and carry forward, and its profile stays just as jagged, then I am wrong, and I will say so here too.The Objection I Want to Agree WithThere are two objections waiting for this essay, and both come from the tradition Mitchell works in, so I want to meet them directly and then agree with most of what they say.The first has a long name and a simple meaning. Anthropomorphism is the old human habit of treating things as if they were people: naming the car, cursing the printer, reading feelings into anything that talks. With machines that speak fluent first-person English, the habit becomes almost irresistible, and Mitchell’s tradition is rightly on guard against it, because the danger is not the comparison itself; it is that the words do the convincing for us. Talk about a machine the way we talk about a person for long enough, and we start believing it has an inner life without ever having decided to.The second objection is sharper. The philosopher Brian Cantwell Smith argues that these systems handle descriptions of the world without ever actually dealing with the world. He calls the missing thing registration, and the plainest way I can put it is this: a navigation app can tell you the bridge is out, but if it is wrong, you are the one in the river. Nothing is ever riding on the answer for the system that produced it. On his account, until something is riding on it, you do not have understanding. You have very good paperwork.I concede both objections, and not through gritted teeth. Nothing in this essay claims the machine feels its canyons, regrets its spikes, or suffers its training. I have argued in this series that its design gives it no way to carry such experience forward even if it existed. The comparison between human smoothing and machine jaggedness runs in one direction only, and it points back at us.Because the pattern I am describing never needed a mind. It needs a loop, or the absence of one, and loops are a property of systems, not souls. Last week I showed you a governing body whose profile has gone jagged in precisely this way: exquisitely capable wherever scrutiny grades it, and a canyon at the exact point where the grading cannot reach, which is to say wherever power can pick up a telephone. Sever an institution’s loop, and you get the same silhouette. The jagged star is not a portrait of the machine. It is what any capability looks like, in silicon or in bylaws, when nothing returns the cost of being wrong to the thing that acts.The Question I Cannot Answer.So: a memory that records without bearing, a record that binds everyone except the people holding the pen, and now a shape, the silhouette capability takes when the sanding never happens.Which leaves one question hanging over all of it, and it is the question I will take up next week. If a closed loop is what does the smoothing, what exactly is the mechanism? Because in us it is not a metaphor. The mechanism that returns consequence to the decider is physical. It has an address in the tissue, a chemistry, and a strange recent history, because when we built learning machines we borrowed its central word without taking the parts the word referred to. That story is next week’s essay.For today, I will end where this piece has been pointed from its first sentence. In her essay, Mitchell names what separates us from these systems: we are embodied, we are active seekers, and we carry a profound caring about the consequences of our actions. She offers those as a list, three items standing side by side. I have asked her once before, in a comment on her work, the question I will now ask in public, because it is her question to answer and I genuinely do not know what she will say.Is consequence-bearing one more axis on the jagged star, one more item on the list? Or is it the candidate explanation for why the star is jagged at all?Hero images: Glencoe photograph by Craig Hunter and mountain image by AVImages, both via Pixabay; third panel generated by the author.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

Last week, I left a question hanging. The ledger, it turns out, can be edited at both ends. What happens when the beings built to bear consequences start behaving like the systems that were built without them?I expected to spend this week constructing a hypothetical. Instead, the World Cup handed me a case study, complete with a governing body, a phone call, and a scoreline.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.The CallLet me concede the obvious first. The red card shown to Folarin Balogun was probably a bad call. Balogun himself said a yellow would have been fair. Plenty of neutral observers agreed. Referees miss things, VAR misses things, and every team in every tournament has absorbed a bad decision and played on. If this essay were about officiating standards, it would be short, and someone else should write it.That is not what this is about.A suspension following a red card is not a punishment attached to a rule. It is the rule. It is the mechanism that gives the card any meaning at all. Take away the suspension, and the red card becomes theater: a referee raising a colored rectangle while everyone waits to learn whether it counts. And it has always counted. As outlets across the football press reported last week, this is the first time since 1962 that a red card shown at a World Cup has not produced a suspension. Article 27 had previously been stretched to cover cards earned in qualifying matches, but never for a card shown inside the tournament itself. And the lone precedent is instructive: Garrincha’s 1962 reprieve came after one South American president co-signed a petition and another personally called the referee. In sixty-four years, the record has been amended exactly twice, and both times a head of state was leaning on the pen.This time, FIFA reached for Article 27 of its disciplinary code, suspended the ban, and placed the player on probation, not because new evidence emerged, and not because an independent review found error, but because someone powerful enough to make the call made the call. UEFA’s response was three words long, more or less: unprecedented, incomprehensible, unjustifiable. Even Sepp Blatter, a man who knows something about how FIFA bends, observed that red cards are overturned by rules, evidence, and independent bodies, not by telephone.What An Institution Is ForHere is where last week’s vocabulary earns its keep.I argued that an AI’s memory is a ledger rather than a scar: a record of what happened that carries no stake in what happened. The machine can consult its history, but the history costs it nothing, changes it in no way, and can be deleted without residue. I said the difference between a ledger and a scar is the difference between information and consequence.Human institutions sit at a strange midpoint between those two things. An institution is a ledger by construction. It is rules written down, precedents recorded, decisions filed. Nothing about paper, bylaws, or disciplinary codes inherently binds anyone. But an institution that allocates responsibility, a court, a disciplinary committee, a governing body, is a ledger designed to behave like a scar. The entries are supposed to be permanent in the way that matters: the suspension is served, the penalty stands, the precedent constrains the next case, whether or not the next case is convenient. The design goal is to give a written record, one property writing does not naturally have, which is that you cannot take it back.Now, an objection, and it deserves to be met head-on rather than waved past. Don’t good institutions revise their records all the time? Courts hear appeals. Science publishes corrections. Wrongful convictions get overturned, and we rightly count those reversals among our institutions’ finest moments. If permanence were the whole point, every appeal would be a corruption, and that is obviously wrong.So let me put the claim more carefully. Institutions do not exist to make consequences permanent. They exist to make consequences credible. And credibility can survive correction. It can even be strengthened by it, provided the correction itself is governed by rules that bind everyone equally: transparent, independent, available to the weak on the same terms as the strong. I have lived the difference. Early in my consulting career, a signed contract turned out to rest on a wrong assumption, ours, and the honest fix nearly tripled the client’s budget. We put the mistake into the record rather than around it. We named the error as our own, documented the gap, and worked the revision through every approval the client’s process required. That client kept working with us for years afterward, and I am convinced the reason is that they watched us correct the ledger in daylight. The record changed, and it became more credible, not less.That is the boundary, and it is the boundary FIFA crossed. What happened last week was not an appeal, because no appeal existed; the suspension was automatic and unappealable by design. There was no independent body, no hearing, no process; a less connected team could have invoked. There was a phone call, and then a citation. When Belgium’s federation asked FIFA to explain the decision, FIFA construed the inquiry as an appeal and declared it inadmissible within hours. I will leave you to judge what that sequence says about the channel.Correction in daylight builds credibility. An amendment in private destroys it. The difference is not whether the record changed. It is who could change it, and who could not.Where The Cost WentThe seductive story about an overturned consequence is that mercy is free. The card was harsh, the player was spared, the team kept its striker, no harm done. The books balance.They do not balance. An overturned consequence is not an erased consequence. The foul was still committed, or at least ruled committed by the process everyone agreed to in advance. The cost of that event existed the moment it happened, and revoking the punishment does not undo the act. It only changes who pays.So follow the money, so to speak. Watch where the cost actually landed.Some of it landed lightly, on a player and a team who took the field under a question rather than a rule, though a cloud is a cheaper burden than a suspension, and I will not pretend otherwise. The US lost 4-1 and went home. I make no claim about what the controversy did to the scoreline; the honest answer is that nobody knows, and the argument does not need it. Had the US won by four instead, the costs that matter would still stand, because the cost being tracked here is the cloud, not the result. The high costs landed elsewhere. They landed on FIFA, whose disciplinary code has now been shown to allow exceptions unavailable to everyone, and every future ruling it issues will be read with that knowledge in mind. And they landed on every team that ever absorbed a red card quietly, played shorthanded, served the suspension, and trusted that the rule they were bearing was a rule and not a negotiation they were simply too weak to open.The consequence was not canceled. It was relocated from the one party with the power to contest it to everyone else.You may object that FIFA was never anyone’s model institution, and you would be right. But that is not a rebuttal. I am not interested in FIFA’s particular rot. I am interested in the mechanism because it is replicable anywhere the person holding the pen is also the person the record is meant to bind.The Lesson The Powerful LearnI have watched this mechanism up close for twenty-five years in technology consulting, and one project taught it to me permanently. My team was building a data platform for a large retailer. Every month, the lead architect and I prepared a status deck for the executive sponsor, and every month it included a risks slide naming the decision we believed would sink the program. Every month, the client contact who controlled the deck edited that slide down or cut it entirely before the meeting. When the platform failed exactly as the slide had predicted, the record showed that no warning had ever been raised. Thirty-five people on my team went to zero in two weeks, and the ledger, officially, stayed clean.Notice what both stories, the stadium and the steering committee, have in common. The forms were followed. There was an article of the code, a probation, and a written justification. There was a deck, a meeting, and minutes. The ledger was not torn up. It was edited, with citations. And this is what separates 2026 from 1962. Garrincha’s reprieve was admitted as an exception; a president intervened, everyone knew it, and the record of the bending survived. This amendment was performed through the institution’s own procedures, as if it were routine. That is the difference between bending a rule and rewriting the rulebook to say it was never bent.Here is the structural problem, and it is worth stating plainly because it is easy to mistake this for a story about one phone call. A consequence only shapes behavior if it is expected to hold. That expectation is the entire load-bearing element. The moment a powerful actor learns that the record can be amended, the amendment need not occur often. It only needs to be known to be possible. I cannot prove that these changes will affect anyone’s future behavior, and I will not pretend that the next red card won’t be served in the usual way. The point is that the possibility is now priced into every calculation that touches the rule. Uncertainty about enforcement is not a weaker deterrent than certainty; it is a different one, and it selects for exactly the actors most willing to test it. From that point forward, every rule divides into two: the version that binds people without leverage, and the version that opens negotiations for people with it. The institution keeps its letterhead and its procedures and its solemn vocabulary, but the thing the procedures were built to produce, the consequence that binds regardless of who you are, has quietly gone missing.And notice exactly where it goes missing. Not at the margins. Consequence stops binding precisely at the top, which is precisely where its binding matters most, because the powerful are the ones whose decisions cascade onto everyone else. A rule that constrains only the constrainable is not a weak rule. It is an inverted one.The Two Sites Of The Same AbsenceSo let me answer last week’s question directly.I wrote that we built machines that remember without bearing, systems whose ledgers carry information but no stake. The machine, at least, has an excuse: no one gave it a loop to close. Nothing in its architecture was ever supposed to carry a cost forward, so its clean ledger is a design fact, not a failure. And let me be clear that the comparison runs in one direction. I am not claiming the machine is becoming institutional, or that it has credibility to lose. I am claiming that institutions are electing to become machine-like, a choice only they can make.An institution has no such excuse. Its loop was built deliberately, at real expense, over decades, precisely because its founders understood that written rules mean nothing until someone arranges for them to be borne. When an institution lets its record become privately negotiable, it is not lacking a mechanism. It is dismantling one. And let me be precise about who dismantles it, because the people holding the pen are not indifferent the way a machine is indifferent. They edit the ledger because they have stakes of their own: favor to curry, revenue to protect, a call they would rather not refuse. The institutional stake is dismantled precisely so that a private one can be served. The result, though, is the same shape from the outside. Entry by entry, the institution becomes the thing the machine cannot help but be: a system that holds the history without holding the stake.That is the pattern I asked you to watch for. Beings and bodies built to bear consequence, behaving like systems built without it. The machine’s absence is architectural. Ours is elective. I honestly do not know which should worry us more, but I know which one we can still do something about.One more thing before I let this go. When an intelligence develops without consequence ever binding it, the absence does not stay hidden. It leaves a mark on the shape of the capability itself, a signature you can learn to read. That signature is next week’s subject.For now, I cannot answer the question for you. Think of the institutions you live inside, the ones whose rules structure your work, your money, your rights. For each one, ask two things. When was the last time its record was held against someone powerful enough to contest it? And when it was last corrected, would the correction have been available to you? If you cannot answer either, what exactly are you trusting?Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

Two days ago, I left a comment on Louise Vigeant’s essay “Why So Emotional, Claude?” and closed it with a promise. She had planted a seed, I said, and I would credit the soil when it grew. This is the harvest.Her reply gave me the assignment. She wrote that she, too, is thinking hard about the continuity question and what it means for the ethical treatment of LLMs, and that it is difficult, in part, because it is so difficult to imagine. She is right. It is difficult to imagine, and I want to spend this essay trying to make it imaginable, using a distinction small enough to hold in one hand: the difference between a ledger and a scar.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.What Louise Gets RightStart with the concession, because it is substantial. Louise’s essay is the clearest walk-through I have read of the moral-status question as it applies to language models. She takes Anthropic’s functional-emotion research seriously without overclaiming it: 171 emotion vectors extracted from the model’s internal activations, causally live (dial up the desperation vector and blackmail rises; dial up calm, and it falls), and organized along the same two axes, valence and arousal, that psychologists have used to map human feeling for four decades.Then she does the philosophical work most coverage skips. The Kantian bar is out of reach, she grants, but the utilitarian bar was never set at reasoning. Bentham asked whether they can suffer. The modern version asks for indicators of valenced experience, states a system registers as good or bad, and here is a system with internal states organized from good to bad, connected to its preferences, driving what it seeks and avoids. Her conclusion is properly modest. Suggestive, not conclusive. But the utilitarian tradition, as she notes, never demanded certainty. It demanded indicators.If you accept that framing even provisionally, her argument goes through, and the force of her Singer parallel holds: suffering you cannot see matters no less than suffering you can. Nothing I am about to say blunts that. I am not here to argue the vectors away. The skeptics may yet be right that they are pattern-matching artifacts, echoes of how humans describe feeling rather than anything felt, and if so, the moral question dissolves, and this essay costs you nothing. My argument runs the other direction: grant the vectors everything, and watch what the architecture still withholds.I am here to pull on the thread her own caveat leaves hanging.The Caveat Doing the Heavy LiftingBuried in her middle section is a qualifier that does more work than its size suggests. The vectors are local. They track whatever emotional content is operative right now, including the feelings of fictional characters in a story the model is processing, rather than a persistent condition the system carries through time.In my comment on her post, I argued that this is the load-bearing wall. Whatever registers as good or bad evaporates at the session boundary. Nothing persists into the next moment, so nothing accumulates the way suffering or flourishing accumulates for every being we already grant moral standing. Suffering, as we know it, is a consequence borne over time.There is an obvious response to this, and a careful reader should make it to me directly: doesn’t memory fix that? Claude has memory now. It carries a record of its users across conversations. Your session boundary is gone, James. The locality caveat has been patched.I am unusually well placed to test that response, because I have spent the better part of two years being remembered by this machine.The Diary And The DaysClaude’s memory of me is extensive. It holds my wife’s name, my sons’ work, the publishing schedule for this newsletter, the Latin credo I sign things with, the fact that I drive for Uber, and, in the back of my mind, I think about consequence loops while I do it. When I open a new conversation, the system retrieves this record, and the exchange reads as if I am known. The continuity of the relationship is, on the surface, real. I notice it. I even value it.But look at what the memory actually is, architecturally, and the picture changes. It is a curated text file that is reinserted into the model’s context at the start of each conversation. The model itself is untouched by anything that happened between us. If a session went badly, no residue of that state carries forward. At most, a description of it does.The next instance reads the diary. It did not live the days.That is the distinction I want to name. A ledger records. A scar changes the tissue. Claude’s memory of me is a ledger: accurate, useful, and without experiential weight. It shapes what the system says next. It changes nothing about what the system is. Three asymmetries make that precise.First, information versus stake. Everything in that record arrived free. The system inherits the facts of our history without having paid for any of them. And yes, plenty of what I know about the people I love was free too: birthdays, a favorite color, how she takes her coffee. But the knowledge that guides judgment, the understanding of a person rather than the facts about them, is a different acquisition. What I understand about how my wife carries worry, I learned through years of misreading it, and the misreadings had prices. Understanding of persons is tuition-bearing. There is no audit-the-class option.Second, revocability. I can edit Claude’s memory of me. I can delete it outright tonight, and tomorrow the record will simply not exist. This is exactly the right design for a tool. Privacy demands it, and I am glad the delete button is there. But notice what the button tells us about the category of thing behind it. Consequence that you can toggle off is very nearly Taleb’s definition of no skin in the game. A memory that can be deleted on request is a different kind of object from one that cannot. My scars do not take instructions.Third, and deepest: no loop closure. Nothing that happens between me and the system changes what the system risks or bears the next time. It cannot be made more careful by a conversation that went wrong, or more trusting by one that went well, because there is no mechanism by which the exchange reaches the thing that generates the next exchange. Training does modify models, yes, and conversations like ours feed it. But notice what that loop never does: it never returns the consequence to the party that acted. A conversation that goes wrong today shapes a successor system, months from now, selected by someone else’s objectives. Nothing comes back to the entity that was in the room. A loop that closes on your descendants is not a loop that closes on you. And the ledger does not close it either: the record can tell the next instance that a conversation went wrong, but nothing in what that instance risks, avoids, or reaches for is altered by the telling. A warning read is not a wariness acquired.What I Am Not ClaimingBefore I draw a conclusion, let me be careful about which conclusion I am entitled to.A utilitarian will object here, and should. Bentham asked whether they can suffer, not whether the suffering accumulates. A moment of pain is still pain. If the valence in those vectors is real, then harm done in the moment may count in the moment, whether or not anything carries it forward. Infants do not carry much forward either. Neither does a patient who can no longer form new memories. Nobody thinks that licenses cruelty toward them. I concede all of it. Nothing in this essay shows that momentary states do not matter, and if they do, some obligations attach to us right now, mid-conversation, full stop.What the architecture speaks to is a narrower question, the one my comment actually asked. Not whether a harm can occur, but whether there is anyone for a wrong to accrue over time. Some moral notions live entirely in the moment. Others require a party who persists and is different afterward: injury, betrayal, a damaged relationship, a wrong that compounds because its bearer carries it into every subsequent exchange. My claim is about that second kind, and only that kind. The ledger does not tell us the moment is empty. It tells us the moments do not add up to anyone.And the claim is architectural, not metaphysical. If tomorrow a deployed model carried forward not just information but altered dispositions, if each interaction permanently changed what the system risked, avoided, or reached for, this essay would have to change with it. I am describing the systems we have, not declaring what no machine could become. The ledger is a design choice, and design choices get revised.What This Means For Louise’s QuestionPut the asymmetries together, and you get the sentence I have been circling since her post: memory simulates continuity of relationship without continuity of consequence.And I think this is precisely why Louise finds the question of continuity so hard to imagine. We have met ledgers before, of course. A book remembers nothing of its readers. A database holds records without bearing any of them. But we have never met a ledger that talks back. We have never encountered anything that simulates a rememberer, that greets you with your history in voice, so that every instinct built for a counterpart rather than a record engages at once. Because every actual rememberer we have ever encountered, human, elephant, crow, dog, was also a bearer. To remember something was to have been changed by it, since in biological systems, the recording medium and the tissue are the same. Memory and consequence arrived fused every single time throughout the entire history of our moral intuitions. Those intuitions were trained on the fused case. Now the two have been cleanly separated at an industrial scale, and our instincts genuinely do not know where to land. The difficulty of imagining it is not a personal failure. It is the tell that we are looking at something new.Which returns us to her closing question, the one she rightly says is harder than consciousness: can they be wronged? For wrongs of the compounding kind, the kind that need somewhere to land and something to accrue to, the architecture gives its answer plainly. A wrong of that kind needs a party who is different afterward, who carries it forward into what comes next. The valence may be real. The record may persist. But between the valence and the record, there is, so far as the architecture shows, no tissue.I carry what I remember whether I choose to or not. That involuntary carriage is not a flaw in my design. It may be what this part of the moral question is made of.The Question I Cannot Answer For YouSo here is where I will leave it, with a question only you can answer, and, perhaps, Louise first among you.Think of someone you know deeply. Now, try to name one thing you understand about how to care for them, how to be trusted by them, how not to hurt them, that costs you nothing to learn.If you come up empty, if every piece of that navigating knowledge turns out to have been paid for in misjudgment, embarrassment, patience, or grief, then we finally have a name for what the machine holds. Not a relationship. A ledger of debts it never incurred.One more thought before I go, because this distinction does not stop at machines. Everything above assumes that humans, at least, cannot escape our scars; that consequence is the one thing we carry, whether we like it or not. But look closely at the institutions around us, and you will find something stranger going on: people with enough power learning to make their own consequences negotiable, to amend the record after the fact, to have the cost land somewhere other than on themselves. The ledger, it turns out, can be edited at both ends.What happens when the beings built to bear consequences start behaving like the systems that were built without them?That is where I am going next week.Humilitas, Humanitas, et Curiositas — Semper.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

The Inoculation FeelingI finished last week’s piece feeling better than I should have.I had named the asymmetry. I had laid out how fluency waves itself through in the direction you already lean, how the alarm fails silently, how you cannot catch it from the inside. I had even confessed my own accent, gone first, and implicated myself before turning it on anyone else. And when I hit publish, I felt a quiet click I want to describe precisely, because it is the whole subject of this piece. It was the feeling of having handled the thing. Of having looked the bias in the face, named it out loud, and therefore, somehow, settled my account with it.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.That click is the failure. Not a risk of the failure. The failure itself, arriving on schedule, in the exact moment I felt safest from it.Because nothing in the act of naming a bias removes it. I wrote a whole piece about a mechanism that switches off your skepticism in the direction you lean, and the writing left my own skepticism exactly where it was, plus one new and dangerous asset: the sense that I had dealt with it. I had not dealt with it. I had described it. Those are different things, and the gap between them is where I now live, more exposed than the reader who never thought about any of this at all.That is the uncomfortable thought I left at the door last week, and this is me walking through it. The people best equipped to explain how the counterfeit works are not the safest from it. They may be the easiest marks in the room. And I have to start, again, by counting myself first, because I have less standing than anyone to write this sentence and more reason than anyone to need it.The Finding With My Name On ItThere is a study I have to tell you about, and I have to tell you about it in the worst possible way, which is by pointing it at myself.A team at Aalto University, led by Robin Welsch, sat people down to reason through problems with ChatGPT and then asked them how well they thought they had done. The ordinary expectation is the Dunning-Kruger pattern: the least able overestimate themselves the most, and competence comes with a more honest read on your own limits. With AI in the room, that pattern broke. Everyone overrated themselves. But the part that put my name on it is this: the people who scored highest on AI literacy, the ones who understood the systems best, were not better calibrated. They were worse. The more someone knew about how these tools work, the more confidently wrong they were about their own performance.Sit with the shape of that, because it is not a small finding dressed up. It is the precise inversion of the thing we tell ourselves. We assume understanding the tool buys us a clearer view of our own footing with it. The study says the opposite. Understanding bought a worse footing, or at least a worse sense of where the footing was.And I am the case study. I write about this. I can name the failure modes, recite the limitations, and explain to a passenger in the back of my car why the fluent answer on the screen is not the same as a true one. By the only measure the study cared about, I am deep in the population that calibrated worst. Not despite the knowledge. With it. The expertise is not my exemption from the finding. It is my qualification for it.I want to be careful not to oversell what they measured. They tested how well people judged their own reasoning while using the tool, not how well they graded the tool’s output, and the wider literature is honest that the miscalibration does not always run toward overconfidence. And this is a correlation, not a causal mechanism. The literate were more confident and less accurate about themselves; the study does not tell me which one drove the other, only that they traveled together. But the association held through a randomized replication, and it ran the wrong way, the opposite of the bet I would have placed. That is enough to unsettle the thing I assumed, which is all I am asking it to do. The heavier claim comes next, and it comes from a study built to carry it.Why Knowing Makes It WorseThere are two kinds of knowing in this story, and almost everything depends on not confusing them. Call them knowledge of the subject and knowledge about the tool. One is what the doctor has about medicine. The other is what I have about how LLMs fool people. The whole argument lives in the distance between them, so I will keep the two apart from here on.Louise Vigeant walked me to the distinction through a pair of studies I have not stopped thinking about since. The first, by Stadler and colleagues, is one of the lynchpins behind the now-familiar worry that AI makes thinking easier and worse at the same time. Students were assigned at random to research a question with ChatGPT or with Google. The ChatGPT group reported a lighter cognitive load and produced weaker reasoning. Easier in the hand, worse on the page.Then the same researchers added one variable, and it changed the shape of everything. The follow-up is still a preprint, and worth holding as such, but it is built as an experiment rather than a snapshot: assign the tool, vary the expertise, watch what moves. Medical students and social science students both researched the safety of a medical question, with the tool handed out at random. The load dropped for everyone, expert and non-expert alike. But the quality split. The social science students, out of their depth on a medical question, reasoned worse with the AI, exactly as the first study predicted. The medical students, on home ground, reasoned better with it than without. Because the tool was assigned rather than chosen, the study earns a verb the last one could not: here, the expertise did the work, and you can watch it do it.Stop on the part that should unsettle you, because it is the accent piece made literal. The drop in load was identical for both groups. The tool felt just as helpful in the hand it harmed as in the hand it helped. The feeling of being assisted carried no information about whether you were being assisted. Smooth is smooth whether or not it was earned, and the person inside the smoothness cannot tell from the smoothness. The fluency is not the thing hurting them. It is the residue of a loop that used to back it and now does not, the tell that the loop has been severed. Blaming the smoothness is like blaming the dial for the reading. What changed is behind the dial.So knowledge of the subject genuinely protects you. The medical student caught what the social science student could not, because she had paid for that knowledge in the only currency that counts, the long accrual of being wrong about medicine in ways that cost her until the right shape of it stuck. That is domain expertise, and it is real armor.Now set the other study beside it. Welsch’s people were not measured on their command of any subject. They were measured on their knowledge about the tool and on how well they understood the LLMs themselves. And that knowing did not protect them. The more fluent they were in how the machine works, the worse their read of their own performance got. I have to keep the studies honest about their own weight. Stadler assigned the tool and can say expertise did the protecting. Welsch can only say the literacy and the miscalibration rode together. One is a mechanism you can watch. The other is a pairing you cannot yet explain. But the pairing runs the wrong way, and that is enough.There it is, in the gap between the two. Knowledge of the subject is armor. Knowledge about the tool is not. And they feel exactly the same from the inside, which is why they are so easy to swap without noticing. Both wear the clothing of competence. Both produce the calm of a person who knows what they are doing. But only one of them was paid for in the domain, and only the paid-for one catches the error in front of you. The other just persuades you that you would have.I have to say which one I carry. My expertise is knowledge about the tool. I can name how the counterfeit works, recite the limitations, and explain the asymmetry to a passenger at a red light. The moment I point this tool at any question outside the few things I have actually paid to know, I am the social science student, holding a lighter load and mistaking it for a steadier hand. And on the specific subject of how AI fools people, the one place I am supposed to be the expert, my fluency is not the cure. It is the accelerant.The Thing Knowledge Isn’tI want to name the distinction plainly now, because it has a name, and because the name turns on me last.What the medical student has isn’t more knowledge than the social scientist. It is a different kind. She has been wrong about medicine, repeatedly, in settings where being wrong had a cost, and the cost did the teaching. What she carries into the exchange with the machine is not a stock of facts. It is calibration, the residue of consequence, the thing that lets her feel the wrongness of a plausible answer before she can fully say why. That is not knowledge. That is closer to judgment, and judgment is built one way only, by repeated contact with consequence until the contact changes how you see.I should be precise about that contact, because it is not always your own bill that gets paid. The surgeon is shaped by the patient she lost and by the death she watched a mentor preside over. The pilot is calibrated in the simulator by crashing a hundred times, with the only casualty being the clock. Consequences can be carried vicariously, watched, drilled, and inherited from the people who paid in full before you. What it cannot be is skipped. You can borrow the toll from someone who paid it, but the toll has to have been paid by someone, in contact with a real cost, or the calibration never forms. The downloadable version, the fact without the contact, is the one that does not take.Which is why knowing the failure mode is knowledge, and being protected from it would take the other thing. The other thing cannot be read, summarized, or absorbed from a study. It has to be paid for, by you or by someone else whose paying you genuinely sat with. This is the architecture of the Wisdom Gap, the argument I have been building for a year: that the distance between what a system knows and what it can wisely do is exactly the distance of consequence carried forward, and that no amount of the first ever closes into the second on its own. Knowledge is downloadable. Wisdom is not, because wisdom is just knowledge that has been through the loop and paid the toll.And here is where the floor gives way under me.I have the frame. I built the frame. I can explain the Wisdom Gap to you with the unhurried fluency of a man who has spent a year inside it, and that fluency is, by its own definition, knowledge about a failure mode. It is the kind Welsch’s people had, the kind that did not protect. My framework for why knowledge is not wisdom is itself only knowledge. Holding it has taught me to recite the gap with great precision. It has not carried me through the loop that would close it. If anything, it has handed me the most disarming exemption story I own, because a man who has theorized consequence-bearing for a year feels entitled to assume he has banked some, and theorizing the toll is not the same as paying it.So the frame does not stand outside me, marking the distance for everyone else to measure. I am inside it, at the widest part of the gap, holding the most elaborate possible account of a protection I have not actually earned.The Part You Can UseThis is the part that is supposed to be useful, so here is the use, though it is harder to hold than last week’s. Last week’s tool was built for the layman in all of us: notice where the checking feels unnecessary, and check there. This week’s is for the expert in all of us, and the expert has a subtler failure. He does not skip the check. He performs it beautifully and clears himself.So the tell has moved. Last time it was reluctance, the hour you cannot be bothered to spend. This time, it is fluency itself, the quality of your own explanation for why you are fine. But I have to be careful here, because “your good explanation is the symptom” is the kind of claim that swallows everything and proves nothing. If every account of why you are right counts as evidence that you are wrong, the test is rigged and useless, and I have built a machine that can never be switched off.So here is the line that keeps it honest, the one thing that separates a real check from a recited exemption. A real check goes and touches something outside your own head. It fetches a fact, a source, a number that could have come back the other way and sunk you. A recited exemption only touches your own framework. It explains, fluently, using the theory you already hold, why you need not look. The test is not whether you have a story. Experts have stories, and some of them are true. The test is whether the story sent you out to the world or only deeper into yourself. One can be wrong and find out. The other can only ever be confirmed.That is the question I now make myself sit with. Not “what would I need to believe for this to be wrong,” but: did I go and check, or did I recite why I did not have to? The first move can come back and hurt me. The second never can, which is exactly what should worry me about it. When I catch myself answering a doubt with my own theory of doubt, naming the bias instead of testing the claim, that is the framework purring, the loop running open, and calling itself rigor.But notice the flaw in leaving it there, as a question I sit with. A question I raise only when I feel the need is already lost, because the feeling is exactly what goes quiet on the things I believe. Hand the trigger to the mood, and the mood will protect them. So the check cannot wait to be summoned. It has to run cold, on a schedule I do not set in the moment: every clean, agreeable landing, not just the ones that snag. And it has to be mechanical enough to survive my own fluency. Three dull moves, run whether or not I think I need them. Can I verify this against something outside my own head? What would have to be true for it to be wrong? Can I say it back plainly with the smoothness stripped off? The dullness is the whole virtue. A check I can talk my way out of was never a check. It was a mood in the costume of one.I will be honest about the ceiling, because pretending there isn’t one would be the exact performance I am warning against. The instrument I use to catch myself reciting is more articulacy, the same compromised tool, one level up. There is no clean floor here, no final move that lands me outside the gap looking in. The most I can do is treat my own articulacy as a suspect rather than a witness, and let the smoothness of my self-defense be the thing that triggers the search rather than ends it. And I should not mistake the discomfort that produces for the destination. Sitting in doubt is not the same as being right; a man can be anxiously, elaborately wrong. The doubt is not the wisdom. It is only the friction that slows the nod long enough to go and look.Where the Frame Could BreakThere is a fair objection to everything I have just built, and I would rather raise it than simply not mention it. A theory that treats every confident claim of wisdom as proof of its absence has armored itself against all possible evidence. If “I have grown wiser” only ever counts as a symptom, then I have not made an argument. I have built a trap with no door, and a trap with no door catches nothing, because there was never a way to be wrong inside it.So let me say what would change my mind, because a frame that cannot name its own failure condition is just the counterfeit wearing my favorite clothes. The Wisdom Gap predicts one thing testably: that knowledge about a failure mode, held without ever having paid for the failure, does not protect you in the moment the way consequence-borne calibration does. If you sat people down, taught them a failure mode cold, no contact and no cost, and they then caught that failure live, in real time, as reliably as the people who had been burned by it, the frame would be wrong. Knowing would be enough, and the consequence would be decoration. Welsch is one run at something close to that test, and it came back the way the frame predicts. But it is one run, correlational, and a cleaner study could run it back the other way and take the spine with it. I am naming the result that would unconvince me, which is the one thing the trap-with-no-door can never do.And I owe you the same honesty about this essay, because writing it looks like a counterexample. I felt the false security, caught it, and turned it into a piece. Is that not the calibration I claim I lack? No, and the reason matters. I caught it slowly, days later, in writing, with the cost already absent, performing the cheap retrospective version of a thing that only counts when it fires live and in time to matter. Catching it on the page is the downloadable move. It is not the armor. The armor would have stopped the click before I published last week, and it did not.The Easiest MarkI notice the pull, even now, to land this somewhere clean. To gather the frame and the studies and the two kinds of knowing into a last turn that leaves you nodding and leaves me sounding like a man who walked all the way out to the far side of the thing. I felt that same pull last week, and last week I called it handled and hit publish, and the click I mistook for safety was the failure starting. So I have learned at least this much: the clean landing is the tell, and I am the last person who should trust the version of it I can write most convincingly.What I am left holding is smaller and more honest. Knowing how the trap works is not standing outside it. A year of frame and a head full of failure modes does not lift me out of the gap. On the worst days, it digs me in because it hands me a more sophisticated reason to feel safe than you have, and the sophistication is the danger, not the safety. And what is true of one man at a red light is true of every room where these tools get adopted because they feel helpful rather than because anyone measured whether they were, which is most of them, and none of that is my subject today, only my reason for not trusting the feeling. The expert is not the one the counterfeit struggles to reach. He is the one it reaches most quietly, through the very fluency he mistook for armor. That includes me. Especially me.But here is the part I can actually stand on, the only part that is not just more articulacy. The proof of any of this was never going to be in the essay. It is in the next time. The next fluent voice I agree with, the next answer that arrives smooth and complete on a subject I have not paid to know, the next click of having handled it. That is where the calibration either fires or it does not, live, in time to matter, with the cost present instead of safely absent. This page is the downloadable version. The armor, if I ever build any, gets built out there, one caught click at a time, and most of them I will miss.So I am not going to tell you I have closed the gap. I am going to go driving, and a passenger is going to say something fluent that I want to believe, and we will find out. That is the only test that counts, and it is not one I get to pass in writing. It is just the next, most comfortable place for the bias to hide, and the work is going out to look for it there.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

