0042: Listener Feedback

Welcome to Concerning AI, this is Ted. And this is Brandon, this is the show where we have two questions. Yeah, is there an existential threat from AI for humanity? Yes.

I think we have the answer is yes. And if so, what should we do about it? And it seems like we are learning some decent answers to that from other people, which is exciting because at the start, we didn't have any hope really. Yeah, one potential answer is to go as fast as we can now.

Yeah, so that we don't end up with a giant hardware overhang, that means a lot of computing cycles that are available for the general intelligence when it finally reaches our level and it can just boom, zoom out to fill that up and get way ahead of us quick. Yep. And the other one is perhaps augmentation. Perhaps a brain computer interface could be a help.

So tell us, stay ahead of the Singularity. Yeah, and I'm reading a series of books right now, which I highly recommend to you, Ted, by Rheemez Nam called the Nexus Trilogy. And it's a pretty good read and it looks a kind of envision a future in which people and groups of people become the super intelligence, which I don't know, I kind of like it. Yeah, I don't know how real it is and stuff, but it's a good read.

I totally recommend it. All right, well, if I get to that, then maybe that'll be fodder for a future show. Possibly. It's all right, very good.

In the meantime, we have a lot of listener feedback. And by the way, if you would like to send us feedback, the best way to do that is to put it in a voice memo and email it to concerningaiongmail.com. And that's what Birdie did. And I'm going to play Birdie's voice mail now.

Hi, I'm Bertie. I'm a long-time listener and big fan of the podcast. I just listened to your Neuralink episode. I'm very glad you're covering it.

Hopefully you'll continue to talk about it. What I want to talk about though is the boring company. Your negative comments about it, not saying it's terribly important, maybe even a bit boring. I mean, clearly, Tesla, SpaceX, Neuralink are more obviously important.

But I don't think we should totally discount the boring company. So I'll try to keep this in terms of AI and your podcast. So the two things that you guys have been mentioning as a possible way of attacking with a problem are go fast on artificial intelligence and brain machine interfaces. To do this, we need as many people as possible, spending as much of their time as possible, working on improving AI and BI, meaning biological intelligence.

So just to talk about this in terms of human clusters, to Mervyn's term for human as a whole, because this is an important point. It's not necessarily humans individually, but a smart. Or if they are individually smart, it's because of many billions of humans over many years collectively, progressively becoming smart. And so we spend a huge amount of time stuck in traffic.

And in the process, filling the sifties, which are filled with many smart people that ought to be working on these problems with a position instead of, you know, stuff that actually harms our brains. I think it's sort of, hopefully, when I'm trying to get up. But what I'm saying is that if we spent less time stuck in traffic, maybe we'd have more time to go fast and work on brain machine interfaces. I mean, sure, not everyone's going to be working specifically on these things, but with everything that's happening at the moment and that could happen in the future, we need people to be spending their time easily, not staring angrily at the back of a car in front of them.

Anyway, I'll stop going on. I have it. Cheers. Thanks so much for sending the voice of a birdie.

I love it. I love it. What's your initial thought about this? Well, birdie sounds really smart.

And I think that it's probably because birdie is smart, but I think he also gets the benefit of the accent. The accent is awesome. I love it. I wish we Americans love our accents, too.

Yeah, we love. Maybe it's us that have the accent. But regardless, the way that you talk sounds great. I like it.

Yeah. So yeah, my answer to being stuck in traffic is to move close to where you need to spend your time. And that's what I've done here. I live very close to my kid's school.

And then I can write a master answer to my work and work on the bus in the way. And so I am not that impressed with the importance of the problem because I think it has an easy solution that doesn't require exotic technology. And it doesn't require people who could be focused on other things to attack it. So I don't know.

How about you, Ted? What's your take? Yeah. So the reason that I was so negative about this on the last show was that it just seemed so absurd for Elon to be spending any of his time on what seems like a problem that doesn't give much of benefit.

So I love that Birdie gives us a benefit. Birdie gives my brain input that it wasn't going to generate on its own. I'm not open minded enough to look beyond my own initial reaction. So I really value that.

I have since seen Elon tweeting about the boring company quite a bit. And it really bothers me. It makes me angry. So I'm still not swayed by Birdie's argument.

People sitting in traffic. Yeah. I mean, I guess that is a real thing. I think I'm also influenced by a colleague of mine who has been really hating on Elon Musk every time I got to lunch with him.

You got to lunch with Elon Musk? That's amazing. Yes. Every time I got to eat with Elon Musk.