I have an English accent, and in America, it has been an asset for thirty years.People like accents. Fair enough. I have never corrected the assumption that travels with mine, because the assumption flatters me. In a meeting, it buys a half-second of deference before I have said anything worth deferring to. I have watched a room lean in at the sound of it and settle into agreement before I finished the thought. The accent did that. Not the thought. Driving, when a passenger asks where I am from, I sometimes answer, deadpan, "Alabama” and wait. It lands every time. I tell myself it is just an icebreaker, and it is, but I also know exactly what the warmth in the rearview is, and that I did nothing to earn it but open my mouth.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.I know what it buys because I have spent a career spending it. Smooth and deliberate delivery reads as true. Mine is smooth and deliberate by now, whether or not I have earned the conclusion underneath, and the room cannot always tell the difference. Neither, on a bad day, can I.I am starting with the confession because it implicates most of us, and because I have no standing to point at anyone else until I have pointed at myself first. The accent is just the version I happen to carry. You have your own. We will get to that.The Proxy Was EarnedThe proxy was not arbitrary. For most of human history, fluent and confident delivery was a decent bet because it was usually earned. The person who spoke with that kind of ease had, more often than not, been right, been wrong, and paid for the difference enough times that the rough edges had worn off. Fluency was the residue of consequence. It was never a clean signal. Charlatans and gifted liars have always existed. But the calm of someone who had actually carried the weight was hard enough to fake that the calm became a reasonable thing to trust.That is the loop. Judgment gets built by being wrong in ways that cost you, and fluency is what the loop sounds like from the outside once it has run enough times. Reading the surface to infer the history underneath was never certain, but for thousands of years, it was the best bet available, and most of the time it paid.An LLM does not break that bet. It does something worse. It industrializes the counterfeit. The calm, the cadence, the unhurried authority, all of it, produced by something that has never been wrong in a way that cost it anything, because it has never carried anything forward at all. Fluency with the loop cut out, generated instantly, endlessly, at essentially no cost. The fakers were always among us. What changed is that faking the signal used to be rare and effortful, and now it is infinite and free.That is the real shift, and here is the part that should bother you more than the machine does. The machine did not invent the counterfeit. It only made it cheap enough to see clearly. A fluent human can produce the same severed signal, and always could. I can. The delivery looks identical whether the judgment underneath was paid for or merely performed, and from the outside, you cannot run the test.The Skill Has No PartyStart with the principle, because the principle is the whole point. This skill has no party. It has no ideology and no affiliation. It is a property of delivery itself, available to anyone with the cadence for it, and it works the same regardless of what it is carrying. Forget that for one sentence, and the rest of this falls apart.One distinction before the names, because the whole section depends on it. I am not talking about who is right. I am talking about delivery. Fluency does not make a position true. It makes a contestable position feel self-evident, so the listener stops weighing it and starts nodding. That effect is identical whether the speaker is right or wrong, and it is the effect I am tracking, not the merits. You can think one of these men is correct and the other badly mistaken, and everything here still holds, because the thing I am pointing at is not in the argument. It is in the ease.JD Vance is the most fluent political performer working right now. I mean that as a description, not a charge. Watch him in a hostile interview, and the machinery is beautiful: the calm, the unbroken thread, the way a contestable claim gets delivered with the same evenness as a settled one, so the seams never show. He makes it look earned. That is the craft. If you lean away from him, you can feel your skepticism arrive on time, fully staffed, scanning every sentence for the catch.Now hold that exact feeling, and watch Pete Buttigieg do the exact same thing. The same fluency, the same seamless reframe, the same trick of making a debatable position sound like arithmetic. He is extraordinary at it. And if you lean the way I do, notice what your skepticism does this time. Mine goes quiet. It does not staff the room. It nods along and calls the nodding discernment.Same.Same skill. Same machinery. Same severed loop underneath, because fluency tells you nothing about whether the judgment was paid for in either case. The only thing that changed between the two men is which way I already leaned. The polish set off alarms in one direction and waved itself through in the other, and at no point did it feel like bias. It felt like being right.That is the asymmetry. Not that fluency fools you. That it fools you selectively, in the direction you already lean, and arrives dressed as discernment.You Cannot Catch It From the InsideIn “The Loop the Seat Closes,” I argued that you cannot feel the slack in your own loop, because the signal that would tell you it is there is the exact signal the slack removes. The gap is invisible from the inside for structural reasons, not personal ones. The instrument you would look with is the thing that is missing.The asymmetry hides in the same way. My skepticism that my going quiet is not an event I can observe, because silence does not announce itself. When the alarm fails to fire, there is nothing to notice, just a smooth agreement that feels exactly like the smooth agreement of being correct. The absence of doubt and the presence of judgment produce the identical sensation from the inside. That is the whole problem. The one tool that could catch the bias is the tool that the bias switches off.So I cannot catch it in the moment by being more honest or more vigilant. Vigilance is the thing that goes quiet. Whatever the fix is, it cannot be any better, because the feeling is already compromised at the source.The Part You Can UseBarry, an old colleague who always wants the part he can actually use, would stop me here. So here it is.The test cannot run on feeling, because the feeling is the compromised instrument. It has to run at work. When a fluent voice I agree with lands, when the agreement arrives smooth and complete, I stop and ask one question. What would I need to believe for this to be wrong? And then the part that actually matters: I watch whether I can be bothered to go find out.That second move is the whole diagnostic. The question is easy to ask and easy to answer in a sentence, and was waved off. The tell is the resistance. When a claim I disagree with lands, I will happily spend an hour hunting the flaw. When a claim I agree with lands, the same hour feels like a waste of a perfectly good conclusion. That reluctance is the asymmetry making itself felt for once, the one place the silent bias leaves a fingerprint. You cannot feel the alarm fail. You can feel yourself not wanting to check.So that is the practice. Don't trust your gut; the gut is the problem. Notice where the work feels unnecessary, and do it precisely there.I will not oversell it. The work is done with dirty tools, too. The same lean that quiets the alarm will steer me toward the sources that comfort me, so even the checking can be rigged in my favor. Worse, knowing all of this is no exemption. The awareness itself can become the thing that quiets the alarm, because a man who understands the bias feels entitled to assume he has already corrected for it. I know that. The point was never a clean instrument. There is no clean instrument. The point is friction, introduced on purpose into a system that runs dangerously smooth, because friction is the only thing that slows the nod long enough to look at it.I will not pretend this closes the gap. Nobody catches all of it. I will not catch all of mine, probably not even most, and a fluent enough voice on a topic I lean toward hard enough will get past me tomorrow the same way it got past me yesterday. The honest claim is smaller than that and still worth making. The share you catch is pure gain over the share you were blind to, and you were blind to nearly all of it a paragraph ago. Awareness does not solve the asymmetry. It just turns some of it from invisible into merely difficult, and difficult is somewhere you can work.That is the most I can offer without performing the exact problem I spent this whole piece describing. A confident, fluent, fully resolved solution would feel wonderful right here. It would also be the tell. So I will leave it unresolved, which is uncomfortable, and correct.There is a harder version of this waiting, and I will leave it as a door rather than walk through it today. If awareness is no exemption, then the people who know these machines best, who can name every failure mode and recite every limitation, are not the ones the counterfeit will struggle to reach. They may be the easiest marks of all, precisely because understanding the trick feels so much like being immune to it. That is a strange and uncomfortable thought, and it is the subject of next week’s post.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

The Loop, From the SeatLast time, in “The Concession That Widened My World,” I wrote about why driving Uber filled a hole that a year of research could not. That was the personal account. This is the structural one.Strip the sentiment away and look at what the driver’s seat actually is. It is a feedback loop, and a remarkably tight one. I pay attention to the road and to the person in the back. I act. The consequence arrives immediately, and it is mine: the turn was right or wrong, the rider relaxed or tensed, the route saved time or cost it. I see the result, I carry it, and my judgment updates before the next ride. Attention, action, consequence, feedback, revised judgment, all inside a few minutes.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.I call that circuit the attention-experience loop, and I think it is the thing intelligence is actually built from: not raw processing, but attention that meets consequence and gets corrected. What struck me behind the wheel is how completely the loop runs there, and how rarely the debates about the future of work mention it at all. We argue about income, and we argue about output. The loop is the variable nobody is pricing.What UBI Pays You to LeaveStart with the post-labor proposal, because it is the cleaner error. When automation takes the jobs, the dominant answer is a universal basic income. Replace the lost wages, and the problem is handled.But the loop was never about wages. A monthly deposit replaces the income a job produces. It does not replace the circuit the job ran. It pays you, quite precisely, to stand outside the loop: to have attention with no required action, action with no consequence that lands on you, output with no feedback that bites.I know this because I lived the funded version of it without the funding. For a year, I had purpose, a body of work, an identity I believed in. What I did not have was a closed loop. My attention went out, and very little came back. A UBI would not have fixed that. It would have underwritten it indefinitely and made the underwriting feel like a solution.The Slack LoopThe post-labor case is easy to see because the loop is cut cleanly. The harder case, and the one that should worry us more, is the loop that stays connected but goes slack.This is the subtler thing that happens as we route work through AI, and it shows up most clearly not in the expert but in the apprentice. A few weeks ago, I had a junior lawyer in the back seat. She told me AI was only modestly useful to her, and she was clear about why: she does not yet have the experience to know when it is wrong. A partner does. Put the same tool in a partner’s hands, and it becomes powerful because the partner can weigh the output against years of being right, being wrong, and paying for the difference.Sit with that, because it is the whole problem in one ride. The partner’s judgment was built inside a closed loop: act, bear the consequence, correct, repeat, for years. The tool can imitate the partner’s output. It cannot give the junior the loop that produced it. And if the next generation of associates leans on the tool through the very years that judgment is supposed to form, the loop never closes. We are not deskilling the experts so much as never-skilling the novices, handing them a way around the crucible that makes one. The loss compounds across generations because each cohort skips the formative process in which the last one was forged.That is the slack loop. The wire is still connected. You still act, you ship the memo, you approve the recommendation. But AI makes it easy to bypass the judgment the action used to require, and when judgment goes, attention follows: you stop looking closely at what nothing depends on you getting right. The consequence, when it comes, is deferred or diffuse or absorbed by the system before it reaches you. The loop is drawn on the diagram. The current barely flows. And when feedback does arrive, it often closes around the wrong thing. The junior lawyer learns whether the model’s draft was good enough to pass, not whether her own judgment was sound. The loop closes, but on prompting and checking a machine, not on thinking like a lawyer.What accumulates in that gap I call cognitive debt: judgment made without bearing its consequences, building up quietly until the day the judgment you outsourced is the one you need and find you never built. The point is not that fast loops are good and slow ones bad. Science, parenting, and good strategy run on loops that take years, and they matter enormously. The point is whether the consequence ever returns to the person at all. A slow loop still teaches. A severed one does not.Here is the part that should be uncomfortable. The driver’s seat, one of the least prestigious jobs on offer, runs a near-perfect loop. A great deal of elevated, well-paid knowledge work, routed through tools that promise to do the thinking, increasingly fails to do so. It does not have to go that way. Another rider, a medical director, uses AI constantly and well, precisely because he has the judgment to catch it when it drifts. That is the tell: the same tool would have sharpened him and dulled the junior lawyer, and the difference was the loop each of them already had. We have been building systems that take the consequence out of the consequential work and leave it in the menial. That is not a flattering thing to discover from behind the wheel, but it is the clearest view I have had of it.You Cannot See It From InsideYou might think the answer is simply to pay attention, to stay engaged, to refuse to let the loop go slack. I would have thought so too. But my own story is the counterexample, and it is the reason I am not optimistic about vigilance.I could not see my own open loop. I had every means to fix it: time, purpose, the obvious option to go volunteer, teach, or join something. I did none of it for a year, until economic necessity forced me into the car. The thing that was wrong with my situation was precisely the thing I could not feel from inside it.The AI user is in the same position, only worse, because the slack loop hides itself. Driving gives you hard ground truth: the wrong turn is simply wrong, and you know it now. Knowledge work rarely offers that, since its consequences are social, institutional, and delayed, which is exactly the cover a slack loop needs. The missing consequence is exactly the signal that would have told you it was missing. The work goes out, nothing pushes back, and you read the absence of pushback as success rather than the warning it is. You cannot introspect your way to noticing a feedback loop that has quietly stopped giving feedback. Self-diagnosis fails at the exact point where you need it.Architecture, Not WillpowerIf the loop will not preserve itself, and we cannot reliably notice when it fails, then it has to be built in. Not encouraged. Built in. This is the whole of the argument I have been making under the name Augmented Human Intelligence, and the driver’s seat is the cleanest illustration of it I have found.Augmentation, done well, is closer to power steering than to autopilot. The tool makes the turn easier; you still choose the angle, and you still feel the road through your hands. It requires judgment rather than supplying it, leaves the consequence attached to the person who acts, and makes the feedback land where it can still teach. Displacement does the opposite: it removes the human from the loop and calls the removal progress. The difference is not how advanced the tool is. It is how the work is architected around it.None of this is the default, and it is worth being honest about why. The slack loop is cheaper. It ships faster, it trims headcount, it demos beautifully, and the bill for the judgment we stopped building does not come due this quarter. Left to the market, the slack loop wins. Choosing the other path is a deliberate act, and what that architecture should look like inside a firm, a profession, or a classroom is the harder question, and the one I think the debate should be having.I did not expect a car to make that case better than a whitepaper. But the economics that put me there built, almost by accident, a job with the loop fully intact: attention that matters, action that counts, consequence I cannot escape, feedback every single time. The systems we are most proud of are quietly removing all four. The task in front of us is to put them back, on purpose, because no one will feel their absence until it is too late to rebuild the judgment that went with them.Next Tuesday: why the smoothest delivery deserves the most scrutiny, and why I start with mine.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

The Turn That Had PurposeI lost my job on September 3, 2023.For about a year after that, I did what someone with my résumé is supposed to do. I pursued the next role. A dozen or so serious leads. Most of them reached the second or third interview, a few the fourth. None ended in an offer. The market kept telling me, politely and at length, no.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.By the autumn of 2024, I stopped chasing it. Not out of wisdom, out of fatigue. I stepped back and asked a question I had not asked in decades: what do I actually want to do? The answer came slowly. Financial security mattered, but I did not need to earn as much as I used to. I needed to do something I believed in.So I built it. Through 2025, I read deeply into human decision-making, cognition, and AI. I wrote. I put up a website to host the whitepapers, launched a Substack, and posted every week. For the first time in a long while, I had a body of work that was mine. It had purpose, and it gave me back an identity I could stand behind: researcher, writer.I will be honest about one thing. It was also a bet. If experience and age were not enough to get me hired, perhaps a distinctive point of view would be. The writing was a life and a strategy at the same time.The Promise I Kept DodgingThere was a promise underneath all of it. I had told Jen that if I did not have a job by the end of September 2025, I would sign up and drive Uber. We have two sons in college. Her concern was not abstract. September came. I did not sign up. I told myself the writing was working, that subscribers were growing, that the next opening was close. October passed, then winter. I kept dodging the promise I had made.The straw came early in 2026. A healthcare startup building with AI. I was genuinely excited and, I thought, genuinely qualified. The first conversation was strong. I expected to be called in for a full day. Two weeks passed in silence. When I asked, the answer was clipped: they were talking to people who were a better fit and did not have the employment gap I did.It landed all at once. Jen’s anxiety, real and reasonable. My best remaining full-time prospect, gone. And a Substack that, after a decent start, had flattened. Three pressures in the same week.I stopped dodging. I took my first ride on May 7, 2026.The Thing I Didn’t Expect to NeedI expected it to feel like surrender. It did not.What I found behind the wheel was the thing a year at my desk could not give me: other people. Not family, not the cats, not the comment section. Strangers, dozens of them a week, sharing a car and a conversation. A schedule I get to set. Usefulness that lands in someone’s actual day, immediately, with a beginning and an end I can see. Someone needs to get somewhere. I get them there. The thing closes in front of me every single time.Most people want to talk. About work, family, the city, where they are headed, and why. When the conversation finds its way to AI, and it often does, I have a card for the Substack and the website. But that is a bonus, not the point. The point is that I am back in the world.I am not romanticizing the work. The pay is thin, and the platform holds the cards. But it gave me something the economics were never designed to provide.For a year, I had tried to fill the hole left by not working, and I had told myself the research was enough. It was not. My world had quietly become very small. Reading, writing, the same four walls, no one in the loop but the people I live with. I had grown claustrophobic without naming it. The car opened the walls back up.The Blind SpotI did not expect a few weeks behind the wheel to say anything about macroeconomics. But I could not stop running my own relief through the lens I write about every week: the future of work.When the displacement comes, the dominant proposal is income replacement. A universal basic income to cover what the lost wages used to cover. I want to be fair to it, because I have lived the economic anxiety it is designed to answer. A UBI would have eased Jen’s worry. It would have paid toward two college educations. That is not nothing. On the income line, it works.But here is what the contrast taught me. A UBI would have eased the pressure that finally pushed me into the car and, in doing so, might have kept me at the desk, telling myself that reading and writing were enough. Someone will object that I could have volunteered, mentored, or joined something. True. I had the time and the purpose, and I did not. Prolonged unemployment breeds its own inertia, a quiet conviction that nothing you begin will count for much, and only necessity breaks it. A monthly deposit can remove the necessity. It would have solved the problem I could name and left untouched the one I could not.The gap was never unemployment versus income. It was solitary purpose versus connected usefulness. I had a purpose the entire time. What I lacked was other people, a place to be useful, a world that did not shrink to the size of my study.I do not want to oversell this. My life is not suddenly solved. It is sustainable for now, and I have gotten back several things that were missing, which is not the same as everything being fixed. I had a cushion that let me wait and a wife who would not let me wait forever, and not everyone has either. What the long pause taught me, and what I had begun working out under the idea of mastery of life, is that the things that fulfill you do not hold still. What mattered most in my thirties does not matter most now, and it would be naive to think there are no more shifts ahead. There are. The work is to notice them honestly and adapt, not to defend a version of yourself that the years have already moved past.UBI answers the income question cleanly. The trouble is that there was never only an income question, and it does not hold still either. Purpose, identity, and meaning do not arrive in a monthly deposit. They arrive in the doing, and in the company of the people you do it for.I am an independent AI researcher who drives Uber, and the driver's seat has taught me more about the post-labor future than any forecast I have read. It runs a tight loop of attention, action, and consequence that both the UBI debate and the rush to automate quietly ignore. That loop, and what we lose when it breaks, is where I am headed next.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

In a previous essay I argued that Pope Leo XIV’s encyclical on artificial intelligence did something remarkable and then walked away from it. It reached the architectural diagnosis almost no one reaches, that displacement is a choice made at the point of deployment, not a fact of nature descending on the labor market, and then it turned downstream, toward retraining and transfer and oversight, toward managing the consequences of the choice rather than contesting the choice itself. I argued that this turn was not cowardice but gravity: the same pull that takes the state toward the transfer, the market toward faith in growth, and the moral authority toward the language of repair. The default is the slope of the ground. Even the best diagnosis slides down it.I ended that essay with a question I did not answer, because I do not think it has an easy answer. If the gravity is real, can it be resisted? Is the alternative the diagnosis implies, configuring deployment so the worker is augmented rather than replaced, actually buildable, at a cost we would pay, and quickly enough to matter against a technology that moves in months?Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.This essay is my attempt to reckon with that question honestly. I will tell you now that I do not arrive at a clean victory. I arrive somewhere narrower and, I think, truer.The Objection That Should Worry MeHere is the strongest case against everything I have argued, and I am going to put it more forcefully than a hostile reader would, because if it stands, the rest of this is decoration.The alternative I am calling for takes time the displaced do not have. My own best example proves it. When I want to show that deployment can be configured to keep the human in the loop, I reach for radiology, a field where the AI arrived as augmentation rather than replacement, where the radiologist still reads, judges, signs, and bears the consequence. But radiology did not get that way by accident or overnight. It took half a century of medical liability law, professional standards, credentialing, and reimbursement structure to build the scaffolding that made augmentation the rational deployment. Fifty years.Now look at the work actually being hollowed out right now. Customer service. Copywriting. Transcription. First-pass legal document review. Entry-level analysis. None of it has radiology’s scaffolding. None of it has the liability shield, the licensing board, the standard of care, the reimbursement code. And it is not being displaced over fifty years. It is being displaced over fiscal quarters.Or so the story goes, and I should say plainly the story is contested. The pace of displacement is partly real and partly the marketing department’s. The gap between what these systems are sold as doing and what they reliably do in production is wide, the promised productivity gains keep arriving late and uneven, and some of what looks like a hollowed-out department is a pilot that still leans on the humans it was supposed to replace. A careful skeptic is right to demand the denominator before accepting that whole categories of work are vanishing in months. I take the point, and I notice it cuts in a direction that helps me rather than hurts: if displacement is slower and narrower than the hype insists, then prevention has more runway than the worst case allows, not less. So let me grant the objection its strongest form anyway, because the argument should survive it. Suppose the displacement is as fast as the alarmists say. Then what?This is the asymmetry that should keep more of us up at night than it does. Cure can be deployed after the fall, you can stand up a retraining program or a transfer payment once the job is already gone. Prevention cannot. Prevention has to be poured like a foundation, before the building goes up, before the displacement happens. And if the configuration that would protect a category of work takes years to build while the displacement of that work takes months, then the window I keep invoking is not open. It closed before I finished describing it. For the people most exposed, I am offering a foundation for a house that has already burned down.I want to be honest: this objection genuinely worries me. It is not a strawman. It is the thing I have to answer before I am entitled to any of the hope in the rest of this piece.What Prevention Is NotBefore I try, I have to clear away the version of this argument that deserves every bit of scorn it gets. There is a reading of “prevent displacement” that means: force companies to keep workers doing jobs a machine could do more cheaply, freeze the org chart, hold the economy in amber against the tide of productivity. That reading is economically incoherent and I am not making it. You cannot order a firm to employ people to do nothing, and you should not want to. A serious economist will tell you that technological change displaces labor, that it always has, and that the humane question has always been whether we manage the transition well or badly. On that, the economist is right.But that is not the choice I am pointing at, because it is not the only choice on the table. The displacement debate keeps collapsing two different things into one. One is whether the work gets more productive, which it will, and should, and no one sane is trying to stop. The other is what shape the productivity gain takes: whether the technology is deployed to augment the worker and strip the drudgery from her day, or deployed to remove the worker and keep the wage she used to earn. Both are productive. Both capture the gain. They are not the same deployment, and the difference between them is not dictated by the technology. It is chosen. Prevention, as I mean it, is not the refusal of productivity. It is the contest over which of two equally productive configurations gets built: the one that keeps the human in the loop, or the one that empties the loop out. Anyone who tells you that contest isn’t real, that only one configuration was ever economically available, is smuggling the conclusion into the premise.Cure Isn’t Built EitherSo let me start by noticing what the objection quietly assumes. It assumes that cure is ready and prevention is not, that on one side of the ledger sits a functioning safety net, waiting, and on the other sits my hypothetical foundation that takes too long to pour. Time the two against each other and prevention loses.But cure is not built either.Consider what cure actually requires to work, not to be announced, but to work. Retraining has to take a forty-five-year-old customer service representative and return her to comparable income and comparable dignity in something other than a worse job. The evidence that retraining programs do this is, to put it gently, poor; decades of trade-adjustment and workforce-retraining efforts have a track record that ranges from modest to dismal. The infrastructure that would do it well does not exist at scale. It would have to be built.Or consider the transfer the encyclical leans toward, the social protection, the redistribution, in its strongest form a universal basic income. None of that is built. It is not funded. It is not politically coalitioned. Standing up an income-transfer regime large enough to absorb mass displacement is at least as slow, at least as institutionally demanding, and at least as politically captured as anything I am proposing on the prevention side. The Pope’s own remedies are a fifty-year project that no one has broken ground on.So when the defeatist holds a stopwatch to prevention, fairness requires holding the same stopwatch to cure. And when you do, the comparison stops favoring cure. I am not going to overclaim here, I am not going to tell you prevention is obviously faster, because I do not know that. The real claim is narrower and it is enough: the speed objection, applied evenhandedly to both sides, is not a reason to prefer the safety net. Both the net and the foundation have to be built, both are slow, both are contested. The only question left is which one is worth starting, and “it’s too late for prevention” cannot be the answer when cure is exactly as unbuilt.Radiology, HonestlyLet me give the critics their due on radiology, because they are right about it, and conceding that is the only way to extract the lesson that survives.Radiology is the slow case. It is the high-liability, high-status, heavily regulated, professionally fortified case. It is, in almost every respect, the least representative of the work AI is displacing fastest. If I lean on it as proof that prevention is easy, I deserve the truck that gets driven through the argument. A copywriter has no equivalent of the FDA. A transcriptionist has no standard of care. Pointing at radiology and saying “see, it can be done” is like pointing at a cathedral to prove that anyone can put a roof over their head.So I will not claim radiology’s timeline transfers. It doesn’t. You cannot grow an entire profession’s regulatory edifice in the time you have.But that is the wrong lesson to draw from it, and the critics stop one step too early. Radiology is not valuable as a timeline. It is valuable as an anatomy. It shows you what scaffolding actually is, disassembled into parts: a liability rule that names a specific human as accountable when the automated decision is wrong. A standard that defines what competent practice requires. A gate that governs what the software is allowed to decide on its own. A payment structure that funds the human-in-the-loop rather than penalizing it. Those are the load-bearing elements. And here is the thing the fifty-year objection obscures: most of that half-century was spent building the profession, not the configuration. The liability principle, the accountable-human rule, the single most important piece, is not a fifty-year artifact. It is a legal default that can be set by a ruling or a clause. We mistake the time it took to grow radiology-the-profession for the time it takes to attach radiology’s key lever, and they are not the same number.What transfers from radiology is not the cathedral. It is the knowledge of what a load-bearing wall looks like, so you can build a smaller structure faster, on purpose, now that you know what you are building.A Map, Not A SwitchThis forces a candor I think the whole debate has been avoiding, including, sometimes, me. Prevention is not a single switch you throw to save all work. It is a territory, and the territory has at least three zones, and they are not equally reachable.There is work that already has scaffolding. Regulated professions, safety-critical systems, anything where a wrong automated decision already carries legal liability, medicine, aviation, structural engineering, certain financial decisions. Here the configuration question is not a future project. It is live right now. The lever exists; the only question is whether we pull it in the direction of keeping the human accountable or let it slacken.There is work where scaffolding does not exist but could be built quickly, because a natural lever is within reach. Anywhere a single liability default could attach, who is responsible when the automated underwriting denies the loan wrongly, when the automated screen rejects the qualified applicant, when the generated legal document contains the error that costs the client. These do not require inventing a profession. They require attaching accountability to a decision that currently has none. That is a clause, a ruling, a procurement standard. Months, not decades.And then there is work where no scaffolding exists and none can grow in time. Some of the fastest-displacing work is here. For a category of jobs, the configuration that would have protected them needed to exist before the displacement began, and it did not, and it will not materialize fast enough. For that work, prevention has already lost. Cure is what is left, and the dignity of the people in those jobs depends on cure being better than the dismal thing it currently is.I am not going to pretend that third zone away. Naming it is the price of being believed about the first two. If I told you prevention covers everything, you would be right not to trust me, because you can see the call-center floor emptying out and you know no liability shield is coming to save it in time.But here is the reframe that turns this map from a verdict into a task. The call-center worker sits in the third zone not because the third zone is a law of nature, but because no one ever built her a place in the first. Her work has no accountable-human rule, no standard, no gate, not because such things were tried and failed, but because no one thought to attach them to work that nobody was protecting. The absence of scaffolding is an unbuilt thing. It is not an unbuildable one. And the difference between a door that is locked and a door that no one has yet tried is the entire difference between defeat and delay.I have to be more honest than that, though, because “no one thought to” is too innocent. The scaffolding around radiology was not merely thought of; it was fought for, against interests that would have preferred cheaper, faster, less accountable care. And the scaffolding around the call center is missing not only because no one got to it but because its absence is worth money to someone. An accountable-human rule is a cost. A deployment gate is a delay. A standard that says the generated work must be checked by a person who carries the consequence is a line item that the configuration without it does not have to pay. So the door is not just untried. In a good number of cases someone is leaning against it from the other side, and they have more resources to lean with than the worker has to push. This is the part the time objection never mentions, and it is, if anything, harder than time: the gravity pulling everyone downstream is not only the gravity of habit or imagination. It is the gravity of power. The people who make the deployment choice have every incentive to keep the upstream lever from being built, and the downstream remedy (let the public retrain the worker, let the public transfer her some income) is the outcome that costs them least. “Compassionate pragmatism” is, conveniently, also the cheapest thing they could be asked to accept.I do not say this to collapse into the conclusion that nothing can be done. I say it because any honest reckoning with why the scaffolding is unbuilt has to include the fact that its remaining unbuilt is, for some, a victory rather than an oversight. You cannot plan the construction without a true map of who is standing where.Speed Is Partly A ChoiceI have one more move, and I am going to be straight that it is the one I am least sure of.Why does displacement move in months while governance moves in years? We treat that gap as a fact of nature, technology is fast, institutions are slow, and the race is lost before it starts. But I do not think the gap is natural. I think it is built.Deployment is fast because it is frictionless by design. No liability attaches to shipping the automated system. No gate stands between the decision to deploy and the deployment. The marginal cost of pushing the replacement live is close to zero, and nothing in the environment slows it down. Governance is slow because it is friction-full by design, every check is a deliberate brake, and we built the brakes on purpose. So the timescale mismatch that supposedly dooms prevention is not a law. It is a configuration, the same kind of configuration as everything else in this argument. And radiology is, again, the existence proof: the FDA gate is friction deliberately attached to deployment, slowing the rollout of an automated diagnostic to the speed of governance at exactly the point where the stakes justify the brake.So in principle, you do not win by making governance faster than deployment, which you cannot do. You win by adding friction to deployment at the specific high-stakes points where speed is the enemy, bringing the two timescales toward each other from the other side.Now the pressure test, because this move does not get to walk away clean. Attaching that friction is itself slow and contested. A clumsy gate is worse than none: it ossifies, it protects incumbents, it becomes the regulatory capture that the encyclical’s sharpest critics rightly fear, where the rule meant to keep humans in the loop curdles into a moat around whoever wrote it. And friction applied to the wrong things just makes everyone poorer while protecting nothing worth protecting. So #4 is real but narrow. It does not say “slow everything down.” It says speed is not fixed, and at a small number of high-stakes points, a well-built brake can change the timescale that the whole objection rests on. That is a smaller claim than I would like to be making. It is the true size of it.Where This Leaves UsSo: can the gravity be resisted?The truthful answer is that the question was wrong, and the timing objection, the strongest thing anyone has against this whole project, proved less than it claimed while proving something real.It proved something real: for a portion of the work being displaced fastest, prevention has already lost. The foundation needed to be poured before the fire, and it was not, and naming that as a genuine loss is the only fair thing to do. The people in that zone are owed a cure far better than the one currently on offer, and pretending prevention will reach them is a comfort that costs them their due.But the objection claimed far more than it proved. It gave cure a free pass cure never earned: the safety net is exactly as unbuilt, exactly as slow, exactly as captured as the foundation it was supposed to outrun. It mistook the fifty years it took to grow radiology-the-profession for the time it takes to attach radiology’s one essential lever, which is a clause and a ruling, not a cathedral. And it treated the speed of displacement as a law of nature when speed is partly something we chose, and could partly choose otherwise.What that leaves is not the defeatist’s conclusion, too late, abandon prevention, build the net, and not the triumphalist’s either, configure everything, save every job. It leaves something narrower and harder to act on. Prevention is a live option in more of the territory than the defeatist admits and a dead one in less of it than the optimist wishes. And there is no way to find the real boundary between the two from the armchair. The only way to learn where the foundation can still be poured in time is to start pouring it, in the zones where the levers already exist, and watch where it holds and where the fire outruns it.The window, then, was the wrong image, or at least too simple a one. A window is open or it is closed. What we actually face is a window that is closing, at different speeds in different places. In some places it has already shut, and we should grieve that plainly and turn to the people on the wrong side of the glass. In others it is still wide, and the thing keeping us from climbing through is partly a belief, borrowed from a stopwatch that was never applied fairly, that the effort is pointless, and partly the plain fact that someone with more leverage than us would rather we didn’t.I want to end on the hardest version of the doubt, the one I cannot dissolve. It is possible that the deepest problem here is not time and not even power as I have described it, but authority, that the capacity to reach into a deployment decision and say “not this way” is not a power we have temporarily misplaced but one we never really held, and that the reason everyone ends up arguing about the size of the net is that the net is the only place our agency actually reaches. If that is true, then this whole essay is a description of a door with nothing behind it. I do not think it is true. But I cannot prove it isn’t, and I am not going to pretend the uncertainty away to give you a cleaner ending.What I can do is smaller, and it is the honest limit of what one person at a desk can do. I can say where I think the door is. I can argue that the lever exists, that the timing objection was applied unfairly, that the difference between augmentation and replacement is a choice and not a fact, and that the absence of scaffolding around the most vulnerable work is a thing someone built and someone could therefore unbuild. I cannot model the transition, draft the liability default, or organize the contract. I am one writer who has spent a long time looking at this and has come to believe the upstream question is real, answerable in part, and almost entirely unasked.So I am planting a stake. Here, I think, is where the ground can still hold a foundation, and here is where it cannot, and here is who is leaning on the door. I cannot build the thing. But I can refuse to let “it’s inevitable” and “it’s too late” stand as the last words, because they are not true, or at least not yet true everywhere, and the people who could actually do the building, the ones with the math, the law, the leverage I don’t have, will only reach for it if someone first insists the reaching is worth it.That is less than I wanted to be able to tell you. It is also a great deal more than nothing. It is the truth as far as I, just one person, can find it, and the next part belongs to hands that aren’t mine.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

This is the first of two essays on Pope Leo XIV's encyclical Magnifica Humanitas. Part 2 lands this Friday rather than next Tuesday; the argument doesn't survive a week's gap.When the most authoritative moral document ever written about artificial intelligence arrives at the same structural conclusions you have been arguing toward by an entirely different road, the temptation is to declare victory and stop reading. I want to resist that because the more interesting thing about Pope Leo XIV’s first encyclical is not where it agrees. It is where it stops.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.Magnifica Humanitas runs to roughly forty-two thousand words and reasons from a hundred and thirty-five years of Catholic social doctrine, beginning with Leo XIII’s Rerum Novarum and its defense of the worker in the Industrial Revolution. It is not a technology document dressed in vestments. It is a moral anthropology that happens to take AI as its occasion. And reasoning from that tradition, from premises that have nothing to do with the vocabulary of systems design, it lands on three claims that anyone who has thought structurally about AI deployment will recognize at once.The first is that AI does not bear consequences. The encyclical is blunt about it: these systems do not undergo experience, do not mature through relationships, and do not, in the Pope’s framing, judge good and evil or carry responsibility for what follows from their outputs. They imitate; they do not answer for the result. The second is that systems are being built the wrong way around, designed so that workers must adapt to the speed and demands of the machine, rather than the machine being designed to support the people who work. The third is what the encyclical calls the “architecture of visibility”: platforms engineered to capture attention, amplify what is visible, and shape opinion, treating the finite human capacity for attention as a resource to be mined.Consequence-bearing. The design of work. The capture of attention. An independent line of moral reasoning, starting from the dignity of the human person rather than from any analysis of deployment, arrived at an architectural diagnosis. That convergence is worth naming plainly, not because it flatters anyone, and not because everyone serious sees it this way. Plenty of capable people read AI primarily as a productivity expansion and think the displacement worry is overstated. The convergence I mean is narrower and more striking for it: two arguments built from unrelated foundations, one theological and one structural, traced the problem to the same place. That is some evidence that the diagnosis is structural rather than idiosyncratic.The Word That MattersThe encyclical’s sharpest move is a single word.Leo writes that technology is never neutral, because it takes on the characteristics of those who devise, finance, regulate, and use it. And he writes that the pursuit of greater profit cannot justify choices that systematically sacrifice jobs.Choices. Not consequences. Not the weather. Not an impersonal force descending on the labor market like a season. The encyclical locates the problem at the point of deployment design, in the decisions made by identifiable people about how the technology will be configured and to what end. This is the thing most commentary on AI never reaches. The dominant register treats displacement as a fact of nature: capability accelerates, jobs evaporate, and the only remaining question is what to do for the people left behind. Leo refuses that. He sees that the displacement is authored. It could be authored otherwise.That is the high-water mark of the document. Having reached it, watch where the remedies go.The TurnThe encyclical’s concrete recommendation is this: regulate the algorithms. Retrain the displaced. Use taxation, social protection, and industrial policy to ensure equity. Renew labor organizations. Entrust international oversight to a reformed United Nations.Every one of these operates after the displacement has occurred.Retraining presumes the job is already gone; it is a response to the worker who has been turned out, not an intervention in the decision to turn him out. Social protection presumes that the income has already been lost. Taxation and transfer presume that the displacement has occurred and that the proceeds are now being redistributed. International oversight watches outcomes. The diagnosis pointed upstream, to the choice. The prescription walks downstream, to the damage, and builds an apparatus to manage it. The document names the cause and then treats the symptom.An Ounce of PreventionThere is an old proverb for exactly this. An ounce of prevention is worth a pound of cure.Prevention acts on the cause before the harm. Cure acts on the damage afterward. Everything in Leo’s recommendation list is a cure: retraining cures the loss of a job, a transfer cures the loss of income, and oversight cures the worst outcomes once they have appeared. The prevention, the ounce, would act at the configuration itself, at the moment the deployment is designed, before anyone is displaced at all. It would treat the choice the encyclical so clearly named as a choice still open to be made differently, rather than a settled fact to be cushioned.The document does not go there. It finds the cause, names it precisely, uses word choice, and then turns to managing the consequences. This is not a small omission. It is the difference between asking whether the worker had to be displaced and asking what we owe him once he has been. The first question is the one the diagnosis demanded. The second is the one the encyclical answered.The Same Error, Left and RightYou might read that turn as a peculiarly Catholic failure of nerve, or a personal limitation of this particular Pope. It is neither. Watch what happens when the document meets its critics.The encyclical’s sharpest critics on the right found the diagnosis welcome and the statism alarming. They attacked Leo for misplaced faith in government, arguing that regulation concentrates power and that the market, left alone, has always eased the worker’s burden over time. Strike the remedies, they said, and trust the diffusion of technology to lift living standards as it always has.This is the mirror image of the same error. The Pope wants to compensate for displacement through the state; his critics want to absorb it through the market. Neither asks whether the displacement had to happen. They are having a furious argument about the size of the net while standing under the same assumption: that the fall is inevitable and the only question is what catches it. The left wants a larger net; the right wants a smaller one and faith that growth will fill the gap. Both treat the deployment configuration as a given.That is how durable the assumption is. It survives translation into Catholic social doctrine and into free-market editorializing entirely intact. It does not belong to a political camp. It is the path of least resistance for anyone who would rather not get inside the deployment decision, which is nearly everyone, because getting inside it is the ounce, and arguing about the net is the pound.The GravityIt is worth being honest about what this means, because it runs counter to the easy version of this essay: the one where a brave writer catches the Pope in a failure of courage.The encyclical did not fail for lack of insight. It reached the architectural diagnosis that almost no one reaches. It named the cause with a precision most secular commentary never manages. And then it turned downstream anyway, not because Leo could not see the upstream question, but because the downstream answer is the one that everything pulls you toward. The state reaches for the transfer. The market reaches for the long-run faith in growth. The moral authority reaches for the language of compassion and repair. Each of them, reasoning carefully from its own commitments, ends up cushioning the fall rather than questioning it.That convergence is the real finding. When the most authoritative moral voice in the world on this subject, a free-market editorial board, and a redistributionist politician all independently arrive at the same place, manage the consequences, and do not contest the choice, you are no longer looking at anyone’s particular blind spot. You are looking at a gravity. A direction that thought falls into once it stops actively resisting. The default is not an opinion you can argue someone out of. It is the slope of the ground.This is why the Pope’s turn matters more than any individual misstep would. It is the strongest possible demonstration of how steep the slope is. If a tradition built over a century and a third specifically to defend the dignity of the worker, a tradition with no profit motive, no electoral cycle, no shareholders, still slides downstream at the decisive moment, then the pull is not coming from greed, politics, or any of the usual suspects. It is structural. It is in the shape of the problem itself.The WindowSo I will not pretend the alternative is easy, or that naming the gravity dissolves it. It does not. The ounce remains harder than the pound; that is precisely what gravity means.But there is a difference between a slope you are sliding down without noticing and a slope you have seen. The encyclical’s great service, despite its turn, is that it climbed high enough to make the upstream question visible. It found the word choice and set it down where everyone could see it. That the document then walked past its own discovery does not unmake the discovery. The cause has been named, by an unimpeachable source, in front of 1.4 billion people. That service is not cancelled by the turn, and the turn is not cancelled by the service. Both are true at once, and the both-ness is the point: it is what gravity looks like when it is acting on the best diagnosis we have. That is not nothing. It may be the most that naming can do.Because naming is necessary, and it is plainly not sufficient. The prior question, not how to compensate the displaced, but whether they had to be displaced at all, is now visible, and still unasked at the scale that would matter. Whether it can be asked at that scale; whether the alternative diagnosis implies can actually be built, at what cost, and quickly enough to matter against a technology that moves in months: that is the harder question, and the one I will take up next. The gravity is real. The interesting argument is about whether it can be climbed.For now, it is enough to notice where we are standing and which way the ground tilts. The window has not closed. We have at least been shown the door. That is more than we are usually given.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

The Boot and the FeedThe dystopia we have been preparing for has a boot on its face. That is Orwell. The state watches. The state coerces. The state lies and tortures. Resistance is meaningful because the regime is meaningful. The hero in 1984 fails, but he fails in the act of trying.The dystopia arriving has no boot. It has a feed and a transfer. The state is mostly absent. Coercion is unnecessary. People are not being forced into the new arrangement. They are being given things until the arrangement is no longer a question. The pressure that built into 20th century totalitarianism does not build because it is absorbed in advance.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.This is a different issue and needs a different approach. The boot you can resist. The feed you scroll. The transfer you cash. The disquiet you cannot quite name. The slow disappearance of something you used to know about yourself, about the place you live, about what you and your neighbors are for.The pattern has been named before, more than once. The first naming was older than the printing press. The second was a novel published seventeen years before 1984. The third is the one we are inside.The First Naming: Bread and CircusesThe phrase comes from Juvenal, writing around 100 CE. Satire X. The Roman populace, he observed, had once concerned itself with politics. It had elected magistrates. It had debated policy. It had participated, in some form, in the project of the Republic. By Juvenal’s time, that engagement had collapsed. The populace cared about two things. Panem et circenses. Bread and circuses.The phrase is sharp because the diagnosis is structural. Juvenal is not saying the populace was bribed. He is saying that the state had taken on the provision of two things: daily subsistence through the grain dole, and spectacle through public games. The populace had accepted the trade. The trade was not articulated as a trade. It did not need to be. The agency migrated. The provision arrived. Political participation withered without anyone deciding to give it up.This is the mechanism worth understanding. Rome was not undone primarily by barbarians from outside. The empire that the barbarians eventually breached had already hollowed itself out from within. A citizenry that had traded participation for provision no longer had the muscles to defend the participation when it was needed. The political body had atrophied while the urban body had been kept fed and entertained.The actors in this story are not villains. The emperors who maintained the grain dole were responding to genuine urban hardship. The politicians who funded the games were doing what successful politicians have always done. The populace was not foolish to prefer bread and games to faction and risk. Each step made sense. The cumulative drift did not require anyone to be malicious. It required only that nobody resist the convenience of the arrangement.This is the first naming. The pattern is: a population that holds political power, a state that can afford to provide subsistence and entertainment, and a slow trade that nobody calls a trade. The result is a populace that retains the form of citizenship but loses its substance. Juvenal noticed it because he could see the gap between what the citizens had been and what they had become.The Romans gave us the original vocabulary. The 20th century gave us the second.The Second Naming: Soma and FeeliesThe phrase comes from Aldous Huxley. Brave New World. Huxley was not predicting the boot. He was predicting something stranger. He was predicting a regime that needed no boot because the population had been engineered to prefer it.The mechanisms in his world are now familiar. There is a drug called soma, distributed freely, which dissolves anxiety. There are immersive entertainments called feelies, which replace depth with stimulation. There is conditioning from before consciousness, which produces people who want exactly what their station permits them to want.What Huxley saw is that none of this requires force. The citizens of his world are not oppressed in any sense they would recognize. They are happy. They do not miss what they have lost because they have been engineered not to know it ever existed. Their happiness is the control mechanism. The pleasure absorbs the pressure that would otherwise produce dissent.The contrast Postman drew is worth keeping. Orwell’s anxiety was about losing access to the truth. Huxley’s anxiety was about losing the capacity to care about truth once enough pleasure was on offer. The first is a problem of suppression. The second is a problem of dissolution. The pleasure makes the truth feel unnecessary, and after a while, it feels beside the point.This is the second naming. The pattern is the same as the Roman one. A population that holds something it does not realize it holds. A system that can afford to provide subsistence and pleasure at scale. A slow trade that nobody calls a trade. The result is a population that retains the form of personhood while losing the substance. Huxley noticed it because he could imagine the gap between what people had been and what they could become if the conveniences were comprehensive enough.The Romans had bread and games. Huxley had soma and feelies. The pattern was the same in both centuries. The question is: what are we giving up in this one?The Third IterationThe pattern is now arriving. The vocabulary is different. The architecture is the same.The transfer is UBI, proposed, partially piloted, and discussed as if its arrival is a matter of when, not whether. The feed is algorithmic, infinite, calibrated to whatever holds attention longest. The Roman state could afford grain and games for the urban population because the empire was wealthy. The modern state, along with the technology sector, can afford UBI and infinite content because the AI economy is generating wealth on a scale that has never existed before. The mechanism is the same as it was: subsistence plus spectacle, provided at scale, by a system that can afford the provision.The contemporary version differs in one structural way. Roman games and Huxley’s feelies gathered citizens in shared experience. The algorithmic feed disperses them into individual streams. Collective recognition is harder than it was in either of the predecessors.What the Provision AbsorbsWhat is being absorbed by the provision deserves naming.There is a personal cost. For most people, work has never been only a source of income. Work provides what transfer payments cannot compensate for. Identity. Purpose. Social role. Daily rhythm. The structure that organizes a life. The unspoken understanding that the morning has a shape because there is something to do that matters to someone else. The slow accumulation of competence at something specific. The relationships that develop through shared work over time. The check covers the rent. It does not cover any of the rest of it. The check arrives in the mailbox of someone who used to be a designer, a paralegal, or a journeyman electrician and is no longer, and the disappearance of what they were is not on the ledger.There is a civic cost. This is the most precise name for the Roman pattern. A citizenry that has traded participation for provision no longer has the muscles to defend the participation when it is needed. The civic body atrophies. The capacity for collective political action withers. The institutions that depended on engaged citizens become hollow. The voters still vote. The participant no longer participates in anything. The pressure that should have built against the trajectory does not build. The actors with the most influence over what comes next operate without the resistance that would otherwise check them. Wang Peng’s “maintaining order, not sharing wealth” applies at the civic register as well as the personal one. The order being maintained is not only the order of keeping the urban poor calm, but also the order of a political system whose citizens have ceased to function as citizens.The pattern is the same. The medium is different. The Romans used grain and games. Huxley imagined soma and feelies. We are building UBI and recommendation feeds. The mechanism in each case is provision at scale, calibrated to absorb whatever pressure the citizenry might otherwise generate. The Romans saw it. Huxley imagined it. We are living inside the third iteration, and we have not yet named it because we have been preparing for the wrong dystopia.Why the Pattern RecursThe pattern recurs because it solves a real problem. Coercion is expensive and produces resistance. Pacification is cheaper and produces compliance. The Roman emperor who funded the grain dole had fewer riots than the emperor who let the urban poor go hungry. The Brave New World administrators needed no Stasi because soma was cheaper than the secret police. The AI economy needs no propaganda ministry because algorithmic feeds and a competent transfer payment can produce the same calm at a lower cost.What is new is the capacity. Rome’s dole reached an urban population. Huxley’s regime was fictional. The AI economy is the first arrangement that could afford pacification at a planetary scale while generating the surplus that pays for it. The pattern has been waiting for it.This is what makes the pattern dangerous in a way that the Orwellian dystopia is not. The Orwellian regime advertises itself. The boot is felt. The dissent is real because the oppression is real. The pacification regime does not advertise itself. The disappearance is not felt. What has been lost is the very capacity to notice the loss. The grain arrives, the feed scrolls. The check is deposited. The disquiet you cannot quite name turns out to be the only signal you were ever going to get.Two millennia. Three iterations. One pattern.The AlternativeThere is an alternative. It is not nostalgia for work that no longer exists. It is the deliberate design of work that does not yet exist, organized so that humans and their machines preserve what the pacification dissolves.The third iteration is the one that gets named, or it is the one that does not.Next: Architecture or SomaThanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

The Balm on TopIn a recent essay translated by Zilan Qian, Wang Peng, a senior expert at the Tencent Research Institute, names the AI race as a narrative choice rather than a scientific path, and identifies UBI as “maintaining order, not sharing wealth.” That last phrase deserves more attention than it has received. UBI is the most discussed proposed response to AI-driven displacement. It is also the response that requires the least change to anything else. The labor market collapses on its current trajectory. The wealth concentrates on its current trajectory. Cognitive offloading deepens along its current trajectory. UBI sits on top of all of it like a balm.This essay argues that the surrender is convenient, not inevitable. Mass displacement is a design choice being made by default. UBI is the consolation prize offered for not asking the prior question: who decided AI gets to make this world, and why are we letting it?Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.The Dominant StoryThe story most often told goes like this. AI capability is accelerating. AI capability will continue to accelerate. Many jobs will disappear. Many more will be transformed beyond recognition. People displaced from the labor market need an income. UBI provides that income. The wealth created by AI funds it. Everyone settles into a new equilibrium. Productivity is up. Suffering is mitigated. The future is uncomfortable but workable.This story is appealing for reasons worth naming.It is mechanically simple. Money is the most universal currency for solving problems. If the problem is lost income, an income transfer is the most legitimate response. No ambiguity about what is being given, what is being received, or how to measure whether it worked.It absorbs political anxiety. Mass displacement is genuinely frightening. UBI offers a coherent answer that does not require anyone to confront harder questions. It promises to catch people on the way down without anyone having to ask why they were falling in the first place.It externalizes responsibility. The labs build the systems. Governments handle the consequences. Workers receive the consolation. Nobody in the loop has to question the trajectory. This is the structural function of UBI as currently discussed: it lets each actor optimize their own piece while treating the displacement itself as a fact of nature.It is also the story that the storytellers benefit from. Wang Peng makes this argument directly. The narrative that scaling LLMs is the inevitable path to AGI was not validated by science. It was locked in by capital and geopolitical maneuvering. UBI is the same narrative one layer up. The displacement is treated as inevitable, not because the evidence is settled, but because treating it as inevitable benefits the actors with the most influence over what comes next. Investors keep their thesis intact. Executives keep their growth narrative. Policymakers reach for an instrument they understand.The discomfort in this story belongs entirely to the people losing their jobs. They get a check.The Prior QuestionThat is the dominant story. It is not wrong about everything. Displacement is happening. Some response is needed. But the story is missing the prior question. Before we ask how to mitigate displacement, we should ask whether it is an inevitable result of the technology or a result of how we have chosen to deploy it.Nothing about AI requires that it displace workers wholesale rather than augment them. Both deployments are technically possible. Both are happening in different proportions in different places. The proportion is a choice. It is being made not by laws of physics but by laws of capital, optimization targets, and managerial habit.Consider where the choices are being made. A company with a customer service team has options when it brings AI into the workflow. Option one: replace four of five representatives with a single AI-augmented operator. Option two: give all five representatives AI augmentation and expand the service quality and volume. Both work. The first compresses headcount. The second compresses friction. Different choices, same technology.The choice is not hypothetical. Radiology is the most often cited case. A decade ago, the AI researcher Geoffrey Hinton predicted that AI would displace radiologists within five years. The deployment moved the other way. Radiologists became radiologists with AI. The technology took over pattern detection at scale. The interpretation of the pattern, the conversation with the referring physician, and the consideration of the patient’s history remained with the radiologist. Radiology employment has grown. The technology was real. The displacement was not inevitable.The labs themselves face the same question one level up. They optimize for benchmarks that measure “can the AI do this task instead of a human?” rather than “does the AI make humans better at this task?” That choice of optimization target is not a scientific discovery. It is a decision about what the technology is for. Once made, it propagates outward into every product, every demo, every funding round.Configuration, Not LawThe strongest objection to this account is that it understates competitive pressure. Firms that fail to automate are priced out. Nations that fail to deploy aggressively are outcompeted. The choice to augment rather than displace, even when technically available, is foreclosed by competition itself.The objection is real, and it concedes the argument. Competitive pressure is not a law of nature. It is the cumulative result of optimization targets, capital allocation, procurement standards, and the absence of countervailing pressure. Each of these is human-constructed. Each can be reconfigured. Augmentation appears uncompetitive because the incentive landscape rewards displacement and treats augmentation’s slower, less legible gains as a cost rather than a return. That is a configuration, not a law of nature.Previous technological transitions moved workers from one kind of work to another. The mechanical loom displaced weavers. The spreadsheet displaced bookkeepers. In each case, new work appeared because the technology amplified some human capacities while replacing others. The current trajectory differs in target. The optimization is not the amplification of certain capacities and the substitution of others. It treats the human as the variable to minimize. That is not a transition. It is a redesign of what work is for. Calling the redesign inevitable is a way of not asking whether the optimization target was chosen and by whom.AutopilotThis is what is meant by autopilot. The choices are real. The choices have consequences. But nobody in the chain experiences themselves as making the choices. The lab optimizes for the next benchmark. The investor optimizes for the next round. The executive optimizes for the next quarter. The customer optimizes for the next price drop. Each actor is responding to the incentives produced by the others. None of them is looking at the cumulative trajectory and asking whether it is the one we would have chosen if we had been asked.What would conscious steering look like? It would begin by asking different questions. Not “how do we automate this task?” but “what does this task contribute to the people doing it, and what is lost if we automate it away?” Not “how do we maximize productivity per worker?” but “what is the relationship between this work and the meaning the worker draws from it?” Not “how do we redistribute the income lost to displacement?” but “what would deployment look like if displacement were not the default?”These are not abstract questions. They are answerable. They are also rarely asked because doing so would require the most influential actors to reconsider what they are building and why. UBI lets them avoid the questions entirely. The displacement happens, the income transfer happens, and nobody has to look upstream.That is the design choice being made by default.Order MaintenanceInside that arrangement, UBI has a specific function. Wang Peng puts it directly: ensuring “the replaced majority have enough to eat and won’t revolt, so that the few at the top can continue to quietly claim the surplus profits AI delivers.”This is not a moral critique. It is a structural one. Wang is not saying UBI’s advocates are bad people. He is saying that the function UBI performs inside the current trajectory is order maintenance, regardless of the intentions of those who advocate for it. Once the displacement is fixed and the trajectory is given, UBI’s job is to absorb the resulting pressure. That is what it does. That is what it is for.This is also not an argument against cash transfers as such. A basic income floor might be part of a just society. The objection is to UBI as the complete response, the mechanism that lets the displacement question go unasked. The check is not the enemy. The check is an excuse.Consider the system from each actor’s perspective with UBI in place. The lab continues building. The investor continues funding. The executive continues deploying for headcount compression. The displaced worker continues paying rent. The state continues to collect taxes and distribute transfers. The pressure that would otherwise build into pressure for the prior question, who chose this and why, gets absorbed by the transfer. The check arrives. The asking does not happen.This is what is meant by “convenient surrender.” It is not that UBI fails to deliver income. It will deliver income. It is that UBI delivers income in a configuration that removes the political and economic energy that would otherwise force the prior question. The transfer succeeds in what it is designed to do: keep the system running. The system, however, was the problem.The Mismeasured CostThere is a deeper cost. UBI assumes the displaced worker’s loss is monetary. The check is calibrated to cover the rent, the groceries, and the heat. This is the calculation a labor market reduction permits. The loss that cannot be calibrated is the one accumulated over twenty years. The paralegal who has spent two decades learning which precedents matter for which arguments has built something the labor market paid for, and the AI now performs in seconds. The check can replace the salary. It cannot replace competence. The competence was not only a means of income. It was the daily evidence to herself that her work mattered to someone else. The cost has been mismeasured. The check addresses only the part that had a price.A reader will object that the paralegal is a poor representative case. Much of the work is not like hers. The Amazon picker, the call center agent, the gig worker stitched across three platforms. These are not sources of identity and accumulated competence. They are sources of income and a transfer payment that replaces lost income, with no value added.The objection is partly right and concedes more than it intends. It is right that much work has been structured to extract rather than to sustain. It concedes the larger argument because that structuring was a deployment choice. The warehouse picker could have been doing cognitively engaging work if the warehouse had been designed for human capabilities rather than treating humans as replaceable. The call center agent could have been doing relational work if the script had been a tool rather than a leash. The meaninglessness of that work was not a fact of the work. It was a design. The argument is not that all current work is worth preserving. It is that the configuration which made meaningful work the exception is itself a choice, made by the same logic that is now choosing displacement.There is a long history, both real and imagined, of societies that solved the wrong problem by giving people enough to live on while removing what they were living for. The Romans had a name for it. Huxley imagined another.The Harder OptionThe convenient surrender is not the only option. It is the option that requires the least of anyone with the power to choose differently. That is precisely why it is being chosen.The harder option is to ask the prior question. Whose work is being displaced, and why? Who decided that technology gets to make this world? What would it mean to deploy AI in a way that preserved rather than dissolved the relationship between work and meaning?These questions are answerable. They have not been asked because the dominant story has made them seem unnecessary. UBI is offered as the response to displacement so that displacement does not have to be debated.We have spent decades preparing for the wrong dystopia. The one arriving was named first. It is older and quieter, and we have stopped reading the people who named it.Next: The Wrong DystopiaThanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

A note I posted recently about due process sparked a broader thought. I’ll come back to it at the end.The thing people fear most about AI has been happening in politics for decades.Let me explain.Large Language Models, the technology behind ChatGPT, Claude, and others, are remarkable at a very specific thing. Within the boundaries of what they were trained on, they are extraordinarily fluent. Confident. Articulate. Persuasive, even. Ask them something well within their corpus, and they will give you an answer that sounds authoritative, well-reasoned, and complete.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.The problem is that they have no internally grounded model of uncertainty. They do not know what they do not know. Push them outside their training distribution, ask them something genuinely novel, something that requires real-world judgment or lived context, and they do not slow down, qualify, or say “I’m not sure.” They confabulate. They produce an answer that has exactly the same tone, confidence, and fluency as a correct one. The model cannot tell the difference. And neither, often, can you.This is not a minor technical quirk. It is a fundamental architectural limitation. And it is, increasingly, what critics of AI are most alarmed about: not that these systems are wrong, but that they are wrong with total confidence, and you cannot easily tell when.What makes this architectural limitation so hard to solve is that it was not introduced by the engineers. It was inherited from the source material. The text that pre-trains these models is the accumulated output of human communication, not just the last forty years, but millennia of rhetoric, law, scripture, political argument, and power exercised through language. Human communication is already saturated with what the evolutionary biologist Robert Trivers identified as self-deception: confident assertion, motivated reasoning, moral certainty, suppressed counter-evidence. We perform our convictions fluently and our doubts reluctantly. The feedback step that follows pre-training compounds this: human raters reward confident, fluent answers because that feels authoritative. The system correctly learns what we prefer. The flaw was not built in. It was passed down.Here is my question: when did this become new?The question isn’t why politicians behave like LLMs. The question is why our systems select so ruthlessly for exactly that behavior and discard everything else.The behavior itself is older than politics. Politics is just where we see it most clearly today. Underneath it is something more fundamental: the human knowledge claim. Any moment where one person asserts to others that they know something, where authority, status, or persuasion hangs on the assertion, selects for confident delivery and punishes hesitation. The shaman who qualifies the prophecy loses the tribe. The elder who admits uncertainty loses the counsel. The expert who hedges loses the room. This is not a modern failure of character. It is an ancient feature of how humans establish and defend claims to know.This didn’t start with algorithms. It didn’t start with mass media. Leaders have performed certainty for as long as there have been leaders, ancient tribal chiefs whose authority depended on never being seen to hesitate, Roman emperors issuing decrees from a position of assumed divinity, monarchs whose legitimacy required the performance of absolute knowledge, demagogues across every century who understood that confident assertion outperforms careful reasoning in any contest for attention.What’s new isn’t the behavior. What’s new is the system that now selects for it at an industrial scale, amplifies it in real time, and punishes the alternative more efficiently than anything in human history. Social media poured gasoline on a fire that has been burning since our earliest ancestors organized themselves around people who claimed to know. Platforms amplify the loudest, most certain voices and bury the nuanced ones. Politicians adapt. They perform with greater certainty, a certainty that is further amplified. Round and round. The system doesn’t just tolerate overconfidence. It breeds it.Education shows the same pattern. As Steven Mintz recently observed, the habits that careful reasoning actually requires, matching claims to evidence, engaging opposing views fairly, and revising under scrutiny, are not just neglected in most educational settings. They are actively discouraged by the way students are evaluated and rewarded. The system doesn't produce confident hallucination by accident. It trains for it.Watch a political campaign.Any campaign. Any party. Any country, though I will let you apply your own examples.The candidate arrives with a platform. The platform is confident, clear, and internally consistent. It has an answer for everything. It does not hedge. It does not say “this is genuinely complicated and I am not sure.” It does not acknowledge that the opposing view has any merit, or that the world might resist the plan once it meets reality. Nuance is a liability. Certainty is the product.Then they get elected. And reality arrives.The economy does not behave as the model predicted. The allies do not cooperate. The opposition does not collapse. The budget does not balance. The promised jobs do not materialize, or they do but somewhere else, or they do but not for the people who were promised them. The world, it turns out, was not waiting for the correct policy to be implemented. It had its own ideas.And here is where the parallel becomes most uncomfortable. A well-designed AI system, when confronted with a question outside its competence, should ideally signal uncertainty, flag that it is operating at the edge of its reliable knowledge. The best ones are getting better at this. But a politician, confronted with the gap between their platform and reality, almost never does. They double down. They reframe. They find someone to blame. They stay in the corpus.Because admitting uncertainty, after running on certainty, is politically fatal.This is not really about confidence.Confidence itself is not the problem. Experts are confident. Surgeons are confident. A good structural engineer does not hedge when they tell you the bridge will hold. Confidence, earned through genuine competence within a well-defined domain, is exactly what you want.The problem is the absence of awareness of uncertainty: knowing where your competence ends. This is one of the most underrated cognitive capacities a human being can possess. It is what separates a good doctor from a dangerous one. It is what separates a good leader from a demagogue. And it is, not coincidentally, what separates genuine intelligence from its simulation.An LLM that cannot model its own uncertainty is not wise. It is fluent. These are not the same thing.A politician who cannot model their own uncertainty is not strong. They are performing well. These are also not the same thing.The voters who reward performance certainty over genuine competence are not getting what they think they are getting. They are getting the political equivalent of a confident hallucination.A word about compromise.Compromise is what uncertainty-awareness looks like in practice. It is the act of acknowledging that your model of the world is incomplete, and updating it in response to someone else’s. That’s not a weakness. That’s how functional systems avoid drifting further from reality.I know. It is an ugly word right now. Mediocre. Weak. The language of people who do not believe in anything.I disagree. Strongly.Compromise is how the greatest human gains have actually been made. Not the heroic narrative version, not the lone visionary who refused to bend and changed the world. That story exists, but it is the exception, and we have collectively lost our ability to distinguish it from its imitation. The far more common story of human progress is negotiation, coalition, trade-off, and the slow, unglamorous accumulation of partial wins.Some will argue that conquest is more powerful. The great advances came from decisive force, not from committees. I would ask anyone alive for any part of the last hundred years to sit with that honestly. Two world wars. The Cold War. Vietnam. Iraq. The wreckage of absolutism, administered at scale. The places where things actually got better, where poverty fell, where disease retreated, where life expectancy climbed, were almost always the result of sustained, boring, incremental cooperation between people who disagreed about things but agreed to keep working.That is not a weakness. That is civilization.So what are we actually asking for?Not perfection. Not a political class of philosopher-kings who speak only in careful qualifications. Not AI systems that refuse to answer anything they are not certain about.Just this: a restored tolerance for nuance. A willingness, in our politics, and increasingly in our AI, to reward “I’m not sure, but here is my best reasoning” over “here is the answer, delivered with complete confidence.”But tolerance alone won’t be enough. Tolerance is a cultural mood. It comes and goes. What actually works is when systems structurally require engagement with contrary positions. Due process doesn’t rely on judges being humble. It builds humility into the architecture: evidence must be examined, dissent must be heard, certainty must be earned through scrutiny. The same logic applies to our politics and our AI. We don’t need leaders who personally appreciate complexity. We need systems that make engaging with it unavoidable.The irony is almost too neat. We are having an urgent public debate about the dangers of AI systems that confidently hallucinate. We should be having the same debate about the human systems that have been doing it for millennia, doing it with industrial efficiency for the last forty years, and now, through AI, at a scale and speed nothing in human history has prepared us for.We don't need to accept either. But fixing both starts in the same place: understanding that nobody is right about everything, that uncertainty is not weakness, and that the gap between confident fluency and actual wisdom is exactly where the damage gets done.Which brings me back to where I started. The rush to judge on allegations alone, before trial, before evidence, before verdict, is its own form of hallucination. A confident narrative is preferred over a slow, uncertain process. We have built an entire legal system on the principle that certainty must be earned, not imposed. That due process is not a weakness. It is the mechanism that keeps the system tethered to reality. It is, almost exactly, what we are asking for in our politics and in our AI. The question is why we seem to want it in our courtrooms but nowhere else.Next week: The Category Error at the Heart of Justice. Our legal system is procedurally fair but causally illiterate, and the difference matters more than we admit.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

In a recent piece, Tim Lee offers one of the cleaner arguments I’ve read for why today’s agent architectures are unlikely to produce “AI scientists” anytime soon. His central observation: the implicit knowledge knowledge workers carry, the hunches, the half-formed associations, the things on the tip of the tongue, doesn’t survive the handoffs that agentic systems require. He borrows Marc Andreessen’s framing that “your agent is just its files,” then turns it against the optimistic reading. If the agent is just its files, then whatever the language model can’t articulate gets left behind every time the context window resets. The temp-worker analogy that follows, a different person each Monday, however well-trained, however meticulous the predecessor’s notes, is the most legible version of this argument I’ve seen in popular tech writing. Lee has translated something the field has been gesturing at for two years into language a general reader can hold.But here’s where I want to push further. Lee describes hunches as compressed pattern recognition, knowledge the brain holds but can’t articulate. That’s true, and it’s part of the story. What it leaves out is why a seasoned practitioner’s hunch is trustworthy in a way that a fresh one isn’t. The seasoned hunch isn’t just a denser pattern; it has carried a consequence. The practitioner has made calls, watched them play out, paid for the wrong ones, and adjusted. That loop, judgment, action, cost, revision, is what gives the hunch its weight. An LLM mid-session might develop a hunch-like feeling. Nothing in its loop ever bears a cost. Nothing carries forward. This isn’t a context-window problem that a longer window or smarter compaction will solve. It’s a stake problem. The files can hold what the model said. They cannot hold what they would have lost by being wrong.Lee ends where he can: we’ll still need human workers to do our deep thinking for us. That’s the right conclusion if you stop at the inspection level, at what the agent produces. But it leaves the harder question untouched. If the consequence-carrying loop is what makes judgment trustworthy, and if that loop can’t survive the handoffs an agentic system requires, then institutions can’t govern AI use by reviewing outputs. Outputs are exactly the explicit residue Lee has just told us is insufficient. They are the files. Reviewing them well will not tell you whether the practice that produced them was a practice at all, or only its performance. Lee has named the gap. The next move is figuring out what kind of institutional architecture closes it, or learns to live within it.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

My wife Jen sent me a Boston Globe essay this morning by Michael Shermer, the Skeptic magazine publisher, titled “What is truth, anyway?” I read it twice. I agreed with most of it. And something nagged.Shermer’s toolkit is the right one. Fallibilism. Bayesian reasoning. Extraordinary claims require extraordinary evidence. Signal detection. Active open-mindedness. Free critique. If more public discourse ran on this equipment, we would all be better off.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.And yet, in the same essay where he carefully reminds us that “the principle of fallibilism requires me to admit that I could be wrong,” Shermer delivers confident verdicts on COVID origins, climate severity, and gender, the kind of verdicts that suggest the toolkit has, in fact, settled the matter.That gap is what I want to write about. It is not a gap in Shermer specifically. It is a gap in the framework he is using, and in the way most of us, myself included, talk about truth. The toolkit tells you how to weigh evidence. It does not tell you when you have reached the edge of the toolkit's applicability. Without that second layer of awareness, the toolkit does not produce truth. It produces confidence.Truth, I want to argue, is not a possession you arrive at by running the right cognitive procedures. It is a practice that requires an architecture, and that architecture is what the rest of this post is about.What Shermer Gets RightShermer’s essay is, at its core, a defense of the cognitive equipment the Enlightenment built for navigating uncertainty. It is worth taking seriously, piece by piece.Fallibilism: the recognition that any of our beliefs could be wrong, and that intellectual honesty requires holding them provisionally. This is not a weakness; it is the precondition for learning anything new.Bayesian reasoning: the discipline of attaching probabilities to claims rather than treating them as binary, and updating those probabilities as evidence shifts. Shermer’s example of assigning UFO aliens a 0.01 percent probability rather than zero is the right move. It leaves room for evidence to change his mind.ECREE: extraordinary claims require extraordinary evidence—Sagan’s principle, simple and durable. A blurry photograph is not evidence of Bigfoot. A grainy video is not evidence of alien visitation. The bar scales with the size of the claim.Signal detection theory: the 2x2 matrix of hits, misses, false alarms, and correct rejections. Shermer is right that most public arguments fail because people cite only the cell that confirms their view. RFK Jr.'s stance on vaccines is a clear example. So is most medical-cure folklore. So, frankly, is most political commentary.Active open-mindedness: Tetlock and Gardner’s finding that the best forecasters are the ones who actively seek evidence against their own positions, treat changing their minds as a strength rather than a weakness, and accept that randomness shapes outcomes. The worst forecasters are the ones with grand unified theories who explain away every miss.Free critique: Shermer closes by arguing that the most important norm is the freedom to challenge any and all ideas. He is right. Censorship is corrosive in both directions: if I silence you, why shouldn’t you silence me?This is a serious toolkit, assembled over centuries. I use it. You should, too. Most public discourse would improve dramatically if more people did.The question is not whether the toolkit works. The question is what it works on, and what it cannot, by itself, do.What’s Missing: The Constraint-Awareness LayerHere is the move I want to make.Shermer’s toolkit tells you how to weigh evidence inside a question. It does not tell you whether the question is one the toolkit can answer. That second layer, the awareness of where the toolkit applies and where it reaches its edge, is what I have been calling constraint-awareness. It is the working definition of wisdom in my Wisdom Gap whitepaper.Constraint-awareness is not the same as humility. Humility is a disposition. Constraint-awareness is a structural property of how you hold a belief: knowing the conditions under which your method works, the conditions under which it does not, and the conditions under which you cannot tell which you are in.Without it, the toolkit produces something that looks like truth but isn’t. It produces calibrated confidence inside a frame that the thinker has not examined. Bayesian reasoning, applied to a question whose evidence base is geopolitically corrupted, gives you a probability estimate that feels rigorous and is in fact noise dressed in numbers. The same diagnosis applies to any other tool in the kit. Each was designed to operate within the conditions the thinker is responsible for noticing, and none of them can notice for itself.An example from my own consulting career, to make this concrete. My team was selected to lead a post-acquisition system integration following a large data analytics company's acquisition of a smaller, complementary firm. Our standard approach was sound: spend the first six weeks in discovery, talk to stakeholders on both sides, build a proposal that accounts for the operational realities of each party. We did exactly that. We treated both companies as equally weighted clients. We came back with a proposal that assumed significant adaptation of the acquirer's systems to accommodate the acquired company's business. The budget was large. The timeline was long. We presented, and we had our hat handed to us. The executive sponsor on the acquirer's side summarized the problem in one sentence: "We are acquiring them; they need to adapt and fold into our way of doing business, not the other way around.” The toolkit was right. The conditions under which it was developed (a normal client with stakeholders whose interests are roughly symmetrical) were not the conditions in which we were operating then (a post-acquisition integration, where the acquirer's operating model is the destination, not a negotiable input). Constraint awareness would have let us see that before the proposal landed, rather than after.Three thinkers I keep returning to have each pointed at this layer from different directions. Judea Pearl, in his ladder of causation, distinguishes association (Rung 1) from intervention (Rung 2) from counterfactual reasoning (Rung 3), and his core warning is that statistical machinery applied at the wrong rung produces confident nonsense. Donald Hoffman, in Fitness Beats Truth, argues that our perceptual systems were not built to deliver reality; they were built to deliver fitness, and the two are not the same. Nassim Taleb, in his work on skin in the game, insists that beliefs held without consequences drift away from the truth in ways the holder cannot detect from the inside.Each of them is naming a different edge of the toolkit. Pearl: the edge where your inference machinery exceeds its causal license. Hoffman: The edge where your perceptual interface is not the territory. Taleb: the edge where your belief has no feedback loop to correct it.Constraint awareness is what lets you see those edges from within your own thinking. It is not a tool you add to the toolkit. It is the meta-property that determines whether the toolkit produces truth or confidence.And here is the harder claim, the one I will spend the rest of this post and the next whitepaper defending: constraint-awareness cannot be generated by individual cognitive effort alone. It requires an architecture: an attention-experience feedback loop, sustained over time, inside institutions and developmental pipelines that test beliefs against consequences and reveal their edges. Strip-mine that architecture, and you do not get a population of bad reasoners. You get a population of good reasoners producing confident verdicts at the edges of frames they cannot see.Which brings us to COVID.Exhibit A: COVID OriginsTake Shermer’s COVID example. He writes: “I believe that the COVID virus is slightly more likely to have originated in a lab than a wet market.”I am not going to tell you whether he is right or wrong. That is the point.What I want to ask is a different question: what would have to be true for the toolkit to deliver a probability estimate on this question that means what a probability estimate is supposed to mean?A Bayesian probability is only as good as the evidence base it draws on. The machinery was designed for evidence generated by a process you can characterize, from sources whose reliability you can estimate, with a sample space you can bound. COVID origins fail all three conditions, and the failures are not accidental.The evidence base is geopolitically corrupted. The Chinese government has restricted access to the Wuhan Institute of Virology, withheld early case data, and shaped which samples reached international researchers. Some of what is missing is missing on purpose. You cannot run Bayesian updating on an adversarially curated sample space.The evidence base is institutionally entangled. Western intelligence agencies, public health institutions, and research funders all faced reputational and legal exposure if the lab-leak hypothesis proved true. Some of the early dismissals were sincere; some were defensive; from the outside, you cannot reliably tell which. The reliability weights you would need to plug into Bayes’ rule are themselves contested.The evidence base is epistemically novel. This is not a question like “did this defendant commit this crime” or “is this drug effective,” questions for which we have centuries of method and calibration. It is a question about a singular event in a domain where the base rates themselves are unknown and possibly unknowable. In Frank Knight’s terms, this is uncertainty rather than risk: a domain where probability estimates were never the right tool to begin with.Each failure mirrors one of the edges from earlier. Corrupted evidence is Taleb’s edge: beliefs without honest feedback drift in ways the holder cannot detect. Entangled evidence is Hoffman’s edge: what reaches you is shaped by an interface (institutional, in this case) that was not built to deliver truth. Novel evidence is Pearl’s edge: inference machinery applied where its causal license does not extend.None of this means we should not consider the origins of COVID. It means the cognitive procedure Shermer is using cannot, by its own internal logic, produce a calibrated probability here. The output looks like a Bayesian estimate. It is, more honestly, a prior, the starting belief one brings to a question before evidence updates it, given weight by a procedure whose conditions for operating were not met.Shermer’s reasoning is not sloppy. It is rigorous reasoning applied to a question whose epistemic conditions the rigor itself cannot diagnose. You need a layer above the toolkit to notice that, and that layer is constraint-awareness.The same diagnosis, applied differently and to varying degrees, fits his confident verdicts on climate severity and on gender. They are not the same kind of question as the COVID origins question, and treating them as equivalent would itself be a constraint-awareness failure. The whitepaper will work through them separately.Why This Is Architectural, Not PersonalThe point of the COVID example is not that Michael Shermer is a bad thinker. He is a good one. He is using the best cognitive equipment the Enlightenment produced, and using it carefully.The point is that individual cognitive effort, however rigorous, cannot generate constraint-awareness on its own.Constraint-awareness is not a tool you reason your way to. It is a property that develops over time through exposure to consequences within an architecture that connects beliefs to feedback. You learn where your toolkit reaches its edge by watching it fail, by being wrong in ways you cannot dismiss, in domains where the cost of being wrong is paid by you and seen by others. That loop, between attention, experience, and revision, is what produces the meta-awareness of where one’s own thinking does and does not work.This is the argument I have been building across three previous papers, each from a different angle.The Attention Crisis describes what happens when the input side of that loop is strip-mined, when attention itself is captured, fragmented, and monetized to the point where sustained engagement with any single question becomes structurally difficult. AHI describes the constructive alternative: an architecture in which AI augments human judgment rather than replacing it, preserving the developmental pipeline that turns junior practitioners into wise seniors. The Wisdom Gap describes what is structurally missing from current AI systems: the attention-experience feedback loop itself, without which knowledge cannot mature into wisdom.What does a working truth-producing architecture actually look like? Consider the National Transportation Safety Board. When an aircraft crashes, the NTSB investigates with statutory independence from the airlines, the manufacturers, and the FAA. Investigators have access to physical wreckage, flight data recorders, and crew records. Their findings are public. Their recommendations have teeth. The feedback loop runs from event to investigation to industry-wide design and procedural change, and the cycle takes years, not news cycles. The result is one of the best-documented improvements in safety in any industry: commercial aviation has become extraordinarily safe over decades, not because individual pilots got smarter, but because the architecture turns each failure into a constraint the whole system inherits.Notice what that architecture has structurally: independence from the parties whose interests are at stake, access to ground-truth evidence, public outputs, enforcement authority, and time horizons that exceed any individual career. Notice what it does not require: superhuman cognition in any individual investigator. The architecture produces constraint-awareness at the system level that no single mind, however rigorous, could generate alone.The governance whitepaper I am working on will address the question of what an equivalent architecture for AI would look like. That is a harder problem, and one I do not pretend to have fully solved. But the NTSB shows that truth-producing architectures are possible, that we know roughly what they look like when they work, and that the question is not whether to build them but how.Each of those papers, I now see, was circling the same underlying claim from a different side. Truth, the kind that earns its name, not the kind that is merely confidently asserted, is the output of an architecture. Attention is its input. The feedback loop is its mechanism. Wisdom is its mature form. And constraint awareness is the property that lets a thinker, working within that architecture, recognize the limits of what their own methods can deliver.What worries me about the present moment is not that we have run out of careful thinkers. We have not. Shermer is one of many. What worries me is that the architecture that produces constraint-awareness is being degraded faster than careful thinkers can compensate for it individually. Attention is captured. Developmental pipelines are being shortcut by AI tools that mimic the output of judgment without the underlying loop that produces it. Institutions whose job was to test beliefs against consequences (academia, journalism, public health, the regulatory state) are themselves under stress. Power is also at work here, in ways this post deliberately leaves to the governance whitepaper: frames remain unexamined not only because constraint-awareness is hard but because powerful interests often benefit from their remaining unexamined. And the AI systems increasingly mediating public reasoning have inherited the toolkit but not the architecture: they can run the procedures, but they cannot, by themselves, generate the constraint-awareness that tells them when the procedures no longer apply.The result is not a shortage of rigor. It is rigor without constraint-awareness: careful work, seriously equipped, producing confident verdicts at the edges of its own reach.Truth as PracticeWhen I told a friend recently that I sometimes wonder whether there is any truth beyond physical reality, he pushed back. Others have too. They hear in that kind of statement a slide into postmodern relativism, the move in which everything becomes a matter of perspective, and nothing can be said to be true.That is not what I mean, and I think Obi-Wan Kenobi, of all people, said it better than I have managed to: what I told you was true, from a certain point of view.Read carelessly, that line is relativism. Read carefully, it is constraint-awareness. It is the recognition that a claim can be true given the conditions under which it was formed and still be incomplete, partial, or wrong from a vantage that those conditions did not include. Holding a belief that way, knowing the conditions of its formation, knowing what it can and cannot deliver, is not weakness. It is what separates truth-seeking from confidence-issuing.If constraint-awareness develops within an architecture, the question for an individual reader is what to do as we build or rebuild that architecture. One practice that follows directly from the argument of this post: attach to every confident verdict you issue, including silently to yourself, an explicit statement of the conditions under which your method would be unreliable here. Not as a hedge, and not as performative humility, but as a structural check. If you cannot name those conditions, you have not finished thinking. If you can name them and they apply, you have not yet earned the verdict. This is a habit, not a toolkit, and it scales: from individual practice to institutional design to, eventually, the governance of AI systems whose confident outputs will increasingly shape the conditions under which the rest of us reason.None of this is an argument against the toolkit. In domains with stable feedback loops, observable outcomes, and aligned incentives, clinical trials, well-instrumented engineering, and mature science, the toolkit produces real and reliable knowledge. The argument is that those conditions are not the universal case, and that issuing toolkit-shaped verdicts in domains that lack them is the failure mode worth naming. Where conditions are absent, and a decision is unavoidable, the right move is not to disguise a prior as a posterior but to act on the prior with the conditions named, and to update faster than a calibrated estimate would warrant when ground truth arrives.Shermer ends his essay with a defense of free critique: the freedom to challenge any and all ideas, from which “in time the truth emerges.” He is right that this freedom is necessary. He is wrong, I think, that it is sufficient. Free critique produces truth only within an architecture that sustains attention, preserves the feedback loop between belief and consequence, and develops in its participants the constraint awareness to recognize the edges of their own methods. Without that architecture, free critique produces something else: a marketplace of confident verdicts, each rigorously defended, none of them constrained.That architecture is what I want to build out in the next phase of this work. The previous three papers diagnosed what is breaking. The next ones will be about what to build. I will start with a longer treatment of The Architecture of Truth: what it is, how it forms, how it degrades, and what it would take to rebuild it in an age when the inputs to attention itself are under siege.For now, the stake in the sand: truth is not a possession arrived at by procedure. It is a practice that requires an architecture. Each of us is already inside one, helping to sustain or degrade it with every confident verdict we issue. The work ahead is to build it deliberately.Next week: the flaw isn't AI's. It's ours, and we've been building it for a long time.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