So my colleague has been hating on Elon Musk in my options, I guess, are to dig my heels in and say, Elon is great no matter what, or to prove that I'm not a fanboy, and that he is objectively great in certain ways by hating on him in other ways. So hating on the boring company is a great way for me to prove that I'm not a fanboy. So you get benefit both in both directions. Really, if you dig your heels in, then you get to support Elon, and that really supports our brain biases, and then you get to have the bias in the other direction.

So you win either way. Yeah. I'm not sure I quite understood that. But I like the term win either way.

I'm a phrasey, win either way. All right. Yeah. So let's move on.

We've got another voice mail about the boring company from Eric. Ready for that one? There we go. Yeah.

Yeah. And then I have an idea about what we should learn from these. OK. There we go.

All right. And hi, Ted. It's Eric. Yeah.

You guys are being doubting Thomas' about the boring company. How boring of you? So I think it is actually a transition for Hyperloop. The concept being that these Hyperloop pods, they're video from Hyperloop.

It shows these Hyperloop pods picking people up at their doorstep, taking them to the station. And they don't even have to get out of the Hyperloop pod. They just, the Hyperloop pod just boards the train with them in it. And at the other end, same thing.

It unboards the train or disemarks from the train and takes them right to the doorstep. And if you had lots of Hyperloop trains coming in and these little pods just flooding the roadway, it would obviously create horrible traffic jams. But if there was already a set of tunnels in place that had strangely not been used by Tesla cars for a long time or cars, then that would make Hyperloop just work that much better at each end. Eric.

I love it. Again, that people are thinking in more expansive ways. I mean, that's what we want. That's why we want voice memos.

It's hard for me to understand the whole pod idea. Maybe I don't know enough about the Hyperloop to know how that's supposed to work and what these pods are. I thought it was just like a train that you go to the station and you get in the vehicle and then it zooms you off. Whereas this tunnel thing, I don't know.

I mean, I guess that they could somehow be integrated with these two systems. But transportation on Earth, unless we're talking about SpaceX providing the transportation on Earth, I just can't find myself getting excited about it. It just feels too small. It feels too trivial.

What I can find myself getting excited about is the learning that we're getting from this. And you know what that learning is? No. Negative comments that we make on the show generate more feedback.

This is good for us to be hating on Elon or anything else. Yeah, I think that's true. Is that where internet trolling got started? Maybe so.

Maybe that is law number one of the internet. Something like that? Yeah. Haters going to hate.

Yeah. Nice to love Elon and thank you, Eric and Bertie, both very fun. Yeah. So should we go on to the next one?

Yeah. The next one is actually an email from Arjun that you're going to reach for us. All right. So this feedback is from a long time ago.

Is this from a long time ago? Or is it's just coming in new? It's just coming in new. But it's from an episode that we recorded a very long time ago.

This episode you're listening to right now is episode 0042. And the email Brandon's about to read comes in response to episode 0001. Ah. Awesome.

OK. So this is Arjun. Too concerning AI. This is a blast from the past.

Hello, guys. Great podcast. Just started episode one. Here's some feedback.

Intelligence behavior are closely related, but not the same in my view. The third component is sensory input systems. Then he split things up into a list of three. Because he's talking about three components.

One, sensory input systems of an organism are able to observe the universe around the organism and provide data about its state to the organism. Two, intelligence is the ability to use the sensory input, to create a model of the universe, to explain things, to predict, and come up with manipulations that are needed. I'm doing a little bit of summarization here. Three, manipulative systems of the organism actually manipulate the environment, moving stuff around, influencing the outcomes.

Consciousness is linked to a certain level of intelligence. And if you have a certain level of intelligence, it will automatically create a model of the organism, therefore consciousness. For Argent, it predicts that sentient AI will be the result. Once we create AI, it has all three of them of these important systems working.

And we start working on the semantic abilities of the AI. But he offers a nice word too. German has a nice word for understanding, which is beg griefen. Begriffen?

Beg griefen. Literally, this means using your hands on something. And this, number three, to allow an even faster modeling in the environment. If you can manipulate things, you can learn faster, and I think with reference to infants and stuff.

How do they learn? The manipulation is interesting, because the idea of being embodied is maybe a first step. The ability to manipulate would be maybe a further step. Yeah, I guess you could be passively embodied, I guess.

Is that kind of what you're saying? Self-driving car doesn't directly manipulate, except its own position. The world is a given to it, and it just moves around in it. It doesn't manipulate anything.

Yeah, I suppose that's true. Versus you think about the infant learning with the baby rattles that's bringing closer and further, and whacking itself in the head with, and checking, and sticking his mouth, and then dropping on the floor, and having the adult pick it up and give it back. Yeah, moving it from hand to hand, all that kind of stuff. Yeah.

OK. Personally, I think our efforts to create semantic AI by just focusing on number two, which is the creating model, will fail. Imagine a human brain that's never been given access to sensory systems and manipulative systems, apart from its ability to retext and write text. Will that brain be able to understand our world?