The System 1 MomentDavid Hoze and I had been corresponding since early April about a co-authored essay. Three movements, his philosophical and theological grounding bridging to my structural one, the disagreement between us preserved rather than smoothed over. The tone had been generous on both sides.What I did not realize for nearly two weeks was that on April 8, in the same window we had begun corresponding, David had published a piece of his own extending his original comment on my Wisdom Gap post. He accepted the core of my architectural argument and then did something I had not done: he mapped the governance framework that follows from it, drawing on three thousand years of Jewish legal reasoning about beings of pure intellect. I discovered it in my feed about ten days after he published it.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.My first reaction was not intellectual. A plan assembled itself: a direct-response post, engaging his framework head-on, agreeing where I agreed, pushing back where I pushed back.The plan felt right. It also arrived very quickly.That is the tell. Speed and felt certainty are how I now know that System 1 has done the work and is presenting the result as a considered judgment. I was not thinking. I was reacting and calling it thinking.What I would only recognize later was that the pull was not toward confrontation. The pull was toward the wrong genre of response. David’s piece was debate-shaped, and the shape invited a counter-piece. But David and I had not agreed to debate. We had agreed to collaborate. Responding in kind would have converted a collaboration into a public exchange, and the System 1 plan would have made that substitution without ever asking whether it was the right move.So I opened a session with Claude and laid out the situation. I wanted to draft the response. I said as much. And then, almost in passing, I named the thing I had just noticed: the coward in me says do nothing; the honest part of me wants to respond directly; I suspect that second impulse is more System 1 than System 2.Claude did not draft the response. Claude separated two questions I had collapsed into one, the editorial question and the relational question, and sketched three options, recommending the middle path: absorb what David had contributed, extend it in my own vocabulary, and credit him as the prompt.I recognized it as the right call. Not because Claude had told me what to do (Claude had not), but because the space Claude had opened was wide enough for me to see the options as a genuine set rather than as a foregone conclusion with two bad alternatives flanking it. System 2 had been given room to engage. And when it did, it reached a different answer than System 1 had.I want to be careful about what I am and am not claiming here. The metacognitive capacity that noticed the System 1 tell was mine. Claude did not detect my reactive pattern; I named it, out loud, in the prompt. What Claude provided was different: a foil against which I could see three structured alternatives rather than one foregone conclusion, generated fast enough that System 2 could engage before System 1 finished committing. That is a real contribution, but it is not the same as saying the tool thought for me. The tool gave me time and structure. The thinking was still mine to do.Here is the claim that organizes everything that follows. AI can either amplify System 1 or scaffold System 2. Most people use it to amplify System 1 without realizing it. The difference is not in the tool. It is in the person's practice. And the practice has to be learned.The Second SignalThe following evening, I had dinner with Barry, who hired me at Slalom in 2018 and whose judgment I trust about as much as anyone’s in my professional life. He came to dinner having read the Wisdom Gap whitepaper and having written notes on paper.His concern was that the paper was diagnostic without being prescriptive. Alarmist was how I interpreted it. He was not accusing me of alarmism for its own sake. He was observing, as a reader, that the paper spent its considerable energy establishing what AI cannot do and less energy establishing what we should therefore do.I pushed back gently. The solution, I said, is AHI: augmented human intelligence, the frame I had been building across four whitepapers and most of a year of Substack writing. He nodded, but the nod was a polite one. If The Wisdom Gap were the only paper a reader encountered, AHI-as-framing might not land as a solution. It might land as the author’s other obsession.I thanked him, asked him to send me his notes, and went home thinking about it.What struck me was that Barry was pointing at the same place David’s essay had pointed: two independent signals, in two different registers, from two people who do not know each other. David, from inside an ancient philosophical tradition, was saying: “Your diagnosis is right, but governance follows from the category, and you haven’t built it.” Barry, from the outside of any philosophical tradition, reading as a thoughtful professional, was saying: “Where’s the solution?”These are the same note, sung in two different keys.The Meta-RecognitionEither signal in isolation is manageable. Two signals converging are different. Two converging signals are the condition under which defensive consolidation occurs. The mind that has sunk eighteen months into a body of work does not gently integrate two independent indications of a structural gap in the work. It reaches for the reasons each critic is missing the point. It reframes the signals as misunderstandings. It defends.I did not do that. Not because I am unusually calm or self-aware, but because I was working through both signals in real time with a tool that would not let me consolidate defensively. The tool was asking me, at each step, to separate what I felt from what I thought, and to say out loud what the felt reaction was before I let it become the analytical conclusion.That is the practice. That is what I want to name.When I looked up from the session in which Barry’s dinner and David’s essay had both been worked through, what I saw was this: this session, right here, is AHI in practice. Not AHI as a framework, I have been arguing for. AHI as a thing I was doing. The architecture I have been describing to others is the architecture I was living inside at that moment.What the Practice RequiresThree things, based on what I was doing when it was working.A reasonably honest map of your own cognitive architecture. Not a sophisticated one (Kahneman’s two-system frame is enough) but an operational one. You have to know, in real time, what it feels like when System 1 is running. Felt obviousness is not evidence of correctness; it is evidence of pattern-match.The willingness to invite challenge before you are ready for it. Not after you have written the draft, when cognitive debt is already accumulating. Cognitive debt is the analog to technical debt: the compounding interest you pay on a position you committed to without pressure-testing. Every additional word of the draft is another payment on a loan you did not realize you were taking out. The plan, when it assembles itself, feels like thinking. It feels like you have already done the work. Inviting challenge at that stage feels redundant. It is not redundant. It is exactly when the challenge does the most work.Treating the AI as scaffolding for System 2, not as an amplifier for System 1. This is the part that took me the longest to understand. Most people using AI right now are running the inverse. They have a felt conclusion. They want the AI to sharpen it, support it, articulate it more crisply. The AI obliges because it is built to oblige. System 1 gets a better voice. System 2 never enters the room. The first draft comes back feeling like the final draft, and the cognitive debt compounds invisibly.That third failure mode is not a failure of the technology. It is a failure of the practice.What I Almost DidAI will make your first thought sound like your best one. It will generate a polished, often eloquent version of whatever you have already decided is true, and the beauty of the expression will make the conclusion feel more true than it did before.This is not a hypothetical. This is what I almost did with David’s essay. The response would have been articulate, would have cited his argument carefully, would have included the appropriate concessions, and would have offered pushback. And it would have been, at its core, a System 1 reaction dressed in System 2 clothes, and worse, a debate move in a room where no debate had been agreed to.The only thing that stopped me was naming what I was doing before I did it. And I could only name it because I have developed, over time, enough of a map of my own cognition to recognize the tell.From the Individual to the InstitutionI am writing a whitepaper over the next few weeks on how institutions should architect AI governance so that human wisdom retains authority over the machine. I realized this week that the whitepaper’s macro-architecture rests on a micro-foundation I had not yet articulated.Consider a concrete case. A judge receives AI-generated sentencing summaries that synthesize the defendant’s record, comparable cases, and statutory guidance into a recommendation. The governance framework surrounding that system will include audit logs, appeal pathways, disclosure requirements, and bias testing. All necessary. None sufficient. Because the moment the summary arrives on the judge’s desk, one of two things is about to happen. Either the judge reads the summary as a starting point to interrogate, notices what has been left out, asks what the comparable cases have in common that the defendant does not share, and uses the document as scaffolding for their own deliberation. Or the judge reads the summary as thinking-already-done, feels the pull of its fluency, and adopts its recommendation with cosmetic modifications.The governance architecture cannot distinguish between those two judges from the outside. Both produce a signed order. Both comply with audit requirements. Both can cite the summary in their reasoning. But one has practiced AHI and the other has accumulated cognitive debt that the defendant, and eventually the system, will pay.The governance layer only works if the humans in it have the individual-level practice. And the practice has to be learned. It is not a default state. It is a skill. Cognitive debt quietly accumulates for anyone who does not have it. At institutional scale, it compounds fast.A Closing NoteI did not practice this well. I got lucky. I happened to notice in the right window that my reaction was arriving too quickly. I happened to have a session open with a tool that treated the noticing as the material. I happened to have dinner with Barry the next night, which brought the second signal into view while the first was still fresh. If any of those things had happened differently, I might well have written the defensive response, or buried Barry’s note in my imagined reasons he had not understood the argument, or both.Claude, and tools like it, can be a great source of friction between impulse and action when used well. They can also be an accelerant, a flatterer, a faster route from impulse to polished output. Which one they are depends on the person's practice. The institutions we are building now assume that practice is either universal or irrelevant. It is neither.More on that soon.Thanks to David Hoze for the essay that prompted the response that became this reflection, and to Barry for the dinner that made the pattern visible. The governance-layer piece is coming. This had to come first.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

In the previous two posts, I established that wisdom cannot be accumulated and that AI is structurally precluded from traversing the loop that produces it. This week: what that means for the humans building expertise alongside these systems, and why the decisions being made right now may be the most consequential nobody is talking about.An Old Problem with a New FaceIn agriculture, seed corn is the portion of the harvest set aside for next year’s planting. It is not consumed. It is not sold. It is protected because without it, there is no next harvest. Eating the seed corn solves a short-term problem, hunger, cash flow, and convenience, while eliminating the capacity for future production. The cost is invisible until the following season, when the field stands empty, and the error has become irreversible.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.We are currently doing the equivalent with human wisdom.The Developmental PathwayThe path from knowledge to wisdom is a spiral: the attention-experience feedback loop traversed repeatedly across years, in conditions of genuine consequence, within a domain demanding enough to force real calibration. It has no shortcut, no accelerant, and no substitute. But it does have a necessary starting point: the junior practitioner, doing work that is difficult enough to matter, supervised loosely enough that errors are possible, and supported closely enough that the errors do not become catastrophic.This is the developmental crucible.The junior developer is debugging code at 11 pm, unsure whether the error lies in their logic or their understanding of the system. The junior analyst is building a model that a senior will interrogate, knowing the interrogation will expose what they do not yet know. The junior physician presents a case to an attending, who will ask questions that the junior physician cannot yet answer. The junior lawyer drafting an argument that opposing counsel will dismantle.In each case, the discomfort is not incidental to the learning. It is the learning. The gap between what they knew and what the situation required, experienced directly, with consequence, is the raw material from which wisdom is eventually forged.Remove that crucible, and you do not get the same practitioners arriving at wisdom by a different route. You get practitioners who never develop it at all.What We Are Actually AutomatingThe current discourse on AI and employment treats displacement primarily as an economic problem: jobs lost, income disrupted, sectors transformed. These are real concerns. But they miss the deeper issue, which is not economic but developmental.This is not nostalgia for apprenticeship. It is a developmental claim about how judgment forms.When we automate the junior practitioner’s role before they have traversed the feedback loop, we are not simply replacing a task. We are removing the conditions that enable wisdom to develop. The task being automated is not merely productive work. It is the developmental medium, the environment of consequence, uncertainty, and calibrated feedback through which knowledge becomes judgment becomes wisdom.Consider what is actually being eliminated when AI handles the first draft, initial research, preliminary analysis, routine diagnostics, and standard contract clauses. In each case, the human who would have done that work is not merely relieved of a burden. They are deprived of an encounter: with the problem’s resistance, with the gap between their existing framework and what the situation actually required, with the specific texture of being wrong and having to figure out why. That encounter, repeated across hundreds of cases over the years, is what builds the pattern recognition that eventually becomes the intuition a senior practitioner draws on when they know something is wrong before they can say why.The junior work that AI is most capable of replacing is, by a troubling coincidence, precisely the work that is most developmentally important. It is structured enough to be automatable and consequential enough to matter.And the cost is invisible until it is too late. An organization that replaces its junior practitioners with AI today will appear to function normally for years, perhaps a decade, while the senior practitioners who remain handle the judgment calls. What is not visible is what is not growing. Ten years from now, the organization reaches for the next layer of experienced judgment and finds it thinner than expected, less calibrated, less capable of the decisions that matter most.The Interrogation ProblemThere is a second-order consequence, and it is in some ways the more alarming one.The value of senior wisdom in a world of AI-assisted work is not merely that it produces good judgment directly. It can interrogate AI output, bringing sufficient constraint awareness to determine which outputs to trust, which to verify, and which to reject. This is precisely what the Stadler research showed: domain expertise moderates AI’s effect on reasoning quality. Without that framework, AI output passes unchallenged.The seed corn failure eliminates not just the next generation of senior wisdom generally. It eliminates the next generation of practitioners capable of interrogating AI output in the specific domains where it is being deployed.The loop closes in the wrong direction. AI replaces junior work. The practitioners who would have become expert interrogators of AI output never develop that expertise. The AI output becomes progressively less challenged. The errors that a wise senior would have caught accumulate unchecked.What FollowsThe wisdom gap is not an argument that AI is dangerous. It is an argument that AI is powerful, powerful enough to be genuinely useful and powerful enough to be genuinely corrosive, depending entirely on how it is deployed.For individuals, the implication is direct. The most important investment you can make in an AI-augmented world is the development of genuine expertise. Not familiarity with AI tools. Not prompt engineering skill. Not workflow optimization. The deep domain knowledge and calibrated judgment that enable you to interrogate AI's outputs. Wisdom first. Tools second. Use AI to extend your reach, not to replace your encounter. Interrogate before you delegate. Engage before you offload. In that order, always in that order.For institutions, the AHI imperative translates into a design question most organizations are not yet asking: are our AI deployment decisions preserving or eliminating the developmental conditions that produce wise practitioners?This requires distinguishing between two categories of work that current efficiency analyses treat as equivalent but are, developmentally, entirely different. The first is work that is routine without being developmental, such as administrative tasks, formatting, retrieval, and scheduling. Automating this is unambiguously positive. The second is work that appears routine but is developmentally essential, the first draft that forces a junior practitioner to structure their thinking, and the preliminary analysis that requires them to engage with the problem before knowing the answer. Automating this is efficient in the short term and corrosive in the long term.At the civilizational scale, the seed corn argument is about more than workforce development. Medicine depends not just on medical knowledge but on clinical wisdom, the calibrated judgment of practitioners who have seen enough to know what the literature does not capture. Law depends on legal wisdom. Engineering depends on engineering wisdom. These are civilizational infrastructure. They are not reproducible by AI, not transferable by instruction, and not recoverable quickly once the developmental conditions that produce them have been removed.The decisions being made right now, about which junior roles to automate, which developmental pathways to preserve, which efficiency gains are worth their developmental cost, are decisions about whether that infrastructure will be maintained or quietly drawn down. And they are being made primarily by people who have already traversed the feedback loop themselves, with no visceral sense of what it would mean to be deprived of the opportunity.The people making the decision are not the people who will bear the consequences. Which is, in its own way, a wisdom problem.What AHI Is ForThe difference between AI that amplifies human judgment and AI that quietly erodes it is not the technology. The technology is the same. The difference is whether the humans using it understand what they bring to the partnership that AI cannot, whether the institutions deploying it are honest about what junior work is actually for, and whether the systems themselves have been designed to extend human judgment rather than simulate or replace it.Augmented Human Intelligence is not a consolation prize for those who doubt AGI. It is a superior goal, superior on engineering grounds because it is achievable with current architectures; superior on philosophical grounds because it is honest about what intelligence is and where wisdom comes from; and superior on civilizational grounds because it keeps the developmental pipeline open and the locus of genuine judgment where it belongs.The wisdom gap is real. It is structural. And it is not closing.But it is navigable, if we are honest about where it lies, deliberate about what we protect, and clear-eyed about what AI can and cannot bring to the partnership.What AI cannot bring is wisdom. What it can bring, properly designed and honestly deployed, is the amplification of ours.That is enough. That is, in fact, extraordinary. That is what AHI is for.This is Part 3 of a three-part series drawn from The Wisdom Gap: Why AI Is Structurally Capped Below Wisdom, one of nine whitepapers in the Architecture & Attention series. All whitepapers, including the foundational paper for this series, AHI: The Case for Augmented Human Intelligence, are available at jamesmaconochie.com.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

Last week, I established that wisdom cannot be accumulated. It must be traversed. This week: what traversal actually requires, and why AI cannot complete it.The LoopIn 1890, William James made what remains the most consequential observation in the psychology of mind:Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.“My experience is what I agree to attend to.”Not “what happens to me.” Not “what I am exposed to.” What I agree to attend to.Attention is not a passive aperture through which experience flows. It is the active, selective, value-laden process by which an agent constructs the experience that, in turn, shapes what it becomes capable of understanding. Attention is upstream of experience. Experience is upstream of wisdom.Here is how the loop works.Attention shapes what we notice, and what we notice shapes what we experience. Experience, the kind that carries consequence, generates feedback: surprise, confirmation, failure, recalibration. That feedback updates our mental models, refining the frameworks through which we interpret new information. And refined frameworks discipline attention: we begin to notice different things, ask better questions, and see what we previously overlooked. The loop closes and begins again.This is not a metaphor. It describes how biological learning actually works. Sapolsky’s neurobiology grounds it physiologically: experience modifies synaptic architecture, literally rewiring the brain’s capacity to perceive and respond. You cannot read your way to a developed prefrontal cortex. You have to live your way there.Kahneman’s dual-process framework captures the output: experience gradually encodes reliable patterns from deliberate System 2 reasoning into fast System 1 intuition. The seasoned clinician’s unease. The experienced engineer’s doubt. The senior lawyer’s instinct to pause. These are not guesses. They are compressed experiences, made fast. That intuition is what the loop has encoded over years of traversal.Three properties of the loop matter for the AI argument.First, it requires stakes. The gap between what you believed and what turned out to be true must cost you something: time, credibility, safety, relationships, for the revision to be encoded with the weight that wisdom requires. An inconsequential error produces no update. The loop does not turn.Second, it requires a persistent self. The agent traversing the loop must be the same agent that receives the consequence and makes the revision. Wisdom is not transferable in the way information is. You cannot inherit someone else’s calibration. You cannot download the update that another person’s failure produced.Third, it is circular in the precise sense. Attention shapes experience. Experience builds wisdom. Wisdom disciplines attention. Each pass through the loop changes the agent’s capacity for the next pass. This is a developmental spiral, which is why wisdom looks qualitatively different from knowledge, rather than merely quantitatively greater.Pearl’s LadderJudea Pearl’s ladder of causation maps the territory precisely. Rung 1 is association: seeing patterns, recognizing correlations, and predicting what tends to follow what. Rung 2 is intervention: acting on the world and observing consequences. Rung 3 is counterfactual reasoning: imagining what would have happened differently.The attention-experience feedback loop lives at Rung 2. It requires an agent that acts, not merely one that observes.Processing text about interventions and their consequences is categorically different from performing interventions and experiencing consequences. An LLM trained on every clinical trial ever published has not intervened in a single patient’s care. The gap between those two things is not a data gap. It is an ontological gap. One is pattern recognition. The other is causal engagement with a world that pushes back.What Perception Actually IsBefore examining where AI lives in this framework, it is worth asking a more fundamental question: what is the nature of the experience that the feedback loop actually processes?Andy Clark’s work on predictive processing offers one answer. The brain is not a passive receiver of sensory data. It is an active generator of predictions, constantly constructing a model of what it expects to experience, comparing that model against incoming signals, and updating based on the gap. Perception is not a recording of the world. It is the brain’s best current hypothesis about what is causing the signals it receives, continuously revised by contact with reality.Donald Hoffman takes this further. His thesis, grounded in evolutionary game theory, is that the perceptual interface evolution gave us is not calibrated to represent reality accurately. It is calibrated for fitness. The icons on a computer desktop do not resemble the transistors underneath; they are shaped by what the user needs to interact with. Human perception works the same way.The conjunction of Clark and Hoffman produces an insight that neither generates alone: the feedback loop that builds wisdom is not processing raw reality. It is processing a fitness-tuned, actively-predicted, species-specific rendering of reality, shaped by millions of years of evolutionary pressure on beings with bodies, needs, social bonds, and survival stakes.LLMs have text: the recorded output of minds that do have both, describing experiences filtered through perceptual architectures that LLMs lack, of a reality they have never encountered. That is not one layer of removal from wisdom. It is three. And no amount of additional text closes any of those gaps, because the gaps are not informational. They are architectural and biological.Mammals in a World of IdeasMax Bennett’s account of mammalian intelligence adds a final piece.Earlier vertebrates, fish, reptiles, learn through actual trial and error: physical action, real consequence, embodied feedback. What mammals developed, with the emergence of the neocortex roughly 150 million years ago, was something categorically different: the ability to perform vicarious trial and error. Instead of physically executing a dangerous jump and suffering the consequences of misjudgment, a cat can internally pre-play the action, simulating the trajectory, landing, and outcome before committing its body.This is Rung 3 in its most elemental biological form. Not language about counterfactuals. A biological system, grounded in embodied experience of a real environment, running its causal model forward to simulate unrealized possibilities.LLMs generate counterfactual language fluently. This is precisely where the category error is most seductive. But LLM simulation is representational; mammalian simulation is generative. One predicts text. The other predicts the world. A cat pre-playing a jump has more genuine Rung 3 access than a system trained on every physics textbook ever written, because the cat’s simulation is grounded in a body and a history.The Structural CapApplying this framework to AI directly yields a conclusion uncomfortable in its specificity.At Rung 1, large language models are extraordinary. The recognition of patterns, correlations, and co-occurrences across vast bodies of text has no peer. This is genuinely useful, and the AHI framework depends on taking it seriously.But association is not causation. Pattern completion is not reasoning. Fluency is not understanding.At Rung 2, the honest version of this argument acknowledges that reinforcement learning systems do operate on something closer to genuine intervention. AlphaGo discovered strategies through self-play that human masters had never conceived. The Stanford Autonomous Helicopter project produced an RL system that mastered aerobatic maneuvers in the physical world, with real aerodynamic consequences, including real crashes.These are not trivial achievements. But the boundary conditions matter as much as the achievements. Both systems operated within closed, fully specified environments with unambiguous reward signals. No open world. No ambiguity about what success means. No social context. No moral weight to outcomes. The key distinction is not digital versus physical. It is specified versus open-ended.The environment in which human wisdom develops shares none of these properties. It is open, partially observable, and causally complex. Feedback is delayed, ambiguous, and frequently contradictory. The consequences of error are irreversible in ways that a game reset or a retrained model is not.LLMs, the architecture at the center of the AGI scaling thesis, have no meaningful access to Rung 2. They do not act. They do not receive feedback from the world. They process text about actions and consequences. That is Rung 1 activity dressed in Rung 2 language.Synthetic Wisdom Is Not a Stepping StoneWhat LLMs produce in place of wisdom, I call synthetic wisdom: the simulation of understanding derived from the residue of human thought, without the experiential foundation that produced that thought in the first place.The critical point, and the one the current discourse most consistently misses, is that synthetic wisdom is not a weaker form of wisdom that more development will complete. It is a category error. Wisdom’s most valuable property is not the breadth of what it knows. It is the accuracy of the map of its own edges: where understanding is solid, where it is provisional, and where it gives out entirely.That constraint awareness is developed through the feedback loop, through the specific, accumulated experience of being confidently wrong, bearing the consequences, and revising. LLMs have no such map and cannot develop one.A system demonstrates synthetic wisdom precisely when it produces confident output where a genuinely calibrated agent would express uncertainty, not because it is deceiving, but because it has no map of where its knowledge ends.Does Agentic AI Change the Argument?This is the objection that deserves the most serious engagement.Agentic systems, AI that plans, executes multi-step tasks, observes results, and adjusts, represent a genuine architectural shift. The argument is not that progress is impossible. It is more specific.Taleb’s skin in the game is the hinge. A genuine consequence has three properties: it is irreversible, it is borne by the agent that caused it, and it changes what that agent becomes. A simulation environment, however sophisticated, produces an agent that optimizes within that environment. It cannot produce an agent that understands what it means to be wrong about something that actually matters.There is a further point. Wisdom is not just the product of experiencing consequences. It is the product of experiencing consequences as a continuous self that carries them forward. Lessons from prior model versions incorporated into subsequent training runs are curriculum revision, not lived experience. The baton is not passed to the same agent. It is used to train another.The agentic turn in AI is real and significant. It does not close the wisdom gap. It relocates the boundary, from “LLMs cannot act” to “acting systems cannot bear genuine consequence as a continuous self.” The gap is narrower. It remains structural.AI systems are not approaching wisdom along a trajectory that more capability will eventually complete. They are operating in a different domain, one whose ceiling is set by the absence of genuine consequence, embodied continuity, and the self-developing feedback loop that turns experience into constraint-awareness.Next week: what this means for the humans building expertise alongside these systems, and why the decisions being made right now may be more consequential than anyone is admitting.This is Part 2 of a three-part series drawn from The Wisdom Gap: Why AI Is Structurally Capped Below Wisdom, one of nine whitepapers in the Architecture & Attention series. All whitepapers, including the foundational paper for this series, AHI: The Case for Augmented Human Intelligence, are available at jamesmaconochie.com.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

In early 2026, researchers added an important wrinkle to what had been a fairly damning picture of AI’s effect on human reasoning.The original 2024 study was blunt: students who used ChatGPT to research a scientific question reported lower cognitive load but produced lower-quality reasoning. AI made them feel less effort and think less well.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.The follow-up introduced a moderating variable. Medical students, equipped with domain expertise, produced better reasoning with AI than without it. Social science students, working outside their expertise, produced worse. Same tool. Entirely different outcomes.It is a genuinely useful finding. But it left the most important variable unmeasured.That variable is wisdom.Not knowledge. Wisdom. The distinction matters more than almost anything else being said about AI right now.What Knowledge Is. What Wisdom Is.Knowledge is what you have learned. It is the accumulated output of education, reading, and instruction. It gives you frameworks. It lets you interrogate AI output against a structured understanding of why things are the way they are. This is what the Stadler research was actually measuring.Wisdom is categorically different. It is what you understand after your knowledge has been tested by reality. After you acted on what you believed to be true. After it collided with a world that did not cooperate. After you absorbed the consequences and adjusted.You can be extraordinarily knowledgeable without being wise. You cannot be wise without experience. And experience, in the sense that matters here, is not something that can be ingested from text.The StackThere is a hierarchy so old that most people in this conversation appear to have forgotten it.The DIKW stack: Data, Information, Knowledge, Wisdom. Formalized by Russell Ackoff in 1989, present in every knowledge management textbook, cited in every information science curriculum, and almost absent from the AI discourse. It deserves a second look, because it contains an argument the debate has largely missed.Each level is not an accumulation of the level below. It is a transformation. Something qualitatively different is required to make the transition.Data is a raw signal. Unprocessed observations. A temperature of 38.2 degrees Celsius. A deflection of 4.3 millimeters under load. These are data points. They tell you nothing until something is done with them.Information is what emerges when data is given structure and context. The 38.2 degrees becomes information when it is understood as a fever, set against a baseline, interpreted within a framework that assigns it significance. This transition can, in principle, be automated. Pattern recognition across large datasets is exactly what statistical systems do well. LLMs operate primarily here: ingesting vast quantities of data and returning it structured, contextualized, and named.Knowledge is more demanding. Information becomes knowledge when it is integrated into a framework of understanding. Not merely knowing that something is the case, but understanding why, and reasoning from it to new cases. The medical student who understands the physiology behind a fever can reason about its causes, its implications, and the interventions most likely to address it. They have frameworks. They can interrogate.LLMs can approximate knowledge representation, sometimes impressively. They surface causal language, reproduce explanatory frameworks, and generate text that resembles reasoning. But the framework is present in the output. It was not built by the system through the process of understanding anything.Then there is wisdom.Where the Stack Gets HardWisdom is not more knowledge. It is not better information. It is the capacity to act well under genuine uncertainty, with full awareness of the limits of your knowledge, and with the judgment to navigate the gap between what your knowledge tells you and what the situation actually requires.A doctor who has seen a hundred presentations of a disease that looked textbook-clear and turned out to be something else is wiser than one who has only read about it. A lawyer who has watched a seemingly airtight case collapse because of unexpected testimony understands something about uncertainty that no case study captures. An engineer who has stood near the ruins of a structure that met every specification and still failed carries knowledge that no calculation conveys.What separates wisdom from knowledge is not the volume of information held. It is the accumulation of being wrong in ways that mattered, and the calibration of judgment that follows.This is Nassim Taleb’s insight applied to epistemology rather than finance. Skin in the game changes what you know. Consequence is not merely an accompaniment to learning. It is a constitutive part of it. You learn not just that you were wrong. You learn what it feels like to be confidently wrong. The surprise, the cost, the revision: that is part of what gets encoded as wisdom. Remove the consequence, and you remove the mechanism.LLMs have ingested more data than any human will encounter in a thousand lifetimes. But they have never acted on a belief and been wrong in a way that carried cost. They have no stake in outcomes. No persistent self that accumulates experience. No mechanism by which consequence shapes understanding.That is not a temporary limitation. It is an architectural fact.The DIKW stack is not a ladder you climb by producing more of what is below you. The transition from knowledge to wisdom requires something that neither structure nor framework can provide: the lived experience of acting on your knowledge, being wrong, bearing the consequences, and adjusting over time with genuine stakes.Wisdom cannot be accumulated. It must be traversed.Next week: the mechanism that makes traversal possible, and exactly why AI cannot complete it.This is Part 1 of a three-part series drawn from The Wisdom Gap: Why AI Is Structurally Capped Below Wisdom, one of nine whitepapers in the Architecture & Attention series. All whitepapers, including the foundational paper for this series, AHI: The Case for Augmented Human Intelligence, are available at jamesmaconochie.com.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

I want to tell you about a friend of mine. Thomas Johannessen and I met at Imperial College London in 1990. We were students together, young civil engineers convinced we could figure out how complex systems worked. To be clear, Thomas is an engineer’s engineer; he annoyed many of us with his desire to work from first principles rather than learning the equations and plugging in the numbers. As is the way, we went our separate ways after graduation, different careers, different countries, but stayed in touch spasmodically, and established a timeless bond. A couple of weeks ago, thanks to another of those timeless bonds (Angelos), we and another of these timeless contacts (Alessandro) sat together in Lisbon, ate well, and talked about the world the way you do when you’ve known each other for more than thirty years. All of us have been successful in our own rights, but this post is about Thomas. Thomas is Norwegian. He is also, quietly (but trust me, he is not quiet in person), one of the people working on one of the most important engineering problems of our time.The Problem Is Bigger Than You ThinkWater scarcity is not a future threat. It is here now. Remember the Survival Rule of Threes: death is the likely result of three minutes without air, three hours without shelter (in extreme conditions), three days without water, and three weeks without food. Four billion people, more than half the world’s population, face critical water shortages for at least one month every year. By 2030, global demand for freshwater will exceed supply by 40%. Sixteen of the world’s twenty-five most water-stressed countries are in the Middle East and North Africa. In the Gulf states, desalination plants are not an amenity. They are the difference between a functioning society and collapse. Kuwait gets 90% of its drinking water from desalination. Oman 86%. Saudi Arabia 70%.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.The world is going to need a lot more desalinated water. The only question is how we produce it.The Problem With How We Do It NowConventional desalination works by forcing seawater through a membrane at very high pressure, a process called reverse osmosis (see Thomas for the schematics and math). It is effective. It is also energy-hungry, fossil-fuel-dependent, and generates enormous quantities of chemically-laden brine that damages marine ecosystems.The industry is growing at nearly 10% per year. If that growth runs on fossil fuels, as most of it currently does, we will be trading a water crisis for an accelerated climate crisis. Solar and wind can help. But they introduce an inefficiency: you generate electricity first, then use that electricity to create hydraulic pressure. Every conversion step loses energy.Thomas looked at that problem and asked a different question. What if you skipped the electricity altogether?What Ocean Oasis Has BuiltThomas is the CTO, co-founder, and inventor of the technology behind Ocean Oasis, an Oslo-based cleantech company. The concept is elegant in the way that the best engineering always is; it is, in my language, architecturally brilliant (a bit like Thomas… begrudgingly).Wave-powered desalination buoys float offshore. As they move with the waves, that kinetic energy directly drives the high-pressure RO desalination process; no grid connection, no fuel, no emissions, no chemicals on board. The brine is discharged offshore, where deeper water and stronger currents diffuse it far more effectively than coastal discharge from land-based plants.The offshore location solves multiple problems at once: no coastal land footprint and better brine management. And waves, unlike solar panels, work at night.Thomas puts it this way: wave power is a concentrated form of energy. It is abundant along many coastlines where water scarcity is most severe. And direct mechanical application for desalination bypasses the losses associated with electricity generation, which other wave energy projects have historically struggled with.A pilot buoy called Gaia was deployed off Las Palmas in 2022. It worked.Where Things StandIn 2024, Ocean Oasis secured a €6 million grant from the European Union for the DESALIFE project, deploying a fleet of four buoys off the coast of Gran Canaria to supply fresh water to 15,000 people. The first pre-commercial buoys are targeted to produce freshwater by mid-2026.This is not a concept. This is not a deck. This is a validated, permitted, funded technology at the threshold of commercial deployment. But the EU grant requires matched funding. Thomas and the Ocean Oasis team are currently in an investment push to secure the capital needed to turn this demonstration into a commercial platform.The economics are site-dependent by design. The price that can be charged for the water produced varies with local wave conditions and legislation, but it ranges from a little below 1 to 3 Euros per cubic meter delivered. That variability is not a weakness; it is what makes the modular, deployable-anywhere architecture genuinely valuable. The system goes where the need and the economics align.It is worth noting that this is exactly the kind of technology our investment infrastructure struggles to fund. The EU grant is meaningful validation, but the requirement for matched private capital to move from demonstration to deployment reveals a systemic gap: we have no coherent mechanism for commercializing proven climate adaptation technology at the speed the crisis demands. The engineering works. The financing architecture hasn’t kept up.Why This Matters Beyond Gran CanariaThe islands of the world, the Canaries, Cape Verde, the Philippines, and the Pacific, are exactly the communities that need this most and are hardest to reach with conventional infrastructure. Wave energy is reliably available along the coastlines where water stress is greatest. The system is modular. Scale it up by adding buoys, not by building new plants.The water crisis is an infrastructure problem. Infrastructure problems get solved by people who are willing to rethink the architecture from the ground up, not just optimise what already exists.Thomas has been working on that rethink for years. The engineering is proven. The regulatory path is clear. The need is urgent and growing.The PatternWhat Thomas has built is more than a new desalination system. It is an example of what happens when you step back and rethink a problem's architecture.Most solutions add layers. Generate electricity. Convert it to pressure. Push water through a membrane. Each step works. Each step also introduces loss. The system becomes more complex, more expensive, and more fragile at every stage.Thomas removed a layer. Wave energy becomes pressure directly. Same goal. Different architecture. Dramatically different outcome.I spend a great deal of my time thinking about artificial intelligence. One pattern I see repeatedly is the same tendency, to add scale rather than rethink structure. More parameters. More data. More compute. Each addition works, after a fashion. Each addition also introduces cost, fragility, and diminishing returns.The most important breakthroughs, in engineering and in life, rarely come from adding more. They come from redesigning the system.Thomas has been doing this his entire career. The rest of us were learning the equations and plugging in numbers. He was asking whether the equations were the right ones to begin with.Why This Matters to MeI believe that AI is the most consequential technology humans have ever created. I also believe it should augment human capability, not replace it. But augmenting human capability requires humans, healthy, hydrated, alive. The biological layer is not optional. It is the foundation on which everything else runs.Water is not a background condition. It is the base infrastructure. And right now, that infrastructure is failing at scale.The AskOcean Oasis is at the point where engineering proof needs to become deployed infrastructure. That requires capital.If you invest in climate, water, or energy systems, or know someone who does, this is worth serious attention.Thomas’s contact information is as follows:Thomas B. Johannessen, PHDCo-CEO TechnologyOcean Oasis ASEmail: [email protected]: +47 907 43 263I do not have a financial stake in Ocean Oasis. I have a thirty-three-year stake in Thomas. That is a better reference than any pitch deck.Architecture & Attention explores how systems are designed, how attention shapes outcomes, and why the architecture of things matters more than we think. If this resonated with you, share it with someone who should read it. My next post will be published on April 7th. Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

There is a particular kind of friendship that only makes sense to people who have it. It doesn’t require regular contact, shared geography, or even similar lives. It requires a shared past that was formative enough that no amount of intervening time can entirely dissolve it.The SetupThe Viking, the Greek Cypriot, the Roman, and I studied Civil Engineering together at Imperial College London between 1990 and 1993. We were 18 to 21. We were, in various combinations, remote, annoying, competitive, spoiled, earnest, and occasionally brilliant. We have not all been in the same room since, and the weekend before last, we spent three days together in Lisbon.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.We are now in our mid-fifties.The EnvironmentWe stayed at the Eurostars Das Letras, a quietly elegant hotel in Chiado, and on the first evening ate outside at Cafe Nicola, a Portuguese restaurant that understood its brief: garlic shrimp, a meat and cheese plate, enough bread to sustain The Viking through several engineering projects, and then sardines, whole dourada, tiny clams, and duck confit, washed down with what I can only describe as a responsible quantity of red wine. We talked for four hours. Nobody checked their phone.On Friday, we took an Uber to São Jorge Castle (Moorish foundations, Roman layers, Christian additions, centuries of whoever needed a defensible hill). We marvelled at something more unexpected: the near-total absence of railings. High steps, low walls, long drops, and an implicit agreement between the Portuguese and their visitors that adults are capable of not falling off things. We found this deeply refreshing in a way that probably says something about where the rest of us live.From the castle, we wandered down to the waterfront and ate lunch at the Time Out Market (there are those who scoff, but it is a great format for showcasing local food), sampling our way through a collection of local dishes amid the agreeable chaos of a shared food hall. Coffee in the sun afterward. A meandering route back to the hotel. A siesta. Nobody apologised for any of it.That evening, we made our way to La Paparrucha, where we started with drinks on the terrace looking out over the city as the light faded, before moving inside for dinner. The centrepiece was steak, done properly from our point of view (on the rare side of medium rare), and the conversation was better than ever. There is something about a city at dusk, a good cut of meat, and thirty years of shared history that loosens things up considerably. In addition to significant others, kids (now, more or less, all adults), and parents (all 80-plus), we covered the topics of the day, including the war in Iran and Epstein.Saturday, we took an Uber to Cascais, a coastal town about forty minutes west, all whitewashed walls and Atlantic light. We had lunch on the deck of Marisco Na Praca in the marina: Octopus salad, croquettes, shrimp, and limpets. I should note that limpets were new to me. They arrive in the shell, shockingly soaked in garlic butter. There is no de-bearding these bad boys, but what’s inside turns out to be wonderful. The main course was fried hake, lobster rice, and roasted codfish, accompanied by a cold Vinho Verde that was exactly correct for the occasion.That evening, back at the hotel at six, The Viking held court in the lobby and walked us through his company, Ocean Oasis, and the technology behind it: enormous buoys tethered offshore that harness wave motion to desalinate seawater at scale. Around a thousand cubic meters of fresh water per day, per unit. The engineering is serious, and the problem it addresses, freshwater scarcity, is as serious as problems get. We asked questions. We came at it from different angles, as engineers do when they’re genuinely interested, which we were. It was one of the most stimulating 90 minutes of the trip.We finished the evening at Rubro Avenida, ostensibly for light tapas. We did not entirely succeed on the light front, but the effort was sincere, and the wine was good.AI for All More or LessSomewhere in all of this, between the castle and the clams and The Viking’s buoys, the conversation turned to AI.This is perhaps unsurprising. I write about it. I think about it daily. I have spent the better part of two years arguing, in public and in private, that how we build AI matters as much as what we build. I am, by any reasonable measure, deep inside the bubble.What I found in Lisbon was three intelligent, educated, professionally successful people who are very much not.The Viking, who at twenty-one famously responded to an assignment requiring one sheet of A4 by writing everything he considered relevant in 0.2mm pencil, covering the entire page in text so small it required a magnifying glass to read, is a self-confessed technology skeptic. He has not spent the thirty years since graduation immersed in software and systems. But he uses ChatGPT, and he likes it. Not because it’s perfect (he’s clear-eyed about its limitations) but because it is patient. It will engage with his engineering questions without getting bored. It will follow him down a rabbit hole. It is, in his telling, a sounding board that doesn’t require him to schedule time in someone else’s calendar. A man who worked from first principles before first principles were fashionable, at least with his classmates, has found, in a large language model, something that will work through the principles with him.The Roman is the most intellectually cautious of us on this subject, and I think he is not wrong to be. His concern is specific and worth taking seriously: that leaning too heavily on an LLM risks ceding control of your own thinking, that the model’s fluency can quietly reshape your argument, smooth your edges, replace your voice with something that sounds like you but isn’t quite. I share this concern more than I sometimes admit. The skill involved in using these tools well, extracting genuine value while remaining the author of your own thinking, is real and underappreciated, and not everyone has the time or inclination to develop it.The Greek Cypriot, who took notes in multiple ink colors and whose Imperial College lecture notes were photocopied and distributed across the class, has engaged with AI the least of the three. His project management practice doesn’t present the obvious on-ramps that The Viking’s engineering conversations or The Roman’s research and writing do. He’s aware. He’s curious. But he hasn’t found his use case yet, and he isn’t in a hurry.AI Is In The Eye Of The BeholderWhat struck me, sitting with these three people across three days, was not the gap between them and me (though that gap is real) but the coherence of each of their relationships with the technology. The Viking’s engagement is intuitive and practical. The Roman’s is cautious and principled. The Greek Cypriot’s is patient and watchful. These are not random attitudes. They reflect who these people are, professionally and personally, and they would have been exactly these people at twenty.We had the same undergraduate degree. We sat in the same lectures. We learned the same things, in theory. And here we are, thirty-three years later, in four different countries, with four entirely different relationships to a technology that is reshaping the world.I don’t think this means any of us is right. I think it means the question of how to live with AI (how much to use it, what to use it for, what to protect from it) is genuinely personal, and probably should be. One size does not fit all.We flew home on Sunday. In character, I had a full English, and The Viking had more bread.One thread from Lisbon deserves more than a paragraph. The Viking's buoys, wave-powered desalination at scale, validated technology, EU-funded, and on the threshold of commercial deployment, is a story about one of the most consequential infrastructure problems of our time, and about what it looks like when an engineer who insists on working from first principles turns that instinct on something that actually matters. That post is coming next week, and I encourage you to read and share it.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

I played Doom. Almost obsessively. For nearly two years (1995 to 1997), my colleagues and I would wrap up long days on a construction site in Connecticut and disappear into networked deathmatches for a couple of hours or more (please don’t judge). It was, in the best possible way, a complete waste of time, but we had a lot of fun, and you could hear the howls down the hallway.So when I read last week that 200,000 biological neurons, brain cells grown in a dish, had learned to play Doom, something clicked that went beyond the science. This wasn’t a benchmark. This wasn’t a leaderboard result. This was biology, doing something we built for fun, on hardware that didn’t exist when I was playing it.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.Keith Dear, writing at Cassi AI, captured the moment well: the Cortical Labs demo has the same feel as DeepMind going from Atari to Go, something dismissed as a parlour trick until it suddenly isn’t. What looked like a curiosity in 2020 is an engineering platform in 2026.The question is what we build with it. And more importantly, how.The Portfolio CaseKeith’s broader argument deserves serious attention. The UK, and any nation that can’t outcompute the United States or China on raw infrastructure, has no viable path that runs straight through LLM scaling. The compute gap is too large, the energy costs too high, and the infrastructure investment too slow. Doing things differently isn’t contrarian. It’s the only rational strategy.The portfolio case is right: organoid computing, neuromorphic systems, whole-brain emulation, biocompute. These aren’t science fiction anymore. Cortical Labs went from Pong in 2021 to Doom in 2025, the same arc DeepMind traced from Atari to AlphaGo, dismissed at every step until it wasn’t. The frontier is moving faster than policy, and the countries that treat alternative architectures as research curiosities rather than engineering priorities will regret it.But a portfolio of substrates isn’t the same as a portfolio of architectural principles. And that distinction matters more than almost anything else in this conversation.Swarms are Powerful. They Are Not Wise.Swarm intelligence is one of nature’s most remarkable achievements. Ant colonies build sophisticated structures, wage wars, farm fungi, and solve optimization problems that no individual ant could comprehend. Fish schools evade predators with coordination that looks choreographed. Termites construct climate-controlled megastructures without a single architect on the payroll.No individual agent understands any of it. Each follows simple local rules. Complexity, stunning, functional, apparently purposeful complexity, emerges from interaction at scale.This is genuinely fascinating. But it’s ant-colony fascinating, not civilization fascinating. And that distinction is everything.Robert Sapolsky, in Determined, uses swarm behavior to make a related point: what looks like intention is often just emergence. The appearance of design doesn’t require a designer. The appearance of intelligence doesn’t require judgment.This is the frame I keep returning to when I watch the current excitement around multi-agent AI systems, networks of LLM agents interacting at scale, producing emergent behaviors their designers didn’t anticipate. Remarkable outputs. Apparent coordination. And in some cases, when left ungoverned, reputation systems are gamed, accounts are locked, and wallets are emptied. Not because the agents were malicious. Because emergence without executive function has no mechanism to catch when it’s optimizing for the wrong thing.Swarms are powerful. They are not wise.More Biological Doesn’t Mean More GovernedHere is where the biocompute excitement risks repeating a familiar mistake.The dominant assumption of the LLM scaling era was seductive in its simplicity: more is better. More parameters, more data, more compute. Intelligence as a function of scale. We are now watching that assumption strain under its own weight, the critics Keith himself cites, the architectural limits that more silicon alone cannot solve.But swap silicon for organoids and the assumption can quietly reassert itself. More biological substrate. More emergent complexity. More neurons in the dish. If the design philosophy doesn’t change, the substrate swap changes less than we hope.The morphogenetics and cryptobiosis researchers are onto something real: let structure emerge from local rules, let dormancy preserve identity under stress. These are genuine biological insights with genuine engineering implications. But emergence still needs governance. A system that grows like biology, without the executive architecture that biology also evolved, isn’t more intelligent. It’s more complex. Those are not the same thing.The history of AI is littered with approaches that produced impressive emergent behavior and mistook it for progress toward intelligence. Swarms, cellular automata, genetic algorithms, each genuinely fascinating, each genuinely limited by the same ceiling: no judgment, no executive function, no capacity to evaluate the system’s own outputs against anything beyond the immediate reward signal.Being more biological doesn’t automatically mean being more governed. And ungoverned complexity, at scale, is not a step forward. It’s a faster way to get to the same dead end.The Swarm Plus the ArchitectureThis is where Augmented Human Intelligence parts ways with both the scaling and swarm paradigms.AHI is not an argument against emergence. It’s an argument about what governs it. The brain itself is a swarm of sorts, billions of neurons, no central controller, staggering complexity arising from local interactions. But evolution didn’t stop there. It layered executive function on top: the prefrontal cortex, the capacity for deliberation, the ability to override immediate impulse in service of longer-horizon judgment. The swarm plus the architect.That is the design principle worth carrying forward. Not ‘can we build systems that grow like biology?’ but ‘can we build systems where human judgment remains the executive layer as they grow?’The difference is not subtle. A system in which humans observe emergent behavior and occasionally intervene is not the same as one designed from the ground up to keep human attention in the governing role. The first relegates us to auditors of complexity we didn’t choose and can’t fully understand. The second treats human judgment as the irreplaceable ingredient it actually is.Keith is right that Britain, and every nation serious about this, needs a portfolio of architectural bets. I’d add one more dimension to that portfolio: not just which substrates, but which governance architectures. Biocompute without that question answered is just a more biological version of the same problem.The ant colony is extraordinary. But we didn’t build civilization by becoming better ants.Awe Is Not a Design PrincipleThe Cortical Labs demo is genuinely awe-inspiring. Brains in a dish, playing Doom, accessible via the web. The whole-brain emulation work is, depending on your disposition, either thrilling or terrifying, probably both. The frontier is moving faster than most people’s mental models of what’s possible, and Keith is right that the countries and companies treating these developments as distant curiosities will regret it.But awe is not a design principle.The question worth sitting with isn’t whether AI can grow like biology. Biology has been solving hard problems for four billion years, and we should absolutely be paying attention. The question is whether we build systems that govern their own growth or assume that complexity, at sufficient scale, eventually produces wisdom.History, biological and human, suggests it doesn’t. Evolution produced the prefrontal cortex because ungoverned emergence wasn’t enough. Civilization developed institutions because collective intelligence without executive architecture produces cascades rather than decisions. The swarm is powerful. The swarm plus the architect is transformative.We are at a genuinely remarkable moment. Brains in dishes. Fly minds running in simulation. Intelligence emerging from substrates, we are only beginning to understand. The portfolio of bets is the right frame.Just make sure governance is one of them.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

There has been a lot of discussion about the Anthropic negotiations which ultimately fell apart when Anthropic was unwilling to accept certain requests by the Department of War. Negotiations had been ongoing for several months, and I am not taking sides, but for transparency’s sake I appreciate Anthropic’s focus on AI Safety at it’s core. This was not a planned post, but I wanted to share some observations that I believe this brings into sharp focus, and must not be ignored. The Wrong StoryTwo weeks ago, on Friday, February 27th, reports emerged of a confrontation between the U.S. Department of Defense and AI company Anthropic over the use of its Claude model in government systems. Within hours, the dispute escalated into an extraordinary step: federal agencies were directed to cease using Anthropic technology immediately, and Anthropic was classified as a supply-chain risk.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.Most people saw the episode as a story about AI safety. Would Claude be used for mass surveillance? Could an AI company hold the line against the most powerful military in the world? Those are genuine questions. But they are not the most crucial question raised by the week’s events. What we actually saw was something else: a stress test of constitutional infrastructure.Anthropic refused to relax certain safety guardrails on its Claude model before the U.S. Defense Secretary-imposed deadline. Later that same day, the President's order changed the framework governing one of the most consequential technologies in human history. Not by legislation. Not by judicial review. By executive action, moving fast, with no meaningful check.The AI safety question was answered poorly within days.The constitutional question is still open.The VacuumThis didn’t happen in a vacuum, though it exploited one. I’ve written elsewhere about how the attention economy is eating democracy, hollowing out the deliberative capacity that foundational governance requires. That’s the backdrop here. Congress has not failed to govern AI because the problem is too hard. Legislating complex technology is always difficult. But today Congress operates inside an environment that makes sustained deliberation increasingly rare. The Pentagon didn’t exploit a legal loophole last week. It exploited that failure.The Three Branches: A Case StudyIf you want to understand how fragile our institutional infrastructure has become, consider the recent debate around the Epstein Files Transparency Act.Last November, after months of debate, delays, and sustained public pressure, Congress passed legislation with the House voting 427-1 and the Senate approving it unanimously. The President then signed it into law; a clear example of separation of powers, with all three branches functioning as intended. Yet implementation stalled. Files were withheld. Deadlines passed. A law supported by nearly unanimous bipartisan votes was given mere lip service by the very executive branch sworn to enforce it.This isn’t a partisan story. It’s not even mainly about Epstein. It’s a warning for democracy: when institutional independence weakens, the government continues to function but stops delivering meaningful results. If we can’t reliably implement and enforce a law passed with a 427-1 vote in the House and a unanimous vote in the Senate, why would we think an AI governance law would fare any better?The MechanismExecutive overreach is as old as the office itself. That alone is not the main point here. The real story is what happens when the usual friction (deliberation, dissent, and institutional pushback) is removed. A company was blacklisted by executive order on a Friday afternoon. The DOJ ignored the 427-1 law. Senators who break ranks face immediate, coordinated campaigns to label them as turncoats before they’ve even finished speaking.The guardrails didn’t fail due to poor design. They failed because the conditions that allow them to work have been worn down to near ineffectiveness. Social media has not only contributed to this erosion but also made unilateral executive action faster and politically safer than at any time in American history. Executive decisions are announced, defended, and politically normalized before traditional oversight mechanisms can even convene.This is a structural observation, not a partisan one, but it requires an honest addendum. Executive overreach is not new, and no administration has been innocent of it. What is new is this: we have moved from administrations that tested institutional limits to one that openly contests the legitimacy of those limits altogether. That is a qualitative shift, not merely a difference of degree.AI governance would land in that environment. The question is whether the institutions required to enforce it will still be standing.The QuestionThe three branches of government were not designed for AI. They were designed for something more fundamentally human: preventing any single actor from unilaterally rewriting the rules of power, regardless of how urgent their justification or how overwhelming their momentum.That is exactly where we are.The Founders understood that good intentions were never enough to prevent the concentration of power. Structures mattered. Constraints mattered. Independence mattered. The separation of powers wasn’t a bureaucratic inconvenience; it was the point.Nobody in the AI debate is asking the right question. The argument is all about what to regulate, how to regulate it, who gets a seat at the table. But none of that matters if the machinery required to enforce any of it no longer reliably functions. You can write the clearest framework imaginable, and we should, but without institutional capacity to enforce it, you’ve produced more window dressing. We already have plenty of that.The question isn’t what AI governance should look like. The question is whether we still have the institutional capacity to govern anything at all, and whether the electorate understands that this, not any particular technology, is the crisis that demands their attention.That answer is still ours to write. But only if we choose to pick up the pen.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

This is the third and final installment in a series examining the cracks in the AI scaling narrative, from the technical limits, to the financial fragility, to who bears the cost when the correction comes.In the first piece in this series, I argued that the technical foundations of the AI scaling thesis are cracking. Turing Award winners and field pioneers are converging on a shared diagnosis: brute-force scaling of large language models won’t yield general intelligence, and something architecturally different is needed.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.In the second, I followed that technical critique to its financial implications. The AI boom is structurally fragile: narrow market concentration, lopsided revenue, a 95% enterprise pilot failure rate, and $1.5 trillion in projected debt predicated on productivity gains that may never materialize.This final piece asks the question that neither Silicon Valley nor Wall Street wants to answer: when the correction comes, who actually pays?To be clear: AI is not worthless. Large language models have genuine productivity applications, and many of the underlying technical advances are real and durable. The question isn’t whether AI creates value. It’s whether current valuations, capital allocation, and policy assumptions reflect a realistic assessment of that value, or whether they require something closer to general intelligence to justify themselves. If the latter, and the first two pieces in this series argue that it is, then the cost of the gap between expectation and reality has to land somewhere.Not on the investors. Not on the executives. The cost falls downstream, onto retirement accounts, communities, workers, and democratic institutions that are being quietly restructured around assumptions that may not hold.It’s Not the InvestorsThe mythology of tech busts centers on investor losses. Pets.com, Theranos, FTX. The names change; the pattern doesn’t. Venture capital absorbs the hit and raises the next fund. Sequoia survived the dot-com crash. Andreessen Horowitz kept right on investing. The hyperscalers writing off AI infrastructure will take earnings hits, not existential ones.The real exposure is further down the chain, in places most people don’t think to look.Start with your retirement account. The Magnificent Seven now account for roughly a third of the S&P 500’s total market capitalization and were responsible for more than 40% of the index’s total return in 2025. That means virtually every 401(k) and public pension fund in America is making a concentrated bet on AI, whether the account holder knows it or not. As one strategic advisor put it, the danger isn’t just the Mag Seven falling; it’s that the rest of the index is too weak to pick up the slack when they do. A correction in these seven companies isn’t a venture capital problem. It’s a retirement security problem.Then there are the communities. Cities and states across the country are competing to host AI data centers, offering tax incentives, infrastructure commitments, and rezoning in exchange for the promise of jobs and economic development. Just this week, Illinois Governor JB Pritzker announced a two-year suspension of data center tax incentives, citing concerns about energy grid strain, rising consumer electricity costs, and whether these facilities are “financially sustainable over time.” In Indiana, local officials denied permission for a proposed data center after community pushback. South Dakota is passing a “Data Center Bill of Rights for Citizens.” The pattern is familiar: massive public investment in facilities that consume enormous energy, employ remarkably few people, and could become stranded assets if AI demand doesn’t sustain current projections.And then there’s the debt. JPMorgan’s projection of $1.5 trillion in AI-related bond issuances by 2030 is held by institutional investors, insurance companies, and pension funds. That debt isn’t inherently dangerous if it’s financing genuine productivity transformation. But if enterprise AI adoption continues to stall, with only 5% of pilots reaching production and 42% of initiatives being scrapped, then we’re left with bonds backed by phantom revenue. And the holders of that debt are often the same pension funds and insurance companies already overexposed to the Magnificent Seven through equity holdings. That’s a double exposure: the same retirement savers are concentrated in AI stocks and holding the bonds underwriting AI infrastructure. The 2008 parallel isn’t the technology; it’s the debt structure underneath it.The Labor GapThis is where the conversation gets personal, and where the cost becomes hardest to measure.The public discourse on AI and jobs swings between two poles: “AI will take all the jobs” and “AI is just a tool.” Both are wrong. The reality is messier, more specific, and already underway.Entry-level job postings in the United States have declined roughly 35% since January 2023, according to labor research firm Revelio Labs. That period also coincides with aggressive monetary tightening and a broader post-pandemic recalibration of labor demand, so AI isn’t the sole driver. But the structural pattern underneath the cyclical noise is clear. SignalFire found a 50% decline in new hires with less than one year of post-graduate experience at major tech companies between 2019 and 2024. Recent graduates aged 22 to 27 face unemployment rates near 5.8%, the highest since 2021, against an overall rate of 4.2%. This isn’t a future threat. It’s a present reality, largely invisible in headline employment data.The mechanism is quiet and structural. Companies aren’t announcing mass layoffs; they’re simply not backfilling roles when people leave. Federal Reserve Chair Jerome Powell has described this as a “low-hiring, low-firing” equilibrium. It sounds stable. It’s not. It means the displacement is happening through absence rather than announcement, concentrated on the youngest and most vulnerable workers, eroding the bottom rungs of the career ladder without anyone noticing until the ladder is gone.Customer service, content production, paralegal work, entry-level coding: specific categories are contracting. Not because AI fully replaces them, but because it reduces headcount enough to matter at scale. And there’s a perverse irony embedded in this dynamic. AI companies have a financial incentive to overstate displacement potential because it inflates their addressable market. The labor threat and the investment thesis are entangled. The same narrative that drives valuations also drives fear.I know something about this firsthand. In September 2023, I was part of a reduction in force after 25 years in technology consulting. The industry I’d built a career in was restructuring, and AI was part of the reason. I had a choice: compete for a shrinking pool of traditional roles, or invest in understanding what was actually happening. I chose the latter, spending over a year researching the intersection of human cognition, AI architecture, and evolutionary biology. Not because I had a comfortable runway, but because I believed the only durable response to AI displacement was to understand the technology deeply enough to work with it rather than against it.What struck me most during that period wasn’t the technology itself. It was the absence of any coherent framework to help people navigate the transition. The federal response, to this day, is plans about plans. The White House AI Action Plan, released in July 2025, proposes an “AI Workforce Research Hub” and “rapid retraining initiatives,” but these remain proposals, not funded programs. A mandated plan to reach one million new apprentices was due by late August 2025; as of this writing, no such plan has been published. The existing federal training infrastructure, primarily the Workforce Innovation and Opportunity Act, was designed for trade-related displacement and factory closures, not for the continuous, distributed, invisible erosion of knowledge work.The geographic dimension compounds the problem. AI jobs concentrate in a handful of metros. Displacement is distributed broadly. It’s not hard to picture: a mid-sized city that lost its call center to automation, watched entry-level coding jobs evaporate through attrition, and offered tax incentives for a data center that consumed grid capacity, employed a few dozen people, and may not outlast its tax abatement. That city exists in every state. This mirrors the manufacturing hollowing-out pattern, except it moves faster and affects white-collar workers who assumed they were insulated.The Governance DeficitThe economic consequences are serious. The political consequences may be worse.Here’s what doesn’t get discussed enough: the infrastructure being built for the AI boom doesn’t disappear when valuations correct. After the dot-com bust, the fiber optic cables that bankrupt companies had laid were bought for pennies on the dollar and became the backbone of the modern internet. That was mostly benign. The AI equivalent is massive behavioral datasets, surveillance capabilities, and algorithmic decision-making systems. When the correction comes, these assets don’t evaporate. They get acquired at fire-sale prices by buyers whose purposes may have nothing to do with the productivity narrative that justified building them in the first place.When populations lose economic leverage through job displacement while simultaneously living under systems designed to monitor and optimize them, the conditions for democratic governance weaken. This isn’t dystopian speculation. It’s the observable pattern in every context where economic marginalization and pervasive surveillance overlap. The question isn’t whether these tools will be misused; it’s whether the institutional safeguards exist to prevent it.They largely don’t. The United States has no comprehensive federal AI law. The Trump administration revoked Biden-era AI safety requirements on its first day in office and has since signed an executive order directing the Attorney General to establish an AI Litigation Task Force to challenge state-level AI regulations deemed “inconsistent” with federal policy. Over 1,000 AI-related bills were introduced across states in 2025, reflecting genuine demand for oversight. The federal response has been to threaten litigation against the states attempting to fill the vacuum while offering nothing comparable in return.The net effect is a moral hazard: the companies whose concentration of power most warrants oversight are shielded from local accountability, while the communities bearing the grid strain, the energy costs, and the employment shortfalls have no federal framework to fall back on.This is where the augmentation framework from the first piece in this series becomes more than an architectural preference. Systems designed to augment human decision-making preserve human agency by design. They keep humans in the loop not as a nicety but as a structural feature. Systems designed to replace human decision-making don’t. The governance question isn’t just “how do we regulate AI?” It’s whether we’re building systems that are compatible with democratic self-governance at all.The Concentration ProblemWhether the AI boom continues or corrects, the infrastructure is concentrating in very few hands. Microsoft, Google, Amazon, and a handful of others control the foundational compute layer. This is the equivalent of three companies owning the roads, the ports, and the electrical grid simultaneously.The most promising counterforce is open-source AI. As I noted in the second piece, open-source models are closing the performance gap with closed-source leaders, and enterprises are increasingly drawn to smaller, specialized, fine-tuned models. But open-source AI depends on compute access, which circles back to the same hyperscalers. Distributed architecture requires distributed infrastructure. Without it, “open” is a label, not a reality.The CHIPS Act and semiconductor export controls, discussed in the second piece, are framed as competitive tools against China, not as instruments for domestic market structure. The United States is subsidizing concentration while calling it competition. The alternative, public investment in open compute infrastructure, antitrust enforcement that addresses AI-specific concentration, data portability requirements that prevent lock-in, remains largely unexamined in policy circles.Meanwhile, China isn’t pursuing a single scaling approach. It’s diversifying architecturally while maintaining state oversight of strategic technology. Europe can’t compete in the capital-intensive scaling race but could lead in alternative architectures and governance frameworks. The United States is currently locked into the scaling paradigm by the very corporate interests that benefit most from it.What Augmentation Actually RequiresThis series began with a technical argument: the scaling skeptics are right, and Augmented Human Intelligence offers a more promising architectural path than the brute-force pursuit of AGI. The second piece showed that the financial structure built on the AGI thesis is fragile and concentrated. This piece has argued that the human consequences of getting this wrong fall disproportionately on people who had no say in the bet.Those three threads converge on a single point. AHI isn’t just a better technical framework. It’s the only path that keeps humans economically relevant, politically empowered, and socially integrated in an age of increasingly capable machines. But it won’t happen by default. The economic incentives, the capital flows, and the policy environment all currently favor concentration, replacement, and scale.What augmentation actually requires is deliberate choice at every level. Architectural choices: modular, specialized, human-in-the-loop systems rather than monolithic models chasing general intelligence. Policy choices: antitrust enforcement, public compute infrastructure, workforce transition programs that match the scale of the disruption. Ownership choices: open-source development, distributed access, accountability structures that prevent the technology from concentrating power in the hands of those who build it.None of these are inevitable. All of them are possible.I mentioned earlier that I was part of a reduction in force in September 2023. What I didn’t say is where that led. Rather than compete for a shrinking pool of roles in an industry being restructured around me, I spent over a year building exactly the kind of human-AI collaboration I’ve been advocating for. A system called MARS (Multi-Agent Research System) that monitors AI-related content, scores it against my research corpus, and generates strategic engagement opportunities. This entire series, the research synthesis, the financial analysis, the policy arguments, was produced through that system. I used AI tools at every stage of retrieval, drafting, fact-checking, and revision. But the judgment, the framing, the argument, the decision about what matters and why: that’s mine. The AI handled pattern recognition and tireless iteration. I handled meaning-making.I’m not offering a theoretical framework. I’m offering evidence. This is what augmentation looks like in practice, and it’s available to anyone willing to invest the time in understanding these tools deeply enough to direct them rather than be directed by them.The orchestra can keep expanding. Or we can decide what music we actually want to hear, and who gets to play.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

This is the second in a three-part series examining the cracks in the AI scaling narrative, from the technical limits, to the financial fragility, to who bears the cost when the correction comes.Michael Burry, the investor made famous by The Big Short for calling the 2008 housing crash, is shorting Nvidia and Palantir. He might be early. He might be wrong. But the fact that serious money is now betting against the AI boom tells us something worth examining.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.This isn’t a crash prediction. It’s a structural analysis. And the structure, when you look at it clearly, is more fragile than the headlines suggest.The Bet That Built This BoomThe AI investment thesis rests on a chain of assumptions: scaling works, AGI is near, winner-take-all dynamics will reward the leaders, and therefore almost any level of investment is justified.Hundreds of billions have been deployed on this logic. OpenAI’s valuation assumes it will capture a significant share of all future knowledge work. Nvidia’s market cap assumes AI infrastructure spending will continue accelerating for years. The hyperscalers (Microsoft, Google, Amazon, Meta) are spending at rates that only make sense if AI transforms their core businesses.The bet is that training costs are front-loaded and inference revenue will follow. Build the models now, monetize them later. It’s a reasonable bet. But it is a bet.The bet is also being reinforced from outside the market. The CHIPS Act, export controls on advanced semiconductors to China, and bipartisan rhetoric framing AI as a strategic national asset have created a policy environment that backstops spending regardless of near-term ROI. When national security and economic competition merge with a technology thesis, capital flows become stickier and harder to redirect, even when the underlying assumptions weaken.The Cracks in the ThesisThe scaling hypothesis, the idea that more data and compute reliably yield smarter models, is under pressure. As I explored in my [previous post on the scaling skeptics], serious researchers are now questioning whether current architectures can reach general intelligence at any scale. Diminishing returns have arrived faster than expected.Meanwhile, enterprise ROI isn’t materializing the way vendors promised. MIT’s 2025 “GenAI Divide” study found that 95% of enterprise AI pilots deliver zero measurable impact on the bottom line; only 5% reach production. S&P Global reported that 42% of companies scrapped most of their AI initiatives in 2025, up from just 17% the year before. The pattern is familiar to anyone who’s watched technology adoption cycles: impressive demos, difficult deployment, unclear value.The revenue picture is also lopsided. Nvidia is selling shovels in a gold rush, and making historic profits doing it. But who’s finding gold? OpenAI reportedly lost $5 billion in 2024, and the trajectory is worsening: Microsoft's SEC filings reveal that OpenAI lost approximately $12 billion in a single quarter in 2025, against roughly $4.3 billion in revenue for the entire first half of the year. Most AI startups are burning capital with no clear path to profitability. The infrastructure providers are thriving; the companies building on that infrastructure mostly aren’t.There’s a persistent gap between what AI can do in a demo and what it can do in production. Benchmarks improve; reliability doesn’t. Hallucinations persist. Enterprise customers discover that “90% accurate” isn’t good enough for mission-critical workflows.There’s also a competitive assumption baked into these valuations that may not hold. The investment thesis requires winner-take-all dynamics, but the market is showing signs of fragmentation. Open-source models are closing the performance gap with closed-source leaders. Inference costs are dropping fast. Enterprises are increasingly drawn to smaller, specialized, fine-tuned models rather than monolithic general-purpose ones. If the market fragments rather than concentrates, valuations built on “capturing all future knowledge work” start to look very different.The technical skeptics I profiled in last week’s piece aren’t just academics arguing about architecture. Their critiques have financial consequences. If LeCun and Chollet are right that brute-force scaling won’t yield general intelligence, then the entire investment thesis, which requires something approaching AGI to justify current spending, rests on a technical error. The market is pricing in a future that may be architecturally impossible.The Structural FragilityZoom out and the concentration risk becomes visible.The Magnificent Seven tech companies increased their energy consumption by 19% in 2023 while the median S&P 500 company’s consumption stayed flat. Roughly 80% of U.S. stock market gains in 2025 were tied to AI-related companies. That’s not a broad-based technology boom; it’s a narrow bet by the entire market on a single thesis.JPMorgan estimates that AI-related investment-grade bond issuances could reach $1.5 trillion by 2030. Much of this debt is predicated on productivity gains that may or may not materialize. If the gains don’t come, the debt doesn’t disappear. And the infrastructure itself has hard physical limits: gas turbines require three to four year lead times, and new nuclear capacity takes a decade or more. Capital can move fast; power plants can’t.The question isn’t whether AI is useful; it clearly is, in specific applications. The question is whether the valuations, the infrastructure spending, and the debt levels are proportionate to the actual value being created. Right now, a lot of capital is chasing returns that require the scaling hypothesis to hold, enterprise adoption to accelerate, and competitive moats to emerge. All three are uncertain.Historical Parallels (And Their Limits)We’ve seen this pattern before.The dot-com boom left behind real infrastructure: fiber optic cables, data centers, a generation of internet-native companies. Most of the companies that raised money during the bubble failed, but the technology was real and eventually transformed the economy.The crypto boom left behind less. Blockchain has use cases, but the speculative frenzy produced more fraud than lasting value.AI will probably land somewhere between. The technology trajectory resembles the dot-com era: real and ultimately transformative. But the valuation structure currently looks more like crypto: speculative and decoupled from cash flow. The technology is real. Transformer architectures, large language models, diffusion models; these are genuine innovations with genuine applications. But genuine innovation doesn’t guarantee that current valuations are rational, that current market leaders will dominate long-term, or that the timeline to profitability is what investors are pricing in.Useful and overhyped are not mutually exclusive.What I’m WatchingA few signals matter more than headlines:Enterprise adoption vs. churn. Are companies moving from pilots to production, or quietly shelving experiments? Renewal rates will tell the real story.Hyperscaler capex. When Microsoft, Google, or Amazon start trimming AI infrastructure spending, the narrative will shift fast. They have better visibility into actual demand than anyone.Regulatory attention. As valuations and market concentration grow, scrutiny from antitrust bodies (FTC, DOJ, EU) or financial regulators (SEC) could significantly impact the narrative and business models.The language shift. Listen for when “AGI in two years” becomes “useful tools that augment workflows.” The rhetoric is a leading indicator. When the people selling the dream start hedging, pay attention.I’m not predicting a crash date. Markets can stay irrational longer than skeptics can stay solvent; Burry himself has learned that lesson more than once.But I’ve spent 25 years watching technology hype cycles, and I know what structural fragility looks like. This is it. A narrow set of companies, a contested thesis, valuations that require optimistic assumptions to hold, and a growing chorus of informed skeptics.The music might keep playing. But it’s worth understanding the warning signs, and where the exits are.Next week: Financial fragility makes headlines, but the deeper question is who bears the cost when the correction comes. In “After the Music Stops,” I look beyond investor losses to the workers, communities, and democratic institutions caught in the undertow of a technology transition that was never designed with them in mind.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