So, the purchase of the choir here, Arjun, I'm a big embodied in AI guy as well, and feel that there's a lot more under our scheme of consciousness than we think there is. And now comes. Just a note that this episode that Arjun's responding to was called, What is Intelligence? And then we went into consciousness on episode two.

So these ideas are very much connected with one another. But we were at the very beginning, we were just trying to understand, what is this? What is artificial intelligence? The idea of, is artificial intelligence a threat to humanity or an existential threat?

We had to understand the nature of intelligence, or we thought we did. We thought we might. Right. And I think separating intelligence and consciousness still makes sense to me.

I also think that it makes sense to pull out the dimensions of emotions and values, just like Arjun is doing here as well. There are different ways to slice and indice in the space of mind, I guess. And all these ways still make sense to me. But you did.

Now, incidentally, I think that the Tim Urban article about Neuralink that we talked about extensively on last show posits that the science may be not nearly as important as the engineering. If we're going to build a brain computer interface, we might be able to just go brute force, or however we do the engineering, and not understand what is the nature of intelligence? What is the nature of consciousness? What are emotions?

We might never answer those questions. We might be able to create the brain computer interface anyway. So these questions seemed really fundamental to me at the beginning. They seem less important to me now.

Yeah, they seem orthogonal to the brain computer interface, for sure. They may still be really important for creation of superhuman and artificial intelligence. But I'm not so sure. We might need to go revisit that question, actually.

Yeah, good call. I mean, if the superhuman intelligence is just an extension of human intelligence, then that doesn't require that the base be understood. That's right. Yeah.

Good. All right. Anything more on that? From that email, it's email is pretty long.

But anything more that you want to pull out from that one? There's one thing towards the end that he said, will we create AIs to die? It is probably necessary that we let AIs die as well. And then, but it's not hard to understand that the AIs may not feel that this is necessary.

Yeah. Yeah. Food for that. Food for that.

Well, why don't we just bleed right into the next email we got, which was actually from Arjun about that episode two. And again, it's pretty long. And there's a lot of overlap between the two. So I'm not going to read it all.

But let me pull a little bit out. Start. Great episode. I like that.

I like that as a beginning. Like if you guys want to get on the show, give me a compliment about how great we are, because we're human. That's a pretty good way to go, just as a tip. Yeah.

Although we have learned that people respond well to us when we hate on things. So who knows? Maybe we'll respond well to you if you hate on us. Well, that's true.

Both birdies are disagreed with us and their voice memos and they got on the show. So don't feel, try the experiment. See what you are. OK.

So Arjun says, I think emotions and consciousness have nothing to do with each other. Nothing. Wow. Yeah.

I wonder what we said on that episode too. Did we think that they were, I can't remember episode two. I can't remember what I had for breakfast. So all right.

If you had, if you had Nexus, you could just replay it. Oh man. I want it. Wizard hat.

Wizard hat. Pretty brand. Totally want it. All right.

Emotions are a relic from our evolutionary past. Evolution allowed emotions in our brains to come into existence, because having emotions proved to be an advantage for an organism. Emotions allowed organisms to quickly assess the situation as a threat or an opportunity to prepare for action. All of this happened on a non-conscious level.

So an organism can have emotions without being conscious. Totally makes sense to me. Consciousness comes like the definition that was quoted in the beginning of the episode with awareness of the environment. If an organism understands the status of the environment, it's placed in the environment.

It's conscious. But there's a thin line here. It is just not only just that. One could argue that a self-driving car is aware of its environment.

Perhaps it even knows its place in it. But we would not call a self-driving car conscious. So what is missing? I think what is missing is meaning.

A conscious being understands the meaning of the objects around him or her, understands the meaning of its own place in the environment, understands how things are connected and influence each other. Self-awareness is also important in this view. Do we call it an organism conscious when it is aware of its surroundings, understands the meaning of the objects around it and how they're connected, even if it fails to recognize itself in a mirror? So I think there's a lot of really interesting questions about consciousness that we may have said some things and we may have just forgotten everything we said about it back in the day here.

Meaning, meaning that's a loaded one. Yeah. Yeah, meaning as if there is such a thing. Meaning.

Meaning. Meaning. Yeah, I don't know about that. But that's why I'm big into the embodied AI.

Because there is some kind of grounding of the symbols and something that's more than symbolic. So I don't know if I would call that meaning necessarily. Maybe. Because meaning often we often overload meaning to mean ultimate purpose and the greater reason for existence and our place in it and why we do what we do and why we should do what we do.

And so yeah, it depends on kind of which of those directions you're taking meaning. Yeah. I