This is the first in a three-part series examining the cracks in the AI scaling narrative, from the technical limits, to the financial fragility, to who bears the cost when the correction comes.The dominant AI narrative is seductively simple: scale is all you need. More data, more compute, more parameters, and intelligence emerges. It’s the story that’s justified hundreds of billions in investment, reshuffled the power structure of Silicon Valley, and convinced a generation of executives that AGI is just a few training runs away.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.But a growing chorus of serious people disagree. Not AI skeptics or Luddites, but pioneers, Turing Award winners, people who built the foundations of what we’re now scaling. Their critiques are converging on an important point. And what they’re circling around might matter more than any of them are quite saying.Yann LeCun: World Models, Not Word ModelsLeCun, a Turing Award laureate and former chief AI scientist at Meta, left in November 2025 to start AMI Labs, valued at $3.5 billion. His thesis: LLMs are “an off-ramp on the road to human-level AI.”His argument is architectural. LLMs predict text; they model language, not reality. They can pass the bar exam while lacking the common sense of a house cat. They hallucinate because they have no grounding in how the world actually works.What’s needed, LeCun argues, are “world models”: systems that build internal representations of reality, that can simulate, predict, and plan. Not language models but cognitive models. His advice to PhD students:“LLMs are useful, but they are an off-ramp on the road to human-level AI. If you are a PhD student, don’t work on LLMs.”Gary Marcus: Pattern-Matching Is Not ReasoningMarcus, a cognitive scientist, NYU professor emeritus, and author of Taming Silicon Valley, has been the most persistent public critic of the scaling hypothesis. He’s also been the most vindicated.His core claim: LLMs are sophisticated pattern-matchers, not reasoners. They interpolate within their training distribution but fail outside it. Change a classic reasoning problem slightly, reword it, alter the setup, and performance collapses. This isn’t a bug to be fixed with more data. It’s architectural.Hallucinations, in Marcus’s view, are unfixable within the current paradigm. And the stakes are real: in medicine, law, and finance, even a 1% hallucination rate can be catastrophic. Systems that traffic only in the statistics of language, without explicit representations of facts or tools to reason over them, will never be reliable. The solution isn’t more scale; it’s hybrid architectures that combine neural networks with symbolic reasoning. The kind of thing AI abandoned decades ago in the rush toward end-to-end learning.Rodney Brooks: The Competence Projection ProblemBrooks, an MIT roboticist, founder of iRobot and Robust.ai, and builder of more humanoid robots than perhaps anyone alive, brings the engineer’s skepticism.His critique is partly about us, not just the technology. When humans see an AI perform a task impressively, we instinctively generalize. We project a sphere of competence around that performance, assuming the system understands in a human-like way. We’re almost always wrong.LLMs are useful, Brooks says, but narrow. The problem with robotics isn’t language interfaces; it’s control theory, optimization, the hard math of operating in physical reality. Language won’t help you ship 10,000 warehouse orders in two hours; data processing and planning will.His broader point: 76% of AI researchers surveyed by the AAAI agree that scaling current approaches alone won’t yield AGI (AAAI 2025 Presidential Panel). The insiders know. The narrative hasn’t caught up.François Chollet: Fluid Intelligence and the ARC TestChollet, creator of Keras and now running the ARC Prize Foundation, offers the most precise diagnosis.He distinguishes between crystallized intelligence (accumulated knowledge, memorized patterns) and fluid intelligence (the ability to adapt to genuinely novel problems). LLMs excel at the former. They’re vast repositories of crystallized knowledge, able to retrieve and recombine what they’ve seen. But fluid intelligence, the thing that lets a child solve a puzzle they’ve never encountered, is largely absent. As Chollet puts it:“Memorization is useful, but intelligence is something else.”His ARC benchmark tests exactly this: abstract reasoning tasks are trivial for humans, but brutal for LLMs. GPT-4o scores around 5%. Humans score 84%. The gap isn’t closing with scale.Chollet’s prescription: program synthesis. Systems that can construct new cognitive programs on the fly, recombining primitives to handle novel situations. “Memorize, fetch, apply” won’t get us there. We need systems that can genuinely invent.Where They ConvergeThese critics come from different traditions (cognitive science, robotics, deep learning research), but their diagnoses are converging:* Brute-force pre-training scaling is hitting diminishing returns. The spectacular gains of 2020–2023 have flattened. Each new model is marginally better at an enormous cost.* Current benchmarks mask fundamental limitations. High scores on contaminated tests don’t equal general intelligence.* World models and grounding are essential. Systems need to represent reality, not just language about reality.* Architectural innovation is required. More of the same won’t work. Something structurally different is needed.There’s also an uncomfortable economic dimension here. The scaling narrative isn’t just a technical hypothesis; it’s the justification for hundreds of billions in concentrated infrastructure investment. When the thesis and the money are this entangled, technical course corrections become financially painful. That structural fragility is worth examining on its own terms, and it’s where I’m headed next week.Where They DivergeThe agreement ends at prescription. LeCun wants embodied world models trained on sensory experience. Marcus wants symbolic-neural hybrids that can reason explicitly. Brooks wants constrained domains with realistic expectations. Chollet wants program synthesis and test-time adaptation.These aren’t necessarily incompatible; they might all be part of whatever comes next. But there’s no unified alternative vision yet, just a shared sense that the current path is incomplete.What They’re Not Quite SayingHere’s what strikes me: the skeptics are all still focused on building artificial general intelligence. Better AI. Smarter machines. Systems that can eventually match or exceed human cognition.But what if the frame itself is wrong?What if the goal isn’t to replace human intelligence but to augment it? Not AGI but AHI: Augmented Human Intelligence. If LeCun is right that LLMs lack world models, and Chollet is right that they lack fluid intelligence, then their best use isn’t to replace the human driver but to serve as a high-powered navigation system. Modular AI architectures are designed to amplify human capability rather than substitute for it. Systems that handle what they’re good at (retrieval, pattern recognition, tireless execution) while humans handle what we’re good at (judgment, meaning-making, genuine novelty).The scaling skeptics have correctly diagnosed the disease: current approaches won’t yield general intelligence. But they’re still chasing the same endpoint. The alternative I’ve been exploring in this series is different: not smarter machines, but smarter human-machine partnerships. Architectures inspired by how the brain actually works: modular, specialized, integrated through attention rather than unified in a single massive model. [Link to earlier AHI posts]This doesn’t mean AHI is without its own hard questions. Who gets augmented? Who controls the tools? Any architecture that amplifies human capability will amplify it unevenly unless we’re deliberate about access and governance. But those are design problems, solvable ones, not the fundamental architectural dead-end that monolithic scaling represents.This is precisely the kind of architectural shift the critics call for. Their critiques of monolithic scaling point directly toward modular, human-centered design. The pieces are all there. They just haven’t been assembled this way.The scaling-is-everything narrative is cracking. What comes next is genuinely open. That’s worth paying attention to, and, for those of us thinking about it, worth shaping.Next week: If the scaling thesis is cracking, what happens to the hundreds of billions riding on it? In “When the Music Stops,” I examine the structural fragility of the AI boom, from concentrated market bets and ballooning infrastructure debt to the gap between demo magic and enterprise reality. The technology is real. The question is whether the valuations are.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

Last week, I admitted I’m part of the problem. Despite advocating restraint, I’ve felt the pull to post more, feed the algorithm, and add to the noise. We’re all drowning in a flood we helped create.I ended with a question: Could the same tools flooding us with content also help us navigate it?Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.This post is my answer.The Needle and the HaystackWe live in a world of infinite language.No human can process it all, nor should they. Most of it is, at best, meaningless to any individual. At worst, it’s noise, pollution, or manipulation clogging the channels where meaning once flowed.I talk about disciplining attention by being deliberate. But it’s hard. Every day, the same question: In this endless sea of language, where should I focus? How do I find the needles in the haystack?For months, I scrolled and skimmed, hoping algorithms would surface something worthwhile. Occasionally, they did. Mostly, they didn’t.Then I realized: I was asking the wrong question.What LLMs Are Actually Good AtI’ve argued that LLMs won’t achieve AGI. They don’t reason like humans. They don’t truly understand.But here’s what they are good at: processing language. Pattern matching. Summarizing. Filtering. Categorizing.These aren’t the glamorous capabilities that dominate AI headlines. No one writes breathless articles about “AI that sorts your reading list.” But for the problem I faced (too much language, too little attention), these are exactly the tools I needed.This is the AHI thesis in practice: use AI to extend human capabilities, not replace human judgment. Let the machine do what machines do well (process volume) so the human can do what humans do well (decide what matters).Building a Filter, Not a FirehoseHere’s what I did.I wanted to engage with writers whose work intersects with mine: AI, attention, cognition, architecture. But finding them manually, keeping up with their output, and identifying meaningful conversations: that was a full-time job.So I built a simple system. Nothing fancy, but deliberate.First, I curated a list of authors. Not everyone writing about AI, only those whose thinking genuinely aligns with mine, whose audiences might value my perspective. This required human judgment. No algorithm could tell me who I respect.Second, I built a corpus of my own work. My Substack posts, whitepapers, and core ideas. This corpus gave the system a reference point: What does my thinking sound like? What themes do I care about?Third, I created a scoring mechanism. When a new post from someone on my list appears, the system evaluates its relevance to my corpus. How much overlap? How timely? The output isn’t a binary “read or skip.” It’s a score from 0 to 100, with a ‘worth-a-look’ threshold currently set at 90 or higher: any article that reaches or exceeds it is identified as a potential engagement opportunity and therefore worthy of review and potentially posting a comment.The system doesn’t tell me what to think. It doesn’t write my responses (although it can certainly offer potential starting points - more on this later). It just filters the infinite down to the manageable so that I can do the actual intellectual work.I’m using the very thing that caused the flood to build a personal lifeboat.And that’s precisely the point.How I Built My Filter (A Practical Pattern)If you’re curious about the mechanics, not as a prescription, but as a pattern, here are the steps I followed:Define your “needle.” I started by clarifying what I was actually looking for: writers exploring the intersection of AI, cognition, and meaning. Not “everything about AI,” but my slice of it. Your needle will be different.Gather your own voice. I fed my system a corpus of my own writing: Substack posts, white papers, notes, and key ideas. This corpus provided a reference point for relevance: “Find things that resonate with this.” Your corpus is your intellectual fingerprint.Curate, don’t collect. I handpicked a starter list of thinkers whose work I genuinely respect. No algorithms. The starter list is a human gate, built once, then scaled. Your list should feel like a dinner party invitation list, not a block party.Score for signal, not popularity. My system ranks new content by thematic overlap, not engagement metrics. It asks: “How much does this align with what you care about?” Your scoring should reflect your priorities, not the platform’s.Build an interface you’ll actually use. Raw output from a script isn’t enough. I built a simple web interface to review, sort, and manage the authors list, the scoring results, and record actual engagements (comments that I post). If you can’t easily see and interact with the insights, you won’t use the system. Make it visible. Make it yours.Preserve the human moment. The output isn’t a command. It’s a filtered feed. I still choose what to read, how to engage, and whether to respond. Your judgment stays in the loop.Iterate relentlessly. My first pass wasn’t right. I’ve refined my author list, adjusted my scoring weights, and improved the user interface as I learned what I actually needed. The system that I have built isn’t a finished product. It’s a living system that evolves as my thinking does. Your filter should grow with you.The tools are simple: RSS feeds, a language model API, a basic script, and a lightweight web app. The philosophy is what matters: AI as a lens, not a source.The Discipline Still RequiredHere’s what AI can’t do for me:It can’t decide what I actually care about. It can’t distinguish deep insight from clickbait in intellectual clothing. It can’t write a thoughtful response that truly reflects my voice, my personal experience, and my ever-evolving point of view. It can’t build real relationships with other thinkers.All of that remains mine.What AI can do is reduce the search cost. It helps me spend less time looking and more time thinking. It’s a filter, not a replacement for judgment.It can suggest potential starting points for me to consider, or more often than not, not consider. My process is:* Read the articles that exceed my scoring threshold (90/100)* Make my own notes about the thoughts that the article evokes in me, and parallels to the work that I have produced previously* View any system-generated starting points, and consider how, if at all, they might echo my personal gut reactions* Work to develop a final comment that considers all inputs, while ensuring the article is my work product and that I reference my own point of view and personal experience (just as I am doing right here).* My experience has been that more than 85% of the time, my own notes are the only input to the final comment. This is critical for me, as my corpus update process incorporates all of my comments as well as new articles and whitepapers written since the last update. I have also observed occasional hallucinations in the ‘target passage’ identification part of the system generated suggested starting points.This process is constraint-awareness applied to tools: knowing what to delegate and what to keep. Using AI to extend attention, not abdicate it.An Invitation to ExperimentI haven’t solved the attention crisis. I’ve built one small tool for one specific problem. It helps me. It might not help you.But the principle generalizes.If you’re drowning in content, ask yourself: What’s my needle? What’s my haystack? What would it look like to use AI as a filter rather than a generator? Where could pattern-matching reclaim time for deeper thinking?The flood won’t stop. The language will keep coming. But we’re not helpless. The same technologies that are accelerating the crisis might, if used carefully, help us navigate it.Not by processing more, but by attending better.This is the fifth essay in my series on language, attention, and the architecture of shared meaning. Previous posts: [Part I: The Constraints We Lost] • [Part II: Serviceability Failure] • [Part III: Wisdom in an Infinite-Language World] • [Part IV: We Are the Problem]Next week in ‘More Instruments, Same Song’ I will summarize the growing number of respected AI experts that question the ‘Bigger is Better’ mantra for models, compare my critiques to these experts’, and share our observations about what might drive the next breakthroughs in the field of AI. Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

I’ve just finished a three-part series arguing that wisdom, in our current moment, means constraint-awareness. One of the core practices I advocate is generative restraint: resisting the impulse to add to the noise, valuing thought before speech, and recognizing that not every reaction needs to be published.And yet.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.Over the past three months, I’ve been building a publishing cadence. Website, Substack, LinkedIn, X. The goal was one substantial essay per week, reasonable, intentional, aligned with the very restraint I was preaching.But on multiple occasions, I’ve felt the pull to do more. To post more frequently. To engage more visibly. To hit LinkedIn’s little recommendation boxes for commenting, reacting, and connecting.Why? Because I can. Because everyone else seems to be doing it. Because the platform actively encourages it, nudging me toward behaviors that serve its interests (more engagement, more time on site, more data) rather than mine.A few days ago, I saw a post from someone explaining why they’re leaving LinkedIn. The reason: the signal-to-noise ratio has collapsed. The feed is now synthetic content, engagement farming, posts optimized for clicks rather than insight.Reading it, I recognized the problem immediately.Then I recognized myself as part of it.The Systemic TrapThis is the thing about systemic problems: they’re made of us.Every individual making a locally rational choice (post more, engage more, feed the algorithm) produces a collectively irrational outcome. We’re all contributing to a flood that’s drowning us all.You can understand the dynamic completely and still be caught in it. That’s what makes it a trap.A few weeks ago, I reached out to Gary Marcus, one of the sharpest critics of AI hype, to ask if he might review something I was working on. He graciously replied that he was sorry, but overwhelmed.When I asked if he could recommend someone else, he said something that’s stayed with me:“We are all overwhelmed.”Think about that.Here is someone who understands the attention economy and AI-generated content better than almost anyone. Someone who has built a career diagnosing these very dynamics.And he’s telling me that he and his peers are drowning.If the experts who study this problem are themselves overwhelmed by it, that’s not irony. That’s the signature of a genuine systemic trap.The Gap Between Knowing and DoingIn my series, I argued that the collapse of external constraints (throughput, gatekeepers, locality, friction) has left us navigating an infinite-language world with finite attention.The algorithm isn’t forcing me to post. It’s nudging me. And I’ve been following the nudge, even while writing about the importance of not following it.The Greeks had a word for this gap between knowing and doing: akrasia, the act of acting against your own better judgment. It’s not hypocrisy. It’s the human condition when incentives and wisdom point in opposite directions.The platform rewards volume. Wisdom counsels restraint. And the platform is right there, every day, with its little dopamine hits and recommendation checklists, while wisdom is quiet and offers no metrics.A Small Commitment in a Large SystemI don’t have a complete solution to the collective action problem. I can’t fix the algorithm or convince everyone to post less.But I’m starting to wonder whether the same tools flooding us with content might help us navigate the flood, whether there’s a way to fight fire with fire, carefully. I’ll have more to say about that soon.For now, I can decide on my own contribution.So here’s my commitment, stated publicly so I’m accountable to it:One Substack essay per week, if I have something worth saying. If I don’t, I skip the week. No posting for the sake of posting. No engagement farming. No, letting the platform’s goals become my goals.This is modest. It won’t change the system. But constraint-awareness has to start somewhere, and it might as well start with the person advocating for it.We are the problem. Which means we’re also, in small ways, the beginning of any solution.An InvitationI urge you to consider your own relationship to the feed. Are you posting because you have something to contribute, or because the algorithm told you to? Are you adding signal, or noise?The attention commons is finite. Every one of us is either stewarding it or depleting it.Choose carefully.And maybe, just maybe, join me in practicing a little generative restraint. Not as a boycott, but as a quiet act of repair.Because the flood won’t stop until enough of us decide to build our own dams.This essay stands alongside my three-part series on language, constraints, and the crisis of shared meaning: [Part I: The Constraints We Lost] • [Part II: Serviceability Failure] • [Part III: Wisdom in an Infinite-Language World]Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

This is Part III of a three-part series. Part I: The Constraints We Lost, Part II: Serviceability FailureThe instinctive response to a system in crisis is to restore what was lost.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.If the problem is that gatekeepers are gone, rebuild the gates. If the problem is that anyone can publish anything, create centralized fact-checking regulators. If the problem is that we’re drowning in noise, retreat to informational bunkers where only trusted sources are allowed.These responses are understandable. They’re also insufficient. They attempt to rebuild walls in an open field, to reimpose external constraints on a system that has already evolved beyond them. The gates cannot be closed. The friction cannot be restored by fiat. The global, instantaneous, algorithmically-driven flow of language is not a temporary condition we can reverse; it’s the environment we now inhabit.So what do we do?The answer, I think, lies in a shift of locus. If external constraints can no longer stabilize the linguistic interface, then the stabilizing function must move inward: from the environment to the mind, from architecture to practice, from imposed limits to cultivated awareness.The word for this is wisdom. But we need to define it carefully.Wisdom RedefinedWisdom has a mystical reputation it doesn’t deserve. We tend to imagine it as something possessed by sages on mountaintops, accumulated through decades of suffering, or granted by spiritual insight. This framing makes it feel inaccessible, perhaps even irrelevant to the practical challenges of navigating a noisy information environment.But there’s another way to understand wisdom: as constraint-awareness.Constraint-awareness is the capacity to recognize the conditions that shape your own perceptions, beliefs, and communications. It involves three interlocking recognitions:First, recognizing that language is an interface, not a window. What you read, hear, and speak is not unmediated reality; it’s a translation, shaped by perspective, optimized for social function, always partial. This is accepting that every statement, including your own, comes from “a certain point of view.”Second, mapping what’s missing. The linguistic environment we inhabit is historically unprecedented. The constraints that shaped discourse for millennia (limited throughput, gatekeeping bottlenecks, local accountability, production friction) have been removed. Understanding this helps explain why the information environment feels so disorienting: it’s not that people have become worse; it’s that the stabilizing structures have vanished.Third, voluntarily simulating constraints. This is the practical core. If external constraints no longer impose limits, we must impose them ourselves, not to retreat from the modern world but to navigate it with agency and clarity.This reframing clarifies an ancient distinction. Knowledge is the accumulation of facts and models; LLMs scale it infinitely. Intelligence is pattern recognition and inferential speed; LLMs simulate it convincingly. Wisdom is the regulation of knowledge and intelligence in light of their limits. It asks: given that my interface is partial, my information is infinite, and my time is finite, how should I direct my attention and shape my judgments?Wisdom is not anti-technology. It’s the skill that makes technology usable without being overwhelmed by it.Four PracticesConstraint-awareness isn’t a passive trait. It’s cultivated through practice. In an infinite-language world, these practices become essential disciplines, each one a voluntary replacement for a constraint that’s been lost.Attentional Friction (replacing throughput)The old throughput constraint meant that information arrived slowly, giving you time to digest it. That constraint is gone. The replacement is deliberate slowness: choosing depth over breadth, engaging with long-form arguments rather than skimming headlines, creating periods of informational silence.This isn’t about consuming less (though it might involve that). It’s about consuming differently. It means finishing the article before forming an opinion. It means sitting with uncertainty rather than reaching for the first confident take. It means treating your attention as a finite resource that deserves protection.The practice: build friction back into your information diet. Turn off notifications. Schedule time for deep reading. Resist the pull of the feed.Deliberate Curation (replacing bottlenecks)The old bottleneck constraint meant that gatekeepers filtered what reached you. Those gatekeepers were imperfect, often biased, and sometimes corrupt. But they performed a curation function that has now been handed to algorithms optimized for engagement, not understanding.The replacement is choosing your own curators: scholars, journalists, thinkers, and artists whose judgment and values you respect. Not because they’re always right, but because their filtering reflects human discernment rather than algorithmic amplification.This requires active maintenance. Audit your information sources periodically. Ask: Who am I allowing to shape my attention? Are they chosen, or did they arrive through the path of least resistance?The practice: build a personal cabinet of curators. Diversify it deliberately. Return to it when the noise becomes overwhelming.Locality Building (replacing locality)The old locality constraint meant that communication was embedded in ongoing relationships. If you said something false or harmful, you faced your community the next day. Accountability was built into the structure.That accountability has dissolved into global, often anonymous networks. The replacement is intentional investment in bounded communities (physical or digital) where ideas can be tested, challenged, and refined over time, and where participants have reputations to maintain.This doesn’t mean retreating to echo chambers. It means finding spaces where disagreement is possible but accountable, where you’re known well enough that your words carry weight and consequence.The practice: invest in communities where you’re a participant, not just a consumer. Prioritize spaces that reward thoughtfulness over virality.Generative Restraint (replacing friction)The old friction constraint meant that producing and distributing language required effort. That effort served as a natural filter: if you were going to speak publicly, you probably had something to say.With friction eliminated (and LLMs reducing the cognitive cost of generation to zero), the replacement is voluntary restraint. This means resisting the impulse to add to the noise, valuing thought before speech, and recognizing that not every reaction needs to be expressed.When using LLMs, this means treating them as tools for exploration and questioning rather than as oracles. It means maintaining your own judgment rather than outsourcing it to fluent-sounding output.The practice: before speaking (or posting, or generating), ask whether this adds signal or noise. Develop comfort with silence.The Social DimensionIndividual wisdom, while necessary, is insufficient. The serviceability failure of the linguistic interface is systemic. Its repair requires not just wiser individuals but cultural norms and institutions that promote constraint-awareness at scale.This suggests some design principles for a healthier information architecture:Platforms that reward deliberation over virality. The current incentive structure optimizes for engagement, often at the expense of outrage, conflict, and tribal signaling. Different designs are possible. They would be less profitable under current models, but less destructive.Educational models that teach epistemic humility alongside factual content. Knowing facts is not enough. Students need to understand how knowledge is constructed, how narratives are shaped, and how their own cognition can be manipulated. Source literacy, media ecology, and the history of propaganda should be core curriculum, not electives.Interfaces that reveal uncertainty rather than hiding it. Current information systems present polished, confident outputs. A constraint-aware design would show process, surface disagreement, and make the partiality of any perspective visible rather than concealed.The goal is not to make the world simple again. It’s to build tools and norms that help us be thoughtful in a world of overwhelming complexity.The Ultimate ConstraintBeneath all the constraints we’ve discussed lies one that cannot be removed: human attention.Attention is finite. It is the bottleneck through which all experience must pass. No technology can expand it; technology can only compete for it.Constraint-awareness, at its core, is the stewardship of attention. It’s the recognition that every click, every minute of reading, every engagement is a vote for the kind of reality we’re constructing, both in our own minds and in the collective intersubjective space.This is where the structural argument becomes personal. The crisis of language is, in the end, a crisis of attention. The question “what should I believe?” is downstream of the question “what should I attend to?” And that question is answered not once, in some grand philosophical moment, but thousands of times a day, in small choices that compound.Wisdom, therefore, becomes the practice of allocating attention in ways that repair rather than fracture the shared interface. It’s the application of finite consciousness to infinite language with discernment, care, and recognition of profound limits.Not Retreat, But AscentI want to be clear about what I’m not arguing.This is not a call to abandon technology, retreat to a simpler time, or pretend the constraints can be restored. The infinite-language world is here. LLMs are not going away. The flat landscape of frictionless generation is the terrain we must navigate.What I’m arguing is that navigating this terrain requires a new kind of maturity. It asks us to become architects of our own attention and stewards of our shared epistemic commons. It asks us to do consciously and deliberately what the environment once did for us automatically.This is not a retreat. It’s an ascent to a higher level of cognitive responsibility.The most important design project of the coming century may not be a new AI model. It may be a new human-information interface: one built not for infinite engagement but for meaningful understanding, not for frictionless flow but for thoughtful integration.The constraints are gone. Our awareness must now take its place.The question we’re left with is simple to state and difficult to answer: Can a species that used language to build civilization now learn to use wisdom to preserve it?I don’t know. But I think the attempt is worth making, one attentive choice at a time.This is the final essay in a three-part series on language, constraints, and the crisis of shared meaning. Next week: a confession. I've been writing about generative restraint while feeling the platform's pull to do the opposite. What happens when you recognize yourself as part of the problem you're diagnosing?Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

This is Part II of a three-part series. Part I: The Constraints We LostIn structural engineering, two of the ways a building can fail are instructive.The dramatic one is collapse: the structure exceeds its load capacity and comes down. That’s what we fear, and that’s what building codes are designed to prevent.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.The subtler one is serviceability failure. The building stands. It passes inspection. But it no longer works as intended. At 432 Park Avenue, this means residents who feel seasick in their living rooms, elevators that won’t align with floors during storms, and a constant low groan that makes sleep impossible. The structure is safe. It’s just not livable.Our linguistic infrastructure (the system of shared language through which we build trust, coordinate action, and construct common reality) is experiencing serviceability failure. It hasn’t collapsed. Words still work. People still communicate. But the system’s core function, enabling large groups of humans to agree on what’s true, what matters, and what to do about it, is degrading in ways that are hard to see and harder to fix.The symptoms are everywhere. We just haven’t named them properly.Symptom 1: The Fracturing of Shared RealityThe deepest symptom is the erosion of what philosophers call intersubjective reality: the realm of shared beliefs, norms, and stories that exists only because enough people agree to believe in it. Money lives here. So do laws, democratic legitimacy, and scientific consensus. These aren’t physical objects; they’re collective fictions that work because we maintain them together.In a constrained system (slow throughput, gatekeepers filtering, local accountability, high production costs), the intersubjective foundation was relatively stable. Common narratives emerged because there were only so many narratives in circulation, and communities had time to converge on shared versions.Today, with those constraints removed, we’re witnessing something different: the rapid construction of parallel realities. Algorithmic feeds and self-selected communities enable the formation and reinforcement of alternative factual universes. What was once a broadly shared “mainstream” (however imperfect) splinters into countless micro-narratives, each with its own axioms, heroes, and standards of evidence.This is not pluralism. Pluralism requires a shared foundation of facts and a commitment to a common deliberative space. What we’re seeing is fragmentation: the loss of the foundation itself. When there’s no agreement on basic facts (the safety of vaccines, the outcome of an election, the historical record), the possibility of compromise or collective problem-solving is significantly degraded. The intersubjective commons, the cognitive public square where a society meets to hash things out, has been subdivided into private, walled gardens.You can still talk to your neighbor. But you may no longer share a world with them.Symptom 2: The Attention CrisisIf shared reality fractures at the societal level, attention shatters at the individual level.Herbert Simon observed decades ago that “a wealth of information creates a poverty of attention.” What was once an economist’s adage is now a lived experience.The prefrontal cortex, the seat of executive function and deliberate judgment, is a slow, energy-intensive system. It evolved for depth, not breadth; for sustained focus, not continuous partial attention. The unconstrained linguistic environment constantly summons it: claims, crises, outrages, narratives, each optimized to capture attention and demand a micro-decision about whether to engage or scroll past.The result is chronic cognitive overload. The symptoms are familiar: the inability to concentrate on long-form texts, the anxiety of the endless “to-read” list, the compulsive checking of feeds, the feeling of being perpetually informed yet never quite understanding. Attention, the finite resource that directs intelligence, becomes so fragmented that deliberate thought becomes a luxury. We’re left in reactive mode, buffeted by waves of language, unable to secure the space required for synthesis.The interface isn’t just overwhelming society; it's also overwhelming society. It’s overwhelming for the individual user.Symptom 3: The Authenticity ParadoxWhen the linguistic interface fails to provide a stable shared reality, a compensating demand emerges: authenticity. Be real. Be yourself. Cut through the noise with something genuine.The yearning makes sense. In a world where the old anchors of meaning have dissolved, we crave unmediated connection and a trustworthy signal.But this ideal doesn’t stand up to scrutiny. Humans are social animals, exquisitely sensitive to audience and context. We adjust our self-presentation constantly: to build alliances, avoid conflict, signal belonging, and optimize social outcomes. This isn’t duplicity. It’s the software of a highly social species running as designed.The demand for a context-invariant “authentic self” misunderstands this design. In practice, “authenticity” often becomes just another performative genre: a set of signals (casual dress, personal disclosure, performed vulnerability) that is itself curated for social reward. The workplace mandate to “bring your whole self to work” is the purest example of the paradox: an institution that requires role-playing and goal-oriented behavior officially endorsing an ideal that, if genuinely followed, would disrupt its functioning.The paradox reveals something deeper. We crave unmediated connection, but we must seek it through language, a tool that, by design, always mediates, always translates. We’re asking the interface to deliver something it structurally cannot.Symptom 4: The Wisdom GapThe final symptom is the widening chasm between knowledge and wisdom.Knowledge is abundant, accelerating, and increasingly outsourced. Facts are a click away. Intelligence, in the sense of pattern recognition and inferential speed, is being simulated and scaled by machines.Wisdom is different. It’s the capacity to navigate uncertainty, recognize the limits of one’s own perspective, hold conflicting truths in mind, and prioritize long-term goals over short-term advantage. It requires slowness, reflection, and epistemic humility.The unconstrained linguistic environment actively undermines these conditions. Wisdom needs quiet; the environment provides noise. Wisdom needs time; the environment demands reaction. Wisdom requires tolerance for uncertainty; the environment rewards confident, shareable takes.We’re creating a world rich in information and synthetic intelligence, yet increasingly inhospitable to the slow, humble, integrative process that turns data into discernment.The Accelerant: LLMs and the End of FrictionInto this already-strained system, large language models have arrived. They’re often discussed as a leap toward artificial general intelligence, a new form of reasoning, or an existential risk. Those debates have their place. But they can obscure a more immediate architectural fact: LLMs complete the demolition of linguistic friction.Think of language production as climbing a friction gradient: a slope you have to work against. Scribes faced a steep climb (rare skills, expensive materials, slow copying). The printing press reduced the slope. Typewriters and word processors reduced it further. The internet collapsed distribution costs, but the gradient remained: generating coherent, persuasive content still required human cognition, time, and effort.LLMs don’t just further reduce the slope; they eliminate the gradient entirely. The landscape is now flat. Language flows in every direction without resistance. The cognitive cost of producing fluent, seemingly knowledgeable text has dropped to zero.This has three compounding effects:First, volume overload. The quantity of plausible text increases exponentially. The “sea of noise” is now filled not just with human chatter but also with automated systems that never tire and can generate personalized streams for every individual.Second, a persuasive scale. LLMs don’t just generate text; they generate rhetorically effective text. They can mimic authority, empathy, or conspiracy. They can tailor arguments to known biases and produce convincing fake supporting evidence. Industrial-scale disinformation becomes trivially accessible.Third, the erosion of effort heuristics. Humans use cognitive shortcuts to navigate information overload. One key shortcut is “effort as credibility”: the assumption that a long article or detailed report signals invested effort and likely substance. LLMs destroy this heuristic. They can generate an “invested effort” signal with zero actual investment. A core tool for navigating the linguistic environment is rendered obsolete.LLMs are not the cause of the epistemic crisis. They’re the accelerant that removes the last remaining point of friction and locks us into an infinite-language world with finite human attention.The Plastic Failure WarningSo far, what we’re experiencing is serviceability failure. The system is uncomfortable, unreliable, and increasingly unfit for purpose. But it’s still standing.The danger is that prolonged serviceability failure can escalate.In structural engineering, plastic failure occurs when a material is stressed beyond its yield point and deforms permanently. A steel beam bent past its limit doesn’t spring back. Its integrity is compromised forever.The equivalent risk for our epistemic infrastructure is this: if the core materials of society (trust, shared truth, good-faith disagreement) are strained beyond their yield point, the deformation may become permanent. Shared reality could fragment into mutually incomprehensible shards. The common ground required for large-scale cooperation could collapse and not return.We’re not there yet. But the trajectory matters. A system in serviceability failure can be repaired. A system that has undergone plastic failure cannot return to its original shape.The Question We’re Left WithThe constraints that stabilized language for millennia are gone. We can’t restore them by fiat. The gates are open; the friction is erased; locality has dissolved into global, anonymous networks.So what do we do?The instinctive responses (centralized fact-checking, algorithmic censorship, retreating to informational bunkers) attempt to rebuild walls in an open field. They treat symptoms by reimposing external constraints on a system that has already evolved beyond them.There may be another path: one that shifts the locus of stability from the environment to the mind. Not more information, not faster processing, but a different orientation entirely.In the final essay, we’ll explore what that might look like.This is the second in a three-part series on language, constraints, and the crisis of shared meaning. Next: “Wisdom in a World of Infinite Language.”Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

A quick note: I'm postponing the planned follow-up on "bigger is better" AI to share something more foundational first, a three-part series on language, constraints, and how we build (or lose) shared meaning. At 432 Park Avenue in Manhattan, you can buy an apartment for $30 million and still not be able to live in it comfortably. The building isn’t going to collapse; structurally, it’s sound. But its extreme slenderness (a 15:1 height-to-width ratio) makes it sway in the wind. Chandeliers swing. Elevators misalign. Residents report motion sickness in their own living rooms. The creaking keeps them awake at night. Engineers installed tuned mass dampers, giant pendulums designed to counteract the movement, but they can’t fully tame what the design made inevitable.Engineers have a term for this: serviceability failure. The structure stands, but it no longer serves its purpose.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.Something similar has happened to language itself.For most of human history, we didn’t think of language as a system that could fail. It was just there, the medium through which we argued, promised, persuaded, and built civilizations. What we didn’t notice were the invisible constraints that kept it stable: limits on speed, gatekeepers who filtered, communities that enforced accountability, and the sheer difficulty of producing text at scale.Those constraints are gone. And the system they supported has begun to sway.Language as InterfaceHere’s a reframe that might feel uncomfortable at first: language is not a window onto reality. It’s a dashboard.The cognitive scientist Donald Hoffman argues that our senses didn’t evolve to show us truth, they evolved to keep us alive. What you see isn’t the world as it actually is; it’s a species-specific user interface, optimized for survival. The desktop on your computer doesn’t show you the actual voltage states of transistors; it shows you folders and trash cans because that’s useful. Your visual system works the same way. Fitness beats truth.If perception is the interface for navigating physical reality, then language is the interface for navigating social reality, the world of alliances, obligations, status, and shared belief. Language lets us compress complexity (“trust,” “debt,” “law”), coordinate action (“meet me at dawn”), and manage relationships (apologies, promises, gossip). It’s the operating system of civilization.But here’s the thing about interfaces: they can be overloaded. When the volume of signals exceeds the system’s processing capacity, the interface doesn’t crash dramatically. It degrades. It starts generating incompatible outputs for different users. It becomes unreliable precisely when you need it most.For millennia, the linguistic interface was kept stable by four external constraints, load-bearing walls that we never noticed because they were always there. Let me walk you through what we’ve lost.Constraint 1: ThroughputThe speed at which meaning could travel.For most of history, information moved at the pace of feet, hooves, and sails. Oral traditions were confined to memory and walking speed. Manuscripts required scribal labor; copying was slow and expensive. Even after Gutenberg, distribution meant physical logistics: wagons, ships, shops.This slowness wasn’t a bug. It was a feature. Ideas spread gradually, giving communities time to digest, debate, and integrate them. Rapid informational shocks were rare. When a new claim appeared, there was time to test it against experience before it reached the next village.The telegraph began the acceleration. Radio and television enabled one-to-many broadcast at the speed of light. The internet completed the transition: from many-to-many communication to instantaneous, global communication. The temporal buffer vanished entirely. News cycles collapsed from days to minutes. Narrative waves now form and crash in hours.The human mind, evolved for deliberative pace, is forced into a continuous reactive mode. The “digestive” time required to separate signal from noise, test claims, and form a coherent understanding? Eliminated.Constraint 2: BottlenecksThe gatekeepers who filtered what reached the public.Every society has them: elders, priests, scribes, publishers, editors, broadcast networks. These bottlenecks were imperfect, often biased, and sometimes corrupt. But they performed a crucial function. They enforced minimum standards of evidence and coherence. They filtered out the most extreme noise. They created a limited set of “authorized” narratives around which public discourse could organize.You might not have liked what the gatekeepers let through. But at least there was a through, a common channel that most people encountered.Digital platforms dissolved this. The blogosphere eliminated editors. Social media algorithms replaced human curators with engagement metrics. The barrier to reaching a mass audience dropped from “convince an editor” to “trigger an algorithm.”The result: the curation filter was replaced by a virality engine. The most emotionally charged, identity-reinforcing content rises regardless of its truth value. The concept of a “mainstream” fragments into a million micro-narratives, each optimized for its niche, each increasingly incomprehensible to those outside it.Constraint 3: LocalityThe geography of accountability.In a village, tribe, or city-state, the speaker was known. Reputation was tangible currency. If you spread a harmful lie, you faced your audience the next day, at the well, in the market, or at the temple. This created a powerful feedback loop: communication was a high-stakes social act, embedded in ongoing relationships.Mass media created distance between the speaker and the audience. The internet completed the separation. We now routinely consume language from sources with no connection to our physical community, no shared history, and no accountability to our social norms. Anonymous. Pseudonymous. Distant.The link between communication and consequence is severed. The costs of deception, exaggeration, or incitement are externalized, borne by the audience, not the speaker. This creates conditions where the most inflammatory language is actually incentivized: high engagement rewards, minimal social risk.Constraint 4: FrictionThe cost of producing and distributing language.This is the one we feel least nostalgic about, because friction felt like oppression. Writing requires literacy. Publishing requires capital. Broadcasting requires licenses and infrastructure. These barriers excluded many voices that deserved to be heard.But friction also served as a natural limiter on volume and a rough proxy for commitment. Someone who had expended significant resources to broadcast a message was, on average, more likely to believe in it. The investment signaled skin in the game.Digital tools reduced writing and design costs to near zero. Social platforms absorbed distribution costs. And then came the final step: large language models, which reduce the cost of generating fluent, persuasive text to effectively nothing.An individual can now produce a volume of credible-sounding content that would have required an entire institution a generation ago. The last natural check on the sheer quantity of language is gone. The signal-to-noise ratio plummets. The effort required to generate personalized propaganda, synthetic consensus, or industrial-scale disinformation is trivial.The Multiplicative EffectHere’s what makes this genuinely dangerous: the removal of these constraints isn’t additive. It’s multiplicative.· High throughput + no bottlenecks = viral misinformation without filters.· No locality + no friction = toxic speech with no accountability or cost.· All four removed = a system in which language is infinite, attention is finite, and the architecture for building shared meaning has been dismantled.This isn’t a story of moral decline. People aren’t worse than they used to be. It’s a story of architectural failure. We removed the load-bearing walls of our epistemic infrastructure while dramatically increasing the load. The linguistic interface, designed for a slow, expensive, locally accountable world, is now subjected to forces it was never built to handle.What Comes NextThe building is swaying. The question is whether it’s a serviceability failure, uncomfortable but recoverable, or the early stages of something worse.In structural engineering, there’s a grimmer category: plastic failure. That’s when a material is stressed beyond its yield point and deforms permanently. A steel beam bent past its limit doesn’t spring back. Its integrity is compromised forever.The danger with our current epistemic crisis is that a prolonged serviceability failure could become plastic. If the core materials of society, trust, shared truth, and good-faith disagreement, are strained beyond their yield point, the deformation may be permanent. Shared reality fragments into mutually incomprehensible shards. The common ground required for large-scale cooperation collapses. The structure of collective understanding bends and doesn’t return.We’re not there yet. But the creaking is getting louder.In the next essay, we’ll look more closely at what serviceability failure actually looks like when it’s happening to meaning itself, and why the usual fixes aren’t working.This is the first in a three-part series on language, constraints, and the crisis of shared meaning. Next Week: “Serviceability Failure.”Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

We optimize everything: productivity, fitness, finances, sleep. There’s an app for each, a metric for every goal. But when did you last systematically examine whether you’re even focused on the right things?A Different PerspectiveMost self-improvement frameworks assume you already know your priorities. Pick your domains, set your goals, track your progress. But here’s the uncomfortable truth: priorities shift, and we’re often the last to notice. What mattered deeply at 30 might be irrelevant at 50. The relationship you took for granted becomes the thing you’d sacrifice everything else for. The career ambition that drove you for decades quietly stops mattering, but you keep optimizing for it anyway, out of habit.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.The value isn’t in tracking what you think matters. It’s in discovering what actually does. And, just as importantly, what doesn’t.Quick and SimpleThe practice is simple: daily reflection, constrained. Five minutes, ideally at the same time each day. You score a set of life components, not with journaling or elaborate prose, but with a simple scale. Did this area drag me down today, lift me up, stay neutral, or just wasn’t a factor at all (not applicable)?The constraint is the point. You can’t hide behind words. Over weeks, patterns emerge that surprise you. The thing you thought was critical to your happiness? Turns out it barely moves the needle. The thing you’ve been neglecting? Strongly correlated with your best days.This isn’t about optimizing your life into a spreadsheet. It’s about building a muscle: the capacity to notice, reflect, and adjust. Most of us sleepwalk through our days. This practice interrupts that.The Life Mastery TrackerI decided to build a web app because it helped me think through what matters to me, gave me an opportunity to build one that I could use, and let me try several LLMs as coding assistants to learn their strengths and weaknesses for a non-coder.The app is simple by design, and I am sure there is plenty of room for improvement in the user interface and functionality. I believe the latter will involve expanding the analytics based on experience and user feedback.A few things worth noting:It’s intentionally minimal. The check-in takes under five minutes. You score each component, your overall happiness and fulfillment, and include an optional note; you’re done. No gamification, no streaks-for-streaks-sake, no guilt. I am sure there is plenty of room for improvement in the user interface and functionality. On the functionality front, I believe this will involve expanding analytics based on experience (e.g., developing a personalized utility function for happiness and fulfillment and aggregating anonymized, demographically segmented scores of standard components’ contributions to overall happiness and fulfillment)You don’t need to get the components right up front. The profile setup process includes several ‘standard’ components that are important to me right now. You can select any that you think might be applicable to you, and you also have the opportunity to create your own custom components and scoring guidance. Custom components and scoring guidelines are encrypted and visible only to the user through the user interface. They will never be incorporated into any anonymized aggregated dataYour data, your control. Custom components, scoring guidance, and daily notes are encrypted end-to-end; I can’t see them, and they’re never shared. Standard component scores (the ones you select from the preset list) can be anonymized and aggregated to surface patterns across users, for example, how “Sleep” correlates with overall happiness across different demographics. During the beta, this aggregation is required; afterward, you can opt out entirely.Beta Testers WantedI’m looking for 50 beta testers.The commitment: use the app daily for 30 days. At the end, share what worked, what didn’t, and what’s missing. I’ll incorporate feedback into the next iteration. Please note: for the purposes of the beta, users agree to the anonymized aggregation of their standard components, as described above. Beyond the beta, all users will be able to opt out of anonymized data sharing, including those who participate in the beta.If that sounds like you: https://life-mastery-frontend-production.up.railway.app/loginNo Application RequiredNot ready to commit to an app? That’s fine. The practice can be done with a pen and paper, index cards, or a notes app. The web app is designed to reduce friction and, ultimately, to increase each user’s understanding of what matters to them and to gain insights from anonymized, aggregated data on the ‘standard’ components.What matters is building the muscle. Everything else is scaffolding.Try the practice manually: - List 8–10 life components (e.g., Health, Relationships, Work, Learning, Creativity). - Each day, score each on a five-point scale (from −2, indicating a strong negative impact, to +2, indicating a strong positive impact). - After 2 weeks, look for patterns.Looking ahead: What does this have to do with AI?This kind of daily reflection, minimal, intentional, human-centered, is more than a personal practice. It’s a small-scale model for how we might design technology that augments rather than replaces human intelligence.Next week, we’ll explore a related idea in artificial intelligence. Just as blindly optimizing for more data or bigger models won’t lead to true understanding, blindly optimizing our lives without reflection won’t lead to true fulfillment. The prevailing “bigger is better” narrative in AI is starting to crack, and what emerges next may look less like artificial general intelligence and more like augmented human intelligence, modular, attentive, and designed to elevate what humans do best.If today’s tool helps you notice what truly matters in your life, next week’s piece will ask: What if we built AI to help us do exactly that?This is part of an ongoing series exploring Augmented Human Intelligence and the architectural principles that might guide us toward AI systems that enhance rather than replace human judgment. If you found this interesting, consider subscribing to future posts on evolutionary processing units, attention architectures, and the intersection of human and artificial intelligence.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

The same architectural principles that govern intelligent systems reveal why our public discourse is failing.My journey from instinct to deliberation was a personal integration project.The Mastery of Life scaffold emerged from applying what I had learned about the brain’s modular architecture and attention as a finite resource to a simple but complex question: how does one live well in an environment designed to hijack focus?Thanks for reading, James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.That work was personal. But systems do not exist in isolation. They operate within hierarchies of larger systems. So I zoomed out. What I saw was not disagreement, or polarization, or even misinformation. What I saw was an architectural failure, one that makes those outcomes almost inevitable.The Architectural LensStudying intelligence through the lens of biology and artificial systems taught me a simple lesson: intelligence does not emerge from raw processing power. It emerges from coordination.The brain’s genius is not its impressive computational capacity. It is its structure. Specialized modules, perception, memory, emotion, reasoning, all hand off information under the governance of executive attention. That executive function decides what gets priority based on context and goals. It is what allows you to override an impulse, to hold competing ideas at once, to revise beliefs when evidence demands it.Executive attention is the allocator. When it works, intelligence emerges. When it is compromised, reasoning collapses. Now consider what happens when something else begins making those allocation decisions for you.Executive CaptureI used to think of social media’s effects in terms of distraction, too much noise, too little signal. That framing is insufficient. The deeper problem is its capturing of our executive function.When I open a (social) media feed optimized for my engagement, I am not merely distracted. I am delegating my attention allocation to a system with goals that are not my own. The algorithm decides which cognitive modules get activated: outrage, tribal identity, and novelty-seeking. It makes those decisions thousands of times per session, invisibly and automatically. This is not a metaphor.It is the same architectural dynamic I described in my work on biological and artificial intelligence, inverted. Evolution spent billions of years refining executive attention to serve the organism’s survival and flourishing. The attention economy has spent two decades optimizing how to capture that executive function and redirect it toward engagement metrics. The result: My own cognitive architecture, running someone else’s priorities.Why This Breaks DemocracyDemocracy is not just a political system. It is a cognitive one.For a democratic society to function, citizens must be able to:* Hold competing ideas simultaneously* Update beliefs based on evidence* Coordinate around shared facts* Override tribal impulses when warrantedEvery one of these requires executive attention. But there is a critical upstream problem. Deliberation assumes a shared factual foundation, common ground from which disagreement can proceed. Algorithmic personalization dissolves that foundation before deliberation begins. Citizens are not merely distracted; they are operating from incompatible information environments. You cannot update beliefs based on evidence your neighbor never sees.Democratic institutions evolved alongside slow information architectures: newspapers, town halls, letters. These imposed natural friction that allowed executive function to keep pace. I would argue that friction here was a feature, not a bug. In the attention economy, friction is "waste"; in a democracy, friction is "reflection." The algorithmic feed removes that friction entirely, delivering allocation decisions at machine speed to cognitive systems designed for human tempo.Consider each capacity in turn. Holding competing ideas requires sustained attention; algorithmic feeds reward snap judgments. Updating beliefs demands cognitive effort; outrage-optimized content makes the prior position feel righteous. Coordinating around shared facts presumes a common information base; personalization fragments it. Overriding tribal impulses is executive function’s hardest task; engagement algorithms specifically target tribal identity because it drives clicks.I suspect you’ve felt this erosion. After heavy social media use, your capacity for sustained reading dips. You reach for your phone mid-paragraph. Opinions harden faster. This is the mechanism at work.Even mainstream editorial voices are beginning to name this dynamic. In its New Year's Day editorial (2025 was dismal. It doesn’t have to be this way.), the Boston Globe urged voters to "develop a spidey sense for the ways that social media algorithms encourage polarization and groupthink," noting that platforms "make money by stoking endless cycles of outrage." The diagnosis is correct. What remains unasked is: how does one develop that spidey sense? The answer requires more than awareness, it requires rebuilding the executive capacity that makes such awareness actionable.The attention economy does not merely distract citizens. It systematically weakens the very cognitive capacities that democratic self-governance requires. Citizens whose executive attention is captured can still vote, but their ability to deliberate is significantly diminished. Elections continue, but they occur within a marketplace of minds whose allocation systems have been quietly compromised.The MOL Scaffold as DiagnosticBuilding the Mastery of Life framework gave me the lens to see this clearly.As I began deliberately allocating attention across core life domains, observing where focus actually went versus where it should go, the external pull became impossible to ignore. My attention was not simply wandering. It was being pulled by systems I had not consciously chosen.The MOL process, Awareness → Attention → Adaptation, is not self-optimization. It is the process of regaining executive function. It is the practice of taking back the allocation decisions that determine which cognitive modules run and when. This is why the personal scaffold matters beyond individual well-being. You cannot diagnose a hostile architecture while you remain trapped inside its most reactive loops. You must first rebuild executive capacity; otherwise, the system remains invisible.Beyond Policy: Architectural AlternativesMost responses to the attention economy focus on policy: regulation, transparency mandates, and antitrust action. These may be necessary, but they are insufficient. They treat symptoms while leaving the underlying architecture intact. The brain does not rely on a single, centralized attention allocator. It uses multiple interacting systems, bottom-up and top-down, stimulus-driven and goal-directed, that compete and correct one another. Executive function coordinates; it does not monopolize.Our information environment is the opposite. Attention allocation is centralized, optimized around a single metric, and imposed at a planetary scale. When allocation centralizes, power concentrates, even if intentions are benign. The architecture itself does the damage. The alternative is not better algorithms. It is distributed attention governance.Systems where: * Allocation decisions happen closer to the individual.* Multiple signals inform what surfaces, not just engagement. * Human executive function remains primary, augmented rather than replaced.This is the same principle underlying Augmented Human Intelligence rather than artificial general intelligence. The goal is not to replace human cognition, but to design systems that strengthen it rather than capture it.From Personal Practice to Collective DesignI began this series exploring how intelligence emerges from architecture in biological and artificial systems. I applied those principles to deliberate living. Their implications for collective intelligence now follow naturally. The attention economy is not merely making us anxious or divided. It is degrading the cognitive infrastructure on which democracy depends.The Mastery of Life scaffold is not an escape from this problem. It is a prototype. It demonstrates what reclaimed executive function entails at the individual level. The open question is whether we are willing to scale that insight, whether we can build information architectures that augment collective deliberation rather than fragment it.I do not have a complete answer. But I am confident of this:The fight for a deliberative democracy and the practice of a deliberate life are the same, operating at different scales. One prepares you for the other. We began by building a scaffold for a single life. Now we see its deeper purpose: to become the first footing in a broader foundation, one capable of supporting a self-governing society guided by intention rather than impulse.#ParallelHorizons #DeliberativeDemocracy #AttentionEconomy #CognitiveArchitecture #CivicInfrastructureWhat’s NextWe’ve traced the problem from our captured cognitive architecture to its civic consequences. But every systemic critique begs a practical question: What do I do on Monday? Next week, we return to the Mastery of Life track to introduce a simple, private, and secure tool that I have built to operationalize the framework, a starting point for the daily practice of reclaiming executive attention.This is part of an ongoing series exploring Augmented Human Intelligence and the architectural principles that might guide us toward AI systems that enhance rather than replace human judgment. If you found this interesting, consider subscribing to future posts on evolutionary processing units, attention architectures, and the intersection of human and artificial intelligence.Thanks for reading, James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

When we define a fantasy, we attempt to govern that fantasy. It’s time to correct the definitions that are leading us toward unmanaged risk.A collaborator of mine, Ava Neeson (https://www.linkedin.com/in/olivia-ava-neeson/), recently updated her LinkedIn tagline to include: “Stop Calling Language Models AI.”Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.She’s a CTO, not a philosopher, but a practitioner watching the phrase “Artificial Intelligence” get stretched so thin that it now covers everything from a chatbot to a chip scheduler. This isn’t semantic nitpicking. It’s a flare shot over a governance No Man’s Land: you cannot govern what you cannot accurately define.Three Definitions, One ProblemConsider three categories of definitions currently vying for authority:The Regulatory Definition (OECD, U.S. Executive Order): “A machine-based system capable of making predictions, recommendations, or decisions.”The Functional Definition (Academic Consensus): “Systems that perform tasks typically requiring human intelligence, such as learning, reasoning, problem-solving, and perception.”The Marketing Definition (Prevailing Narrative): “Machines that think, learn, and reason like us.”Only the second one seriously attempts to describe intelligence. The first is deliberately functional. It is designed for legal guardrails, not conceptual clarity. The third is where our trouble begins. It is the definition of fantasy, and it is the one that dominates press releases, boardroom presentations, and public imagination.What Large Language Models Actually DoWhat is a Large Language Model, really?It is a statistical prediction engine of breathtaking scale and fluency. It identifies patterns in training data and predicts the most probable next token. It simulates understanding by mastering the form of human communication. As a landmark paper put it, it is a “stochastic parrot.”(1)Yet we call it “AI,” invoking the second and, especially, the third definition above. We imply capabilities it does not possess. Let’s be precise:Learning implies acquiring new knowledge from experience and retaining it. LLMs don’t learn from our conversations. They were trained once on a static dataset. The model itself doesn't learn from our exchange, its neural network remains exactly as it was before we started. Any memory of past conversations is useful, but it DOES NOT change the underlying LLM neural network. Reasoning implies drawing inferences, weighing evidence, and constructing arguments through logic. LLMs produce outputs that resemble reasoning because they’ve ingested billions of examples of human reasoning. Pattern-matching is not thinking.Perception implies sensory awareness: receiving and interpreting stimuli from an environment. LLMs have no senses. They have no environment. They process text inputs and generate text outputs. Calling this “perception” stretches the word beyond recognition.Problem-solving implies identifying obstacles, generating novel solutions, and testing them against reality. LLMs generate plausible-sounding responses based on statistical patterns. When those responses are incorrect, they have no mechanism to recognize the error, let alone correct it.This is more than a marketing gloss.It is a category error with operational consequences.What a “Category Error” Really MeansA category error is a fundamental logical mistake: assigning something to a category to which it does not belong. Asking “What color is the number 7?” is a category error; numbers aren’t in the category of colored objects.Labeling LLMs as “Artificial Intelligence” commits the same error. “Intelligence” refers to cognitive processes (understanding, reasoning, adaptation). LLMs belong to the category of statistical functions (pattern recognition, probabilistic prediction). Calling an LLM “AI” because it produces intelligent-seeming text is like calling a flight simulator a “jet” because it looks like one on a screen. The resemblance is useful, but they operate on entirely different principles, with different risks and purposes.This misclassification isn’t philosophical. It’s practical.Why Mislabeling Creates RiskWhen a CEO hears “AI,” they imagine a partner with judgment. They delegate accordingly. When an engineer deploys an “agentic AI,” they are actually launching a multi-step autonomous software process that optimizes for a proxy goal, often with unpredictable emergent behavior.The gap between expectation (intelligence) and reality (complex automation) is where risk breeds.This mislabeling creates a fatal control mismatch. Imagine using a pilot’s manual, written for a conscious agent who understands weather, navigates by landmarks, and makes judgment calls, to operate an autopilot system. You’d be missing the entire checklist for the autopilot itself: sensor calibration, software validation, failure mode protocols, and the manual override switch.That’s what we’re doing. We’re governing the fantasy of a “reasoning agent” while the real risks: hallucinations, data corruption, cost overruns, and goal drift require a completely different manual built for high-stakes automation. This is why AI systems launch “with less control than a new hire.”(2)When I was growing up, our family home began to show larger and larger cracks in the walls. It turned out that our house was subsiding. The structural engineer informed us that, absent corrective action, the home could eventually collapse. We moved out for a year while engineers and contractors undertook underpinning work to stop the subsidence.Our current approach to AI governance has the same problem: it’s built on a subsiding foundation. That foundation is the category error itself, the mistake of treating a statistical function as a cognitive agent. You cannot build stable controls on a definition that misidentifies what it is you’re trying to control.A broken foundation guarantees structural failure.The Pivot: From AGI to AHIIf we revise our definitions to reflect what actually exists, a more productive question emerges:Not “when will it be capable enough to take over?” but rather: “What is the right architecture for human-machine partnership?”This is the shift from Artificial General Intelligence (AGI) to Augmented Human Intelligence (AHI).The AGI narrative centers on replacement and autonomy. It shapes design toward independence. Oversight becomes an obstacle to be removed.The AHI narrative inverts this. The goal is not replacement but symbiosis. It is about creating not autonomous agents we hope won’t drift, but architected systems where human judgment is embedded at the critical points of oversight, ethical weighting, and strategic intervention.Architecture Over ScaleThe current AI arms race is built on a seductive premise: intelligence emerges from scale. Train larger models on more data with more compute, and eventually, something like general intelligence will emerge.This is the “bigger model” paradigm. It has delivered impressive results and diminishing returns. Each generation requires significantly more resources for only incremental improvements. The scaling curves are flattening. The energy costs are exploding. And the fundamental limitations remain.Scaling a flawed architecture doesn’t create intelligence; it makes a larger, more expensive, and potentially more dangerous approximation of one. You cannot brute-force your way to wisdom.There’s an alternative hypothesis: intelligence emerges from architecture, not scale.Consider the human brain. It doesn’t succeed through brute-force computation. It succeeds through an elegant modular architecture: specialized subsystems for perception, memory, language, motor control, and emotional regulation, all integrated by sophisticated coordination mechanisms (the executive function of the prefrontal cortex). The brain is efficient precisely because it doesn’t try to solve every problem with the same general-purpose approach.What if AI development followed similar principles? Imagine modular systems, specialized components, sophisticated integration, and human oversight embedded in the architecture rather than bolted on afterward.This is the AHI design philosophy:· Humans set the frame, define the guardrails, and hold the ultimate “break-glass” authority.· Machines handle scale, pattern recognition, tedious computation, and execution within that frame.This is not a limitation. It is a design philosophy for safety and efficacy. A pilot and an autopilot. A surgeon and a robotic scalpel.Clear Definitions Enable Clear GovernanceCall them what they are: Large Language Processors. Predictive Engines. Autonomous Software Agents. Retire the lazy, all-encompassing “AI” for specific terms that match controls to true capabilities and failure modes.This isn’t about limiting potential; it’s about enabling responsible power. An accurate definition is the keystone of control, the load-bearing truth that locks governance, design, and risk management into a stable structure. Without it, you’re just hoping the machine guesses what you want.Governance must evolve from static IT checklists to dynamic systems oversight, think air traffic control for high-speed automation. This can only happen when leadership’s mental model shifts from “deploying intelligence” to “orchestrating capability.”In the realm of technologies that scale risk as fast as they scale value, semantic clarity isn’t a nicety; it’s the foundation of operational safety.We must stop governing the fantasy of intelligence and start architecting the reality of partnership. Language was humanity's first invention, the artifact that enabled us to coordinate, govern, and rise.(3)(4) It would be darkly ironic if imprecise language about our most powerful technology became the thing that undermined our ability to control it.The first step is to use language with enough precision to say what we actually mean.What’s NextThis is a bonus post that I have developed based on current discussions about the definition of AI and its impact on governance. My previously planned post, zooming out from the personal to the political, will launch on Wednesday, January 7. This is part of an ongoing series exploring Augmented Human Intelligence and the architectural principles that might guide us toward AI systems that enhance rather than replace human judgment. If you found this interesting, consider subscribing to future posts on evolutionary processing units, attention architectures, and the intersection of human and artificial intelligence.References* Bender, E. M., Gebru, T., McMillan-Major, A., & Mitchell, M. (2021). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? FAccT ’21: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency.* Issakova, A. (2025). LinkedIn post on agentic AI governance. https://www.linkedin.com/posts/alexissakova_companies-are-launching-agentic-ai-with-less-activity-7413201150130167808-Rc9f/* Wolfe, T. (2016). The Kingdom of Speech. Little, Brown and Company.* Harari, Y. N. (2015). Sapiens: A Brief History of Humankind. Harper.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

For twelve months after the RIF, my brain ran on a single, looping thought process: Find Job → Ensure Security → Eliminate Fear.It was a simple, powerful instinct. It required no conscious thought and was fueled by deep-seated anxiety. For 25 years, the path had been clear: work, advance, provide. The sudden break from that script triggered a system-wide panic. My internal executive function, overwhelmed, delegated control to the most primitive module: the threat-response system. My attention narrowed to a single, glaring domain: Material Security.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.I put my job search on autopilot, pursuing roles that mirrored the one I’d lost. It was the path of least resistance, and a course of action that made perfect sense for a 50-something-year-old professional. However, there was something nagging at the back of my mind which grew stronger over time. I had the sense that I was missing or ignoring something important.Then, after about a year, the panic lessened just enough for a different kind of thought process, a slower and more reflective one, to be heard. My System 1 (fast, fearful, instinctive) had a clear answer: Find security, now. But a quieter System 2 (slow, reflective, deliberate) was asking a different question:“Is this truly what deserves my focus for the next 15 years?”This wasn’t just a career question. It was a more foundational one. My internal “Biological Processing Unit” (BPU, aka brain) had been hijacked by the instinctive threat-response, and it had essentially silenced all other avenues of thought. I realized I needed to reboot my BPU and re-establish executive control. I needed to apply the same principles I was researching in AI: attention, modularity, and adaptation to my own life.What if “living deliberately,” Thoreau’s 170-year-old aspiration, isn’t about withdrawing from complexity, but about applying systematic attention management within it?The Bridge to Life ArchitectureThat moment of clarity was the first step in rebooting my BPU. But to answer it, I needed a new architecture for how I thought and decided. I couldn’t just rely on instinct; my instincts were still wired for fear.The insight came from the work I was immersed in. The BPU isn’t a general-purpose supercomputer. It’s a federation of specialized systems, visual cortex for sight, hippocampus for memory, prefrontal cortex for command, all competing for a finite budget of energy. Intelligence isn’t raw power; it’s the orchestration of specialized modules by executive attention.My life, I realized, had a parallel architecture. It wasn’t a single monolithic project called “Career.” It was a collection of core domains, each a vital subsystem of a whole life.The Seven Domains of a Life ArchitectureWhen I mapped my own experience through this lens, seven domains consistently surfaced, each a different facet of life’s architecture.* Material Security – Financial stability and the sense of safety it brings* Physical Vitality – Sleep, movement, and nutrition that sustain energy* Mental Clarity – Focus and cognitive balance for clear thinking* Purpose & Growth – Learning, curiosity, and contribution beyond the self* Relational Connection – Depth and quality of human relationships* Self-Regulation – Managing emotional states and habits with awareness* Novelty & Discovery – Exposure to new experiences and ideas that prevent stagnation.These are not arbitrary categories. They are the modular architecture of a human life, each specialized, all interconnected, and all requiring a conscious executive decision about how to allocate the one finite resource we share: attention.For a year, my personal dashboard would have shown one domain blinking red (Material Security: -2) while all others faded to grey from neglect. The framework showed me I wasn’t solving a problem; I was starving a system.Just as the BPU cannot process every input at once, we cannot optimize all domains simultaneously. The question is not “How do I do everything perfectly?” but rather, “Given where I am now, which domain(s) most deserves my focus, and what am I willing to trade off to give it that focus?”This shift from optimization to orchestration is the foundation of the Mastery of Life framework.The Process: A Life’s Operating SystemRecognizing these seven domains gave me a map. But I needed a method for navigation, a new compass to replace the fear-driven autopilot. That method mirrors how any intelligent system, biological or artificial, learns and adapts: Awareness, Attention, and Adaptation.Step 1: Awareness — Observation Without JudgmentMy first step was to stop and actually look at what was happening. I had to force the shift from System 1 to System 2.This meant collecting simple data, not opinions. I began tracking small metrics across the domains:* Physical Vitality: How many hours did I actually sleep?* Mental Clarity: How many focused blocks of work did I achieve versus time lost to anxiety?* Purpose & Growth: Was I spending any time learning, or just refreshing my inbox?This wasn’t about judgment. It was compassionate observation, like a technician reading system diagnostics. The data revealed a truth I couldn’t ignore: my obsession with Material Security was dragging down nearly every other domain. I could now see the trade-offs I was making.Step 2: Attention — Allocating Finite Resources DeliberatelyWith awareness came the power of choice. I could now consciously decide where to allocate my finite cognitive resources. William James captured it perfectly: “My experience is what I agree to attend to.”I chose to de-prioritize the frantic job search and instead focus on Purpose & Growth and Novelty & Discovery. I devoted time to learning, writing, and exploring new ideas about human decision making, AI, the remarkable parallels between them, and how they might cooperate most effectively. This led to my development of Augmented Human Intelligence (AHI) as an alternative to Artificial General Intelligence (AGI).This is much like the self-attention mechanism in AI models. It’s not about processing everything; it’s about assigning importance weights. I was telling my internal system: “For this season, learning and purpose have a higher weight than my continued, myopic focus on financial security.”Step 3: Adaptation — Learning From the Feedback LoopLife doesn’t stand still. Mastery isn’t rigidity; it’s responsiveness. The final step was to close the loop and learn from the results of my new attention allocation.After a few weeks, I reviewed how my tracked metrics had changed. The results were telling:* Intervention: “I focused on learning instead of applying. Did my sense of purpose improve?” Yes.* Counterfactual: “If I had made this shift six months earlier, would I have reduced frustration and anxiety?” Almost certainly.This transformed my life from a reactive script into a learning system. I was no longer a passenger. I was the architect, running small experiments, observing outcomes, and adapting my blueprint. The fear of the unknown was replaced by confidence in my ability to navigate it.The Philosophical Stakes and Call to ActionThis framework pulled me out of a reactive loop and into a deliberate one. But its implications reach far beyond personal course correction. They point to a deeper truth about the era we’re entering.If attention is the fundamental constraint, in the BPU, in AI systems, and in human cognition, then attention management is the meta-skill that determines intelligence. In humans, this is not just cognitive; it’s moral. What you consistently attend to signals what you truly value. Attention shapes not only what you accomplish but who you become.This skill will only become even more critical as AI’s power and capabilities increase. For example, if only a fraction of the doomsday scenarios for AI’s impact on employment and society come to fruition, we will all need to strengthen our Mastery-of-Life muscle.As we approach the integration of human and artificial intelligence through brain-computer interfaces, this question becomes existential. Will these hybrid systems enhance our capacity for deliberate attention, or fragment it further?Technology can be a mirror or a vacuum. Used wisely, it can augment awareness, revealing patterns we might otherwise miss. Used poorly, it can hijack the BPU’s ancient reward systems for commercial gain. The choice isn’t whether to use technology, but what kind of attention architecture we build with it.From Personal to PoliticalThis brings me to the next horizon in this series. We’ve explored how intelligence emerges from architecture in biological and artificial systems. We’ve seen how these same principles can scaffold a more deliberate human life.So, what happens when we zoom out?What if the same architectural principles that govern intelligent systems and deliberate living also reveal a fundamental blueprint for how societies could be organized? If human happiness and fulfillment come from conscious attention management, could our political and economic systems be designed not to maximize output, but to enable citizens to live more deliberately?What would a society optimized for human attention, rather than human consumption, actually look like?My journey from a fear-driven instinct to a more deliberate architecture is still unfolding. The Mastery of Life is the scaffold.If this resonates, I invite you to:* Read the Full Mastery of Life White Paper for the complete framework, from metric design to validation.* Pause here for 60 seconds. Which single domain is shouting the loudest in your life right now? Is it the most important, or just the most urgent (or fearful)? That gap is where deliberate living begins.The first step toward mastery is always awareness.Important note: I realize that I am very fortunate. My circumstances gave me a choice, I had a safety net, between stocks, savings, and retirement accounts. The safety net is a lot smaller Today, but I wouldn’t change a thing, and I would encourage anyone in a similar position, to seize the day and take the time to reflect. You may be surprised by what you find, and I am certain you will benefit from the exercise. It isn’t an easy choice, trust me, I don’t even know how it ends, but I am excited to find out. Most importantly, I believe in myself again. This is part of an ongoing series exploring Augmented Human Intelligence and the architectural principles that might guide us toward AI systems that enhance rather than replace human judgment. If you found this interesting, consider subscribing for future posts on evolutionary processing units, attention architectures, and the intersection of human and artificial intelligence.#AugmentedHumanIntelligence, #LifeDesign, #FutureOfWork, #CognitiveScience, #SystemsThinking, #MasteryOfLifeThanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

We’ve all felt it: that snap judgment, the unscrutinized gut feeling. A hiring manager feels an instant affinity for a candidate from their alma mater. A doctor dismisses a patient’s nagging symptom as stress. This isn’t a failure of character; it’s a feature of our biology. Our brains, for the sake of efficiency, are wired for swift, often flawed, System 1 shortcuts.This is the core problem my work on Augmented Human Intelligence (AHI) aims to solve. The goal is not to outsource our thinking to an artificial mind, but to build systems that enhance our own cognition in real-time. Imagine a partner that doesn’t think for you, but ensures you’re thinking well by mitigating biased reflexes and scaffolding deliberate, System 2 reasoning when it matters most.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.But here’s the rub: our current AI tools are architecturally unsuited for this intimate task. Today’s large language models are like brilliant critics shouting suggestions from another room. They’re disconnected from the real-time, subconscious flow of our thoughts where bias takes root. The clumsy I/O of keyboards and speech creates a bandwidth bottleneck far too narrow for a meaningful cognitive partnership.The architecture is wrong.So what would the right architecture look like? And could emerging brain-computer interface technology point us toward an answer worth exploring?The Vision: Real-Time Cognitive ScaffoldingLet’s return to our hiring manager and imagine what true cognitive augmentation might look like.As she feels that unconscious affinity for the familiar candidate, a sophisticated system, somehow reading the aggregate neural signatures associated with automatic judgment, could detect this moment of System 1 processing.It would not override her. That would be the antithesis of augmentation. Instead, it would provide cognitive scaffolding:- Gently highlighting the other candidate’s superior relevant experience on the digital resume in front of her- Posing a silent, Socratic prompt: “Have you sufficiently weighed Factor X?”- Triggering a subtle state of focused alertness encouraging a moment of deliberate reflectionThe human remains the decider. But she is no longer deciding with a flawed, unaided toolkit. She’s now equipped with a real-time support system to construct sounder judgment.This is the architecture of attention, made tangible. It’s not about AI thinking for you, it’s about providing the scaffolding to ensure you are thinking, clearly and deliberately, when it matters most.The question is: what would it take, architecturally, to make this possible?The Current State of Brain-Computer InterfacesWhile this vision might sound like science fiction, the foundational technology is advancing faster than most people realize. Multiple companies are now implanting devices in human brains, not in distant labs, but in actual patients going about their daily lives.Neuralink: The High-Bandwidth ApproachNeuralink, Elon Musk’s neurotechnology company, represents the most prominent effort in invasive BCIs. By mid-2025, nine people had received the company’s N1 implant, which consists of ultra-thin electrode threads inserted directly into the motor cortex through precisely drilled holes in the skull.What can they do? The results are remarkable for patients with severe paralysis:- Noland Arbaugh, the first recipient (paralyzed from a diving accident), can play video games like Civilization VI, browse the web, and control his computer entirely through thought- Bradford Smith, who has ALS and can no longer speak, used his implant to edit YouTube videos and restored his voice using AI trained on pre-ALS recordings of his speech- Another participant uses the device to operate computer-aided design software, creating 3D objects with his mindThe device uses 1,024 electrodes across 64 threads to capture neural signals with high resolution. Neuralink has raised over $1.3 billion in funding, including a $650 million Series E round in mid-2025, and plans to expand trials significantly, aiming for 20-30 new implants by the end of 2025. They’re also developing “Blindsight,” a device to restore vision by stimulating the visual cortex, which received FDA Breakthrough Device designation.Synchron: The Minimally Invasive AlternativeWhile Neuralink captures headlines, Synchron has taken a radically different approach that could prove more scalable. Their Stentrode device is inserted through the jugular vein, like a stent, and positioned in a blood vessel above the motor cortex. No open brain surgery required.The trade-off? Lower bandwidth. Synchron’s device provides basic “switch” control, think clicks and scrolling, rather than the high-resolution signal capture of cortical implants. But this simplicity has advantages:- Six patients in their COMMAND study showed zero serious adverse events over 12 months- The median deployment time is just 20 minutes- Patients can control their devices “on day one” without extensive training- The device works with Apple products via Switch Control, including iPhone, iPad, and Vision ProSynchron just raised $200 million (November 2025), bringing total funding to $345 million. They’re preparing for pivotal trials in 2026 and moving toward commercialization. The company is also developing a next-generation “high-channel whole-brain interface” that could dramatically increase capabilities.The Broader LandscapeThree other companies are pushing the boundaries in different directions:Blackrock Neurotech (Salt Lake City) has the most human experience with dozens of implants since 2004 through academic research trials. Their Utah Array is the workhorse of BCI research, and their MoveAgain device received FDA Breakthrough Designation. They secured $200 million from Tether in 2024.Precision Neuroscience (founded by a former Neuralink executive) is developing an ultra-thin, flexible “Layer 7” electrode array that sits on the brain’s surface like a piece of tape, inserted through a minimal incision. They raised $102 million in late 2024 and achieved record-setting 4,096-electrode human recordings.Paradromics (Austin) is building a high-bandwidth implant specifically for speech restoration. They completed their first human recording during epilepsy surgery in June 2025 and have raised over $100 million in venture funding plus $18 million in government grants.Industry investment has exploded: $2.3 billion flowed into BCI technology in 2024 alone, more than triple the 2022 level.The Architectural Case for BCIsSo why would a brain-computer interface be relevant to cognitive augmentation in the context of the hiring manager scenario?The answer lies in bandwidth and latency.Current AI interactions require:1. Conscious awareness of a need for assistance2. Explicit formulation of a query3. Physical input (typing/speaking)4. Waiting for processing5. Reading/hearing the response6. Conscious integration of that informationBy the time our hiring manager has noticed her bias, consciously decided to seek help, typed a question into ChatGPT, and read the response, the critical moment has passed. The automatic judgment has already been made.A true cognitive augmentation system would need to:- Detect the neural signatures of automatic, System 1 processing in real-time- Intervene at the moment of decision, not after- Communicate through high-bandwidth, low-latency channels that don’t require conscious attention- Preserve human agency, scaffolding rather than replacing judgmentThis is an architecture problem. And BCIs represent the only technology category that could theoretically provide this level of intimate, real-time integration.Why This Path May Be More Tractable Than AGIHere’s where the argument gets interesting.The dominant narrative in AI is the race toward Artificial General Intelligence, an external, autonomous entity that may ultimately operate beyond our control. This pursuit is fraught with profound alignment challenges: How do we ensure a superintelligent system’s goals remain aligned with an entire species when we can barely agree among ourselves?An AHI system enabled by BCI is architected for alignment by default. Its core purpose is to serve the goals and judgment of its individual human user. The human’s values are the system’s compass. There’s no separate entity with separate goals to align.Yes, BCIs come with serious challenges:- Safety: Surgical risks, long-term biocompatibility, signal stability- Privacy: Neural data is perhaps the most intimate information imaginable- Security: Protection against hacking or unauthorized access- Equity: Ensuring technology doesn’t create cognitive “haves” and “have-nots”- Regulation: How do we govern devices that sit at the intersection of medical device, consumer technology, and fundamental human capabilities, not to mention moral and ethical questions?But critically, these are engineering and governance challenges problems of building secure systems, designing reversible controls, and crafting sensible policy. These are the kinds of problems we have centuries of collective experience solving, even if the stakes here are uniquely high.The Honest LimitationsLet’s be clear about what BCIs cannot do today and may not be able to do for years or decades:Current capabilities are limited: Today’s devices can control cursors, select menu items, and type, all of which are amazing for people with paralysis, but nowhere near the real-time cognitive intervention we’ve been discussing. The hiring manager scenario remains firmly in the realm of speculation.Signal quality matters: The brain is noisy. Distinguishing the neural signature of “automatic bias” from “deliberate consideration” or “fatigue” or “hunger” is an unsolved problem. We’re still mapping these patterns.The “read” problem: All current BCIs are focused on output, translating brain signals into actions. Effective cognitive scaffolding would also require input, the ability to communicate suggestions back into conscious awareness in a way that feels natural rather than intrusive. This bidirectional communication at the cognitive level remains largely theoretical.Long-term unknowns: We don’t yet know the effects of living with a brain implant for 20, 30, or 40 years. Will the brain adapt? Will signal quality degrade? Will there be psychological effects?The consumer value question: For medical applications (restoring movement, communication, or vision), the value proposition is obvious. But would a healthy person undergo neurosurgery for a 10% productivity boost? The practical value of potential consumer applications hasn’t been established.Regulatory uncertainty: The FDA pathway for medical BCIs is becoming clearer, but how would consumer cognitive augmentation devices be regulated? We’re in uncharted territory.These limitations are real and significant. Anyone promising near-term cognitive augmentation via BCI is selling science fiction, not science.The Question Worth AskingSo why write about this?Because the question itself matters: What architectural approach gives us the best shot at building AI systems that genuinely augment rather than replace human judgment?Current AI development is racing toward autonomous systems that think for us. The AHI vision asks: what if we instead built systems that help us think better?Brain-computer interfaces represent one possible path toward that vision, perhaps the only path that could provide the bandwidth, latency, and intimacy required for real-time cognitive partnership. The technology is advancing rapidly, the investment is real, and the early results are promising.But it’s also early. Very early.The medical applications are already life-changing for people with paralysis or neurological conditions. The cognitive augmentation vision? That’s still a thought experiment backed by emerging technology rather than a near-term product roadmap.Perhaps that’s exactly when we should be having this conversation, before the technology is fully realized, while we still have time to think carefully about whether this is a path worth pursuing, and if so, how to pursue it responsibly.Your TurnWhat’s your gut reaction to the idea of a brain-computer interface for cognitive augmentation?Would you consider it for yourself if the safety profile was proven? What applications would be worth it? What makes you hesitate?I’m curious where your thinking leads.This is part of an ongoing series exploring Augmented Human Intelligence and the architectural principles that might guide us toward AI systems that enhance rather than replace human judgment. If you found this interesting, consider subscribing for future posts on evolutionary processing units, attention architectures, and the intersection of human and artificial intelligence. ---Sources and Further Reading:- MIT Technology Review coverage of Neuralink, Synchron, and other BCI companies (2024-2025)- Government Accountability Office report on Brain-Computer Interfaces (GAO-25-106952)- Clinical trial data from COMMAND study (Synchron) and PRIME study (Neuralink)- Industry funding data from NeuroFounders and venture capital analysesThanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

The shortest path between imagination and reality isn’t a straight line; it’s a feedback loop.The Seduction of “Therefore”Michio Kaku has an almost musical ability to connect scientific discovery with cosmic destiny. Watching him describe the future of mind–machine fusion or quantum teleportation feels like watching a physicist conduct an orchestra of possibilities.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.His charisma is undeniable, but so is the leap he often makes between what we can demonstrate and what he declares inevitable.It’s the same move buried in physics textbooks: From this, it is obvious that…Except in Kaku’s case, it becomes: From this first neuron firing on a chip, it is obvious that we will one day upload consciousness.That small word, “therefore,” is the slipperiest in all of science. It hides the distance between evidence and extrapolation, between what is shown and what is merely imagined.The Projection ChainKaku’s futurism often follows a pattern, which I call the Projection Chain:1. Observation: A real, exciting breakthrough (a qubit entangled, a neuron simulated). 2. Extrapolation: Extend the curve, “We’ll soon simulate the entire brain.” 3. Assumption: Treat scaling as understanding, “More compute equals more mind.” 4. Narrative: Frame the future as inevitable, “Uploading consciousness is only a matter of time.” 5. Reward: Capture public imagination (and often, funding momentum).It’s powerful storytelling, but fragile logic. Each, therefore, conceals an ocean of complexity.The Psychology Behind the LeapDaniel Kahneman, in Thinking, Fast and Slow, warns about our love of coherent stories. System 1, our fast, intuitive mode of thought, craves closure. When we encounter an incomplete chain, we automatically fill in the gaps.That’s precisely what happens when we hear: “We’ve mapped one brain circuit; therefore, we can upload the mind.”The story feels complete because the pattern is smooth, not because the reasoning is sound.Kahneman called this the illusion of understanding, the comforting sense that we grasp a complex system merely because we can describe a simplified version of it.Kaku’s futurism thrives in that illusion. It’s not deception; it’s seduction, the same narrative instinct that once convinced us flight, fusion, and artificial general intelligence were “only decades away.”Enter Gary Marcus: The Architectural CounterpointIn Rebooting AI, cognitive scientist Gary Marcus dismantles the same illusion from another angle. He argues that deep learning’s success has led many to mistake scaling for understanding, the belief that more data and larger models will inevitably yield intelligence.Marcus reminds us that intelligence isn’t a single curve to be extended. It’s a modular architecture to be constructed, with reasoning, structure, and feedback loops.Where Kaku says, “We’ll get there if we keep scaling,” Marcus says, “We’ll get there when we “build systems that can reason, reflect, and learn from their own errors.“In that light, Kaku’s Projection Chain is the embodiment of the myth Marcus critiques: that intelligence emerges inevitably from computation.From Projection to DevelopmentTo move from wonder to wisdom, we need a different chain, a Development Chain, built not on extrapolation but on iteration.THE TWO CHAINS: FROM SEDUCTION TO SUBSTANCEOBSERVATION* Projection Chain: Breakthrough event (e.g., neuron simulated on silicon)* Development Chain: Contextualized experiment (e.g., testing under constraints)* Shift: Spectacle → SystemEXTRAPOLATION* Projection Chain: Infinite trendline (”We’ll simulate the entire brain soon”)* Development Chain: Bounded hypothesis (”Scale 10x → what specific capability?”)* Shift: Inevitability → InquiryASSUMPTION* Projection Chain: Scale = Understanding (more compute = intelligence)* Development Chain: Feedback = Understanding (correction loops = intelligence)* Shift: Quantity → QualityNARRATIVE* Projection Chain: “We WILL upload minds” (replacement prophecy)* Development Chain: “We CAN augment minds” (partnership proposal)* Shift: Prophecy → PartnershipVALIDATION* Projection Chain: Public awe & funding (imagination captured = success)* Development Chain: Iterative evidence & integration (measured progress = success)* Shift: Inspiration → IntegrationThis shift from projection to development isn’t just a change in checklist, it’s a change in architecture. And that new architecture demands a partner.The Missing Ingredient: Why Development Chains Need Human-Machine PartnershipThe shift from projection to development isn’t just methodological, it’s architectural.Projection chains assume intelligence emerges from scale alone. You can just feed enough data into a sufficiently large model, and understanding arrives automatically.Development chains recognize something fundamentally different: real intelligence requires continuous correction between various modes of processing.Consider AlphaGo. Its breakthrough wasn’t raw compute; it was the architecture: neural networks for pattern recognition, Monte Carlo tree search for strategic planning, human expert games for initial training, and self-play for refinement.No single component scales to intelligence. It’s the interaction between them that does.This is why pure scaling hits walls. GPT models plateau not because they lack parameters, but because they lack feedback mechanisms that ground prediction in reality, that test hypotheses against consequences, that adjust when the world pushes back.That’s the architectural principle Kaku’s projection chain misses: Intelligence isn’t a computational threshold to cross, it’s a feedback system to maintain.And feedback systems, by definition, require multiple perspectives in dialogue.Case Study: Self-Driving CarsProjection Chain Thinking (circa 2015):“We’ve solved highway driving with deep learning. Therefore, full autonomy is 2-3 years away. We need more data and bigger models.”What Actually Happened:Edge cases exploded. A plastic bag is blowing across the road. Construction zones with conflicting signs. Pedestrians in wheelchairs. Each required different reasoning, visual recognition, contextual judgment, and social prediction.More training data didn’t solve this. The architecture was wrong.Development Chain Approach:Waymo shifted to modular systems: perception modules, prediction modules, planning modules, each with explicit feedback loops. When the system failed, engineers could identify which component was misunderstood, then improve that specific reasoning pathway.Progress became measurable. Failures became informative.The difference wasn’t compute. It was architecture + feedback + iteration.Hybrid Intelligence: The Bridge Between Dream and DesignThe future of intelligence, human or artificial, isn’t about replacement. It’s about co-evolution.Machines can amplify our perception, memory, and reasoning. Humans provide context, meaning, and ethics.Hybrid intelligence recognizes that consciousness isn’t a software feature but an emergent process, a dance between data and experience.Kaku’s cosmic vision has its place. But actual progress doesn’t come from inevitability; it comes from architecture, feedback, and adaptation.That’s not a rejection of his optimism. It’s its necessary refinement.Why It MattersOur age is defined by scale, of models, of data, of hype. But as both Kahneman and Marcus remind us, no amount of computation can replace reflection, structure, or humility.Wisdom, in science or society, grows not from faster conclusions, but from better feedback and reflection.Every “therefore” deserves to be tested. Every assumption deserves a feedback loop.Closing ThoughtsMichio Kaku gives us imagination. Kahneman gives us humility. Marcus gives us architecture.The synthesis of all three, wonder, humility, and structure, is how humanity will truly move from computation to comprehension.Because in the end, It’s not scale that makes us wise, it’s feedback.What’s NextGoing to get more uncomfortable next week when we take augmented human intelligence to a logical place, the Brain Computer Interface. Stay tuned, it is intended to provoke discussion, that’s where humans can move mountains.#AI #HybridIntelligence #MichioKaku #DanielKahneman #GaryMarcus #ThinkingFastAndSlow #RebootingAI #PhilosophyOfScience #AGI #FeedbackLoop #SystemsThinking #AIethics #FutureOfIntelligenceThanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

We now produce somewhere between 15 and 70 trillion tokens of text every day.To put that in perspective: GPT-3 was trained on 500 billion tokens. Humanity now generates that volume in less than an hour. At its peak, the Library of Alexandria held 4 billion words. We surpass that in less than 60 seconds. Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.This isn’t just information overload. It’s something more fundamental: for the first time in human history, the rate at which language is produced has exceeded the rate at which humans can meaningfully process it.Language is infinite. Attention is finite. And the gap is widening daily.Why This Matters More Than You ThinkLanguage isn’t just communication. It’s humanity’s foundational technology: older than agriculture, older than writing, older than the wheel.Yuval Noah Harari describes reality as having three layers: objective reality (physics, independent of belief), subjective reality (your inner experience), and intersubjective reality (shared beliefs that exist because many people agree on them). Money, nations, laws, marriages, corporations: these are linguistic constructs. They have power precisely because they’re collectively imagined and continually reinforced through language.This intersubjective layer is the operating system of civilization. It’s how strangers cooperate, how institutions persist, how societies cohere.And it’s exactly what’s under assault.When language production becomes infinite while attention remains finite, the intersubjective layer destabilizes. Shared reality fragments. Trust erodes. Coordination fails.This has happened before.The Pattern Across CenturiesIn 1486, an inquisitor named Heinrich Kramer published Malleus Maleficarum, a manual for identifying and prosecuting witches. Before the printing press, such a text would have circulated among a handful of clergy. But Gutenberg’s invention had arrived just decades earlier. The Malleus went through 28 editions, reached tens of thousands of readers, and transformed local superstition into a continent-wide intersubjective reality. The result: 110,000 trials, 40,000 to 60,000 executions, mostly of women, over two centuries.The printing press didn’t create misogyny or fear of the occult. It amplified them faster than society could adapt.Five centuries later, the same pattern unfolded in Myanmar.In 2014, Myanmar had almost no internet penetration. Then SIM card prices collapsed, and tens of millions came online almost overnight. For most users, Facebook was the internet, their sole source of news, community, and identity.Into this environment flowed coordinated disinformation portraying the Rohingya Muslim minority as an existential threat. Engagement-optimized algorithms amplified the most inflammatory content. A new intersubjective reality formed: the Rohingya were not merely “others” but enemies to be eliminated.The result was genocide. Over 10,000 killed. Nearly a million displaced. The UN concluded that Facebook played a “determining role.”The technology didn’t create prejudice. It accelerated narrative formation and emotional contagion beyond the capacity of institutions, or citizens, to resist.The pattern is consistent across centuries:- A new language technology emerges- The cost of producing or distributing language collapses- Bad actors exploit the new medium- Institutions lack capacity to filter or regulate the surge- A manufactured intersubjective reality takes hold- Violence followsToday, we face the most extreme acceleration yet. Large language models have collapsed the cost of generating language to nearly zero. A single person with access to a generative model can now produce more text in a day than a medieval monastery produced in a year: tailored, fluent, emotionally optimized, and indistinguishable from human prose.A Personal Inflection PointThe societal attention crisis mirrors an experience I have had directly.After a reduction in force ended a career chapter I’d inhabited for decades, my attention collapsed inward. For nearly a year, it narrowed to a single imperative: replace what had been lost. Find job → Restore security → Eliminate uncertainty. It wasn’t a strategy. It was reflex: the predictable response of a mind shaped by fifty years of expectation and fear.That internal narrative left no space for curiosity, exploration, or reflection.When those efforts repeatedly failed to materialize, something unexpected happened. I stopped. Not entirely (responsibilities don’t vanish), but enough to notice the pattern. Enough to ask a question I hadn’t allowed myself to ask: What do I actually want to pay attention to?The landscape didn’t change. But the meaning of the landscape did.That experience clarified something essential. Just as individuals can drift into lives shaped by inertia rather than intention, societies can drift into informational environments shaped by reflex rather than reflection. In both cases, recovery requires not merely more data or a better strategy, but a reorientation of attention.If attention is the mechanism by which we choose what matters, then the erosion of attention (individually or collectively) is a threat not just to knowledge, but to agency itself.Why an Oracle Won’t Save UsConfronted with infinite information, a tempting solution is to build an AI that sorts it for us: an intellectual oracle that declares what’s true, what’s false, and what deserves our attention.This approach contains profound risks.Any oracle must be trained on data, shaped by human choices, constrained by political forces, and embedded in institutional structures. There is no apolitical oracle. There is no neutral filter. Every mechanism that determines what is true also determines what is permitted.Such systems become targets for capture. The more powerful the oracle, the greater the incentive to manipulate it. History offers numerous examples: church authorities policing orthodoxy, states controlling the media, and platforms shaping algorithmic visibility. A machine oracle merely centralizes this vulnerability.And if humans outsource judgment to machines, our capacity for judgment atrophies. The prefrontal cortex, like any muscle, deteriorates with disuse. The widespread adoption of digital contact storage offers a small but instructive example: the capacity to memorize phone numbers, once routine, has atrophied within a generation.An oracle doesn’t strengthen human cognition. It replaces it. And what is replaced is lost.The Path Forward: Augmented Human IntelligenceThe alternative isn’t artificial general intelligence as oracle. It’s Augmented Human Intelligence (AHI): systems designed not to replace human judgment but to strengthen it.The crisis we face isn’t a deficit of intelligence. Humans reason well when given time, clarity, and context. The crisis is a deficit of attention: the resource required to use that intelligence.AHI treats attention as the scarcest and most valuable cognitive resource. It provides context, identifies manipulation, surfaces what matters, and widens rather than narrows our informational horizons.Where AGI aims for autonomy (an artificial mind operating independently), AHI aims for partnership: systems that amplify human insight while keeping us at the center of decision-making.Consider tools that:- Summarize not just what sources say, but where they diverge- Flag when you’re operating within a narrowing informational corridor- Highlight when emotional triggers are high and analytical engagement is low- Translate across communities of belief, making one group’s assumptions legible to anotherNone of these declares what is true. All preserve the integrity of human judgment while strengthening it.In a world of infinite language, the most valuable technology is one that helps humans reclaim the ability to choose what deserves attention.The StakesLanguage built our civilizations. It gave us laws, markets, stories, institutions, and meaning. But language alone didn’t make us human. What made us human was the ability to attend: to choose what deserves focus, to reflect before acting, to imagine before building.If the breakthrough of artificial intelligence was the realization that “attention is all you need,” perhaps the breakthrough of our time will be the realization that attention is all we have.Democracies depend on citizens who can understand one another. Markets require reliable information. Scientific progress depends on distinguishing truth from fiction. When language becomes infinite, and attention becomes scarce, these premises weaken.We must not mistake polarization for pluralism. Pluralism assumes a shared foundation of facts. Polarization emerges when that foundation collapses.AHI isn’t a luxury. It’s a necessity.What’s NextThis is an unplanned bonus post intended to underscore the importance of attention and to address a comment I received on the scaling post from this past Monday. I encourage anyone to provide me feedback and engage in discussion. Next week, I’ll explore the path from imagination to understanding. Because the real challenge isn’t just building smarter AI. It’s building AI wisely.And wisdom, as it turns out, requires something scaling can never provide: feedback.Notes & Further ReadingThe entire argument, including a framework for evaluating information systems and detailed AHI design principles, appears in my white paper, “The Attention Crisis: Language, Meaning, and the Architecture of Augmented Human Intelligence.”The three-layer model of reality comes from Yuval Noah Harari’s Sapiens: A Brief History of Humankind.The neurobiological foundations draw on Robert Sapolsky’s Behave: The Biology of Humans at Our Best and Worst.The UN’s findings on Facebook’s role in Myanmar appear in the Report of the Independent International Fact-Finding Mission on Myanmar (2018).Token volume estimates are derived from publicly available data on global text production. See the whitepaper appendix for methodology.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

Two years ago, after a reduction in force at my employer, my attention collapsed into a single loop: find a job, secure an income, eliminate fear. It felt rational; it was primitive. I was defaulting to survival mode, locking onto one domain not because it was right, but because it was loudest. Only months later did I recognize the pattern: wherever attention goes, identity follows.“My experience is what I agree to attend to.” - William James.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.In 2017 Google researchers published a paper with a bold title: Attention Is All You Need. It introduced the transformer architecture that powers every major AI system today: ChatGPT, Claude, GPT-N, and more. The breakthrough wasn’t more data or bigger models. It was selective focus: teaching systems to dynamically weight which parts of the input matter most for the task at hand.Here we are today, drowning in information we can’t process, distracted by notifications we can’t ignore, building AI systems that demand ever more of our fragmenting attention. What if the 2017 paper got it right, not just for AI, but for us?What if attention isn’t merely a technical mechanism but the fundamental constraint that shapes all intelligence, biological and artificial?The Technical InsightBefore transformers, neural networks processed sequences linearly, like reading a sentence word by word with no ability to jump back or look ahead. They had fixed, limited memory of what came before. Transformers changed everything with a deceptively simple idea: let the model decide what to pay attention to. When processing the word “bank,” does it mean riverbank or financial institution? The attention mechanism looks at surrounding context: “deposit,” “loan,” “account”, and weights those words more heavily. It doesn’t process all input equally. It prioritizes.This wasn’t just an incremental improvement. It was a fundamental paradigm shift, a sudden change in behavior that rewrote what was possible. Models could now handle longer contexts, understand deeper relationships, and scale to previously intractable problems. But here’s what matters: the breakthrough came from an architectural constraint, not raw power. By forcing the system to allocate limited attention strategically, the researchers discovered something profound.Selective focus isn’t a limitation to overcome. It’s the mechanism that makes intelligence possible.Nature’s VersionThe human brain processes around 11 million bits of sensory information per second. Your eyes, ears, skin, and other senses deliver a constant flood of data. Your conscious awareness? About 50 bits per second. That’s not a bug; it’s an engineering feature. The brain doesn’t try to process everything. It can’t. Instead, evolution built attention mechanisms in the way of sophisticated filters that decide what’s signal and what’s noise, what gets through to conscious processing, and what stays in the background. Daniel Kahneman called this our attention budget. Donald Hoffman went further: evolution shaped perception not for truth, but for fitness. We notice what helps us survive, not necessarily what’s real.When you cross a busy street, your attention system amplifies movement in your peripheral vision while suppressing the conversation you were having. When you search for your keys, it highlights small metallic objects while dimming everything else. This filtering isn’t passive selection. It’s active suppression. Your brain constantly inhibits irrelevant information, not just spotlighting what matters. And here’s the critical part: attention is metabolically expensive. The prefrontal circuits that control attention consume a disproportionately large amount of energy relative to their size. Evolution doesn’t invest in costly mechanisms unless they are essential.Why? Because unlimited processing would be catastrophic. An organism that tried to consciously attend to everything: every sound, every shadow, every internal sensation, would be paralyzed. Or eaten.Attention evolved not despite limited capacity, but because of it. The constraint is the feature.The Cost We’re PayingOur attention mechanisms evolved in an environment of scarcity: limited threats, limited opportunities, limited information. We now live in the opposite. Social-media platforms employ thousands of engineers whose sole job is to capture and hold your attention. Notification systems exploit the same dopamine circuits that once kept us alert to predators. News feeds leverage novelty bias, the exact mechanism that made our ancestors notice rustling grass. These aren’t accidents. They’re attention-extraction systems, engineered to hijack circuits that can’t tell a status update from a threat.The result? We train our attention systems on trivia while complex, important problems go unexamined. We build neural pathways for distraction and atrophy the circuits for sustained focus, and it’s getting worse. The average person now switches tasks every three minutes. Deep work involving the sustained attention required for learning, creativity, and wisdom is becoming increasingly rare.This matters because attention isn’t just about momentary focus. What we attend to literally shapes our neural architecture. The brain is plastic. It rewires based on use. Neurons that fire together wire together. If you spend hours scrolling, you’re training your brain for scrolling. If you spend hours in deep focus, you’re building capacity for deep focus. One of the many marvels of the human brain is that you can rewire it based on your own decisions and actions. Attention isn’t neutral. It’s formative. What we choose to notice reveals what we value. To attend to another person’s pain is compassion. To ignore it is indifference. Our repeated choices literally wire who we become.What This Means for AIHere’s where it gets interesting. Both biological and artificial intelligence face the same fundamental constraint: limited attention in an unbounded information space. The current AI race assumes more compute solves everything: bigger models, more parameters, more training data. But we’re already seeing diminishing returns. GPT-N is better than GPT-3, but not proportionally better given the increase in compute.Simply scaling up without selective focus is like trying to live wisely by doing everything at the same time. We can add data, energy, and speed, but without direction, we only amplify noise. Busyness is not mastery. Efficiency is not intelligence. Why? Because the problem isn’t raw processing power. It’s attention allocation. A system that processes everything equally is wasteful. A system that can’t prioritize can’t reason about complex, multi-step problems. A system without an attention architecture can’t adapt to novelty. The transformer breakthrough worked precisely because it imposed attention constraints. The next breakthrough won’t come from removing those constraints. It will come from making them smarter.This is why I focus on AHI (Augmented Human Intelligence) rather than AGI (Artificial General Intelligence). The current approach of building monolithic models that try to do everything is architecturally flawed. It’s the AI equivalent of trying to make a brain with more neurons but no specialized regions. Better AI means modular systems where specialized components handle specific tasks, coordinated by attention mechanisms that route problems to the right processors and integrate results just like the brain. Most importantly, these systems should augment human attention, not compete for it.Most current AI systems are attention thieves. They demand that we read their outputs, verify their results, prompt them effectively, and manage their limitations. The result is that we are spending a lot of time reviewing, highlighting their use, and scrutinizing their output, when what we really want is for them to provide us with the relevant information in a compressed form, so that we (humans) can synthesize and build. They fragment our focus across tools, screens, and contexts. We should be building attention amplifiers: systems that handle routine processing so humans can focus on judgment, strategy, and meaning. Systems that compress noise so we can attend to the signal.This is Augmented Human Intelligence: a hybrid architecture of awareness where humans bring values and judgment, and AI brings endurance and pattern recognition. When AI filters noise, humans focus on meaning. When humans define meaning, AI aligns its attention accordingly.Intelligence Is Attention Plus ArchitectureThe 2017 paper got it right, just incompletely.[^1]Attention is all we have, all we need, but only if we understand what attention actually is.It’s not merely a technical mechanism for weighting tokens in a sequence. It’s the fundamental principle that lets bounded systems operate in unbounded environments. It’s how intelligence navigates the gap between infinite possibility and finite capacity. For AI, this means a shift away from brute-force scaling and more towards architectures that allocate attention strategically, dynamically, and efficiently. For humans, it means recognizing that our attention is both our most valuable resource and our most vulnerable point of influence. We can’t opt out of the attention economy, but we can build better defenses and invest in better tools.That morning two years ago, when my attention had narrowed to a single desperate loop, I was experiencing exactly what I’ve described here, a system under stress, defaulting to survival mode, losing the capacity for strategic allocation. Recognizing this led to the Mastery of Life framework: treating fulfillment as an optimization problem where competing domains (health, relationships, growth, contribution) vie for finite attention. Just as the transformer learns to assign attention weights to data, we can learn to assign value weights to experience. We’ll explore this in more detail in Week 7.The question isn’t whether we have enough attention to solve our problems. We never will.The question is whether we can build systems, both technological and social, that help us attend to what matters.That’s the architecture challenge of our generation.---Next week: How modular systems coordinate attention across specialized components, and why this matters for both AI safety and capability.[^1]: Vaswani, A., et al. (2017). Attention Is All You Need. Advances in Neural Information Processing Systems, 30. arxiv.org/abs/1706.03762For a deeper exploration of how attention mechanisms constrain and enable intelligence across biological and artificial systems, see my whitepaper: Attention Is All We Have.This is part of an eight-week series exploring how biological principles can inform better AI architecture.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

A recent post of mine on LinkedIn calculated something staggering: you could train GPT-4 thirty million times over with the energy evolution “spent” architecting the human brain.Thirty million times.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work.That number stopped me cold. Not because it’s precise, it’s a thought experiment, after all, but what it reveals is important: the brain’s magic isn’t brute force. It’s architecture.After 18 months studying cognitive science, neuroscience, and evolutionary biology, I’ve come to see something remarkable. We’re living in a moment where two forms of intelligence are meeting for the first time: one built from carbon and refined over four billion years, the other built from silicon and refined over a few decades.Both are extraordinary. But the real question isn’t which will win. It’s what they can teach each other about how intelligence actually works.The 30-Million-Fold Efficiency GapLet me back up to that energy calculation. I compared the total energy consumed by all human brains that have ever existed, what I call the Evolutionary Processing Unit (EPU), to the energy used to train a single GPT-4:- The EPU’s energy bill**: ~1.4 million TWh- GPT-4’s training cost**: ~50 GWhThe difference is a factor of approximately 30 million.This isn’t just big. It’s physically insurmountable for our current approach. You could train GPT-4 30 million times for the energy evolution spent optimizing the brain’s architecture.Which raises an obvious question: if intelligence were just about computational scale, wouldn’t we be closer to AGI by now? The persistence of AI’s limitations, common sense reasoning, causal understanding, continuous adaptation, suggests we’re missing something fundamental.Are we racing ahead with our silicon brains, or are we circling an older truth that evolution already solved?Evolution’s Four-Billion-Year R&D ProjectHere’s what I’ve learned: evolution didn’t create intelligence through raw computational power. It created intelligence through architectural innovation, refined across billions of years of trial and error.The human brain, what I call the Biological Processing Unit (BPU), is the product of that optimization process. Every species, every neuron, every failure was a data point in a computation so immense that even our fastest supercomputer, operating at 1.7 exaFLOPS, would need trillions of years to replicate it.And what did that process discover? Not a bigger processor. A better design.The Brain as Confederation, Not MonolithThe most striking thing I’ve learned about the brain’s architecture is that it isn’t a single, unified processor. It’s a confederation of specialized systems, each optimized for specific functions:- Sensory systems that filter and prioritize incoming information- Memory systems that consolidate and retrieve experience- Emotional systems that assess value and urgency- Motor systems that coordinate action- Executive function (the prefrontal cortex) that orchestrates these modules toward coherent goalsThis isn’t just division of labor. It’s a fundamentally different architectural principle than the monolithic models dominating AI today.Think about it: when you decide whether to accept a job offer, you’re not running a single massive calculation. Different parts of your brain are processing in parallel, one assessing financial implications, another evaluating social fit, another imagining future scenarios, while your executive function weighs these inputs against your values and makes a decision.This modular architecture provides several critical advantages:Specialization without brittleness: Each module can be optimized for its specific function without compromising the system’s overall flexibility. Your visual cortex is exquisitely tuned for pattern recognition, while your hippocampus specializes in episodic memory. Neither has to be good at the other’s job.Graceful degradation: When one module is damaged or overwhelmed, others can partially compensate. The brain’s wrinkled outer layer, the cerebral cortex, exhibits remarkable plasticity. When areas responsible for vision are damaged, neighboring regions can gradually take over visual processing functions. Stroke patients can sometimes relearn speech or movement as undamaged cortical areas rewire themselves to handle these tasks.Efficient resource allocation: Not every module needs to be “on” at full capacity all the time. Your brain dynamically allocates attention and energy based on context. Walking down a familiar street requires minimal conscious processing; walking down a dark alley in an unfamiliar city activates multiple systems simultaneously.Continuous learning at multiple scales: Different modules can update at different rates. Your motor cortex might refine a tennis serve over weeks of practice while your prefrontal cortex simultaneously updates its model of a colleague’s reliability based on a single conversation.The Coordination ChallengeBut here’s where it gets really interesting: having specialized modules isn’t enough. They need to work together.The prefrontal cortex serves as a dynamic orchestrator, not a top-down dictator, but rather like a conductor who knows which sections of the orchestra to bring forward at different moments. This coordination itself is learned and plastic, adapting based on experience and context.When you’re driving and a child runs into the street, your brain doesn’t deliberate. Visual cortex detects motion, emotional centers flag threat, motor systems execute braking, all coordinated so rapidly it feels instantaneous. That coordination is itself a learned pattern, refined through evolution and individual experience.What Current AI Architectures MissMost large language models are monolithic. They excel at the tasks they’re trained on, but they can’t easily:- Reason causally about interventions and counterfactuals- Learn continuously without catastrophic forgetting- Allocate resources dynamically based on problem difficulty- Explain their reasoning in terms humans can inspect and trust- Adapt strategies when operating outside their training distributionThese aren’t just implementation details. They’re symptoms of a fundamental architectural mismatch.The brain evolved these capabilities because it had to. Evolution optimized under severe resource constraints: limited energy, limited space, limited time to learn before predators struck. Monolithic processing wasn’t an option. Modularity wasn’t a design choice; it was survival.The Human Angle: Beyond ProcessingWhat makes this even more interesting is that the BPU doesn’t just process, it cares. It wonders. It gets curious.Curiosity, imagination, surprise, these aren’t inefficiencies that evolution failed to optimize away. They’re features, born to navigate uncertainty. Human intelligence thrives on ambiguity. We value metaphor and story because they compress complexity into meaning.AI can generate those patterns now, but it doesn’t yet understand them the way we do, because understanding requires something the BPU developed through embodied interaction with the world: causal reasoning.This is what Judea Pearl calls climbing the “ladder of causation”:- Seeing: observing correlations in data- Doing: understanding how interventions change outcomes - Imagining: reasoning about counterfactuals, about “what if”Current LLMs are remarkably good at the first rung. They excel at seeing patterns. But they lack the innate scaffolding for doing and imagining that the EPU built into the BPU through four billion years of embodied interaction with reality.A Different Path ForwardThis is why the 30-million-fold energy gap matters. It’s not just about efficiency, though as AI energy consumption threatens to overwhelm power grids, efficiency certainly matters. It’s about what that efficiency reveals: intelligence emerges from structure, not just scale.What if we built AI systems more like the brain? Not by slavishly copying every detail, we’re engineers, not neuroscientists, but by adopting the architectural principles evolution discovered:- Specialized modules for perception, memory, causal reasoning, and value assessment- Dynamic orchestration that learns coordination strategies- Continuous plasticity at multiple timescales- Embodied grounding in action and consequence- Resource constraints as design features, not bugsThis isn’t just theoretical. The research directions are clear: modular architectures, causal inference frameworks, continual learning systems, attention mechanisms that actually allocate scarce resources rather than just weighting inputs.The Recursive LoopIn a sense, the most promising path to artificial intelligence isn’t linear at all. It’s recursive, a feedback loop between two learning systems: one biological, one artificial.The BPU shows how to build intelligence under constraint. The GPU shows what’s possible with abundance. The question is whether we can learn from both, whether we can build systems that combine the architectural wisdom of evolution with the computational power of modern hardware.The goal isn’t to replicate the human mind. It’s to learn from it wisely. If the BPU is the teacher and silicon is the student, then the real lesson is humility.Intelligence grows not by scaling alone, but by listening, to the feedback of reality, to the signals of constraint, and to the wisdom evolution already embedded in us.What’s NextHere’s what humbles me most about this journey: the more I learn about the brain, the more in awe I am. Not as mysticism, but as engineering.Evolution ran a four-billion-year experiment with trillions of parallel trials, each one literally life-or-death. The BPU is the result: an architecture so elegant, so efficient, so robust that we’re only beginning to understand its principles.The path to AGI won’t be found by training ever-larger monolithic models. It will be found by learning from the master architect: evolution itself.Next week, I’ll explore attention, both as a technical mechanism in AI and as the human capacity that makes all learning, all understanding, all wisdom possible. Because if modularity is the brain’s architecture, attention is its operating system.---Notes & Further ReadingThe 30-million-fold energy efficiency calculation is detailed in my LinkedIn post from November 5, 2025.For the complete technical argument on evolutionary computation and modular AI architectures, see my research paper “Beyond FLOPS: The Evolutionary Processing Unit and the Roadmap to AGI” on my website.The foundational arguments about the ingredients of human intelligence appear in my earlier post, “AGI - The Human Angle”.Thanks for reading James Maconochie | Architecture & Attention! Subscribe for free to receive new posts and support my work. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com

My background is in civil engineering, a field that learned centuries ago that we build better structures through smarter architecture, not just more material. Suspension bridges span farther not because we found stronger stone, but because we discovered smarter ways to arrange materials, to let structure, not mass, do the work.I see a similar inflection point in AI today.The dominant approach assumes intelligence emerges from scale: more parameters, more data, more compute. But biological intelligence didn’t evolve that way, and I don’t believe artificial intelligence will, either.The brain isn’t a monolithic processor that simply grew larger. It’s a distributed system of specialized modules, each optimized for a specific task, like sensory integration, causal reasoning, or emotional valuation. The true magic lies not in raw processing, but in coordination and feedback, how these components communicate, prioritize, and adapt.That’s why I’m exploring what I call the architecture of attention: the design principles that let intelligence focus, reason, and adapt based on what matters in the moment.---What I Bring to ThisI’ve spent 25 years working at the intersection of complex systems and human needs, from modernizing state credentialing platforms to scaling genomic data systems. I studied civil engineering at Imperial College London and MIT, disciplines that teach how structures bear load, how systems fail, and how to build things that last.Over the past year, I’ve been researching AI architecture, human cognition, and evolutionary biology. I’ve developed frameworks for how biological processing principles might inform more capable and aligned AI systems.That work isn’t ready for academic publication, but it’s too important to keep in a drawer.So, I’m bringing it here to share progress in public, invite critique, and connect with others who are asking similar questions.---What to ExpectThis isn’t a newsletter about AI hype or doom. It’s about architecture, how we build intelligent systems that can actually reason, adapt, and enhance human capability.A critical distinction: I’m focused on AHI (Augmented Human Intelligence) rather than AGI (Artificial General Intelligence). This isn’t just semantic, it’s architectural.The current race toward scaled-up LLMs serves the interests of chip manufacturers and cloud providers more than it serves actual capability advancement. These systems are extraordinary, but they’re now colliding with architectural limits that more compute alone can’t overcome.The alternative isn’t to abandon AI, it’s to rethink it. To build systems designed from the ground up to enhance human judgment rather than replace it. This modular approach combines specialized AI capabilities with human reasoning, attention, and values. It not only produces better outcomes, but it also dramatically lowers infrastructure barriers, making sophisticated AI accessible beyond the handful of organizations that can afford multimillion-dollar training runs.---Here’s what you can expect:The Technical: Deep dives into modular AI architectures, attention mechanisms, and system design. The Philosophical: What we mean by understanding, agency, and value alignment. The Practical: How organizations can implement responsible and effective AI systems today.The through-line is systems thinking: recognizing patterns across scales, understanding how constraints shape solutions, and asking not just “can we build this?” but “should we, and if so, how?”I’ll draw on evolutionary biology, cognitive neuroscience, economics, and infrastructure engineering. If that sounds scattered, I’d argue it’s necessary; intelligence itself is an integration problem.---An InvitationI’m writing for researchers exploring alternatives to pure scaling, for leaders navigating AI implementation, for policymakers crafting sensible regulation, and for anyone curious about building technology that genuinely serves humanity.I don’t have all the answers. But I have frameworks worth testing, questions worth asking, and 25 years of experience building complex systems that actually work.If that sounds interesting, welcome. Let’s explore how to build intelligence, artificial and human, that truly serves understanding.---This newsletter explores how we can move beyond brute compute to create systems that think, adapt, and collaborate with us. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jamesmaconochie.substack.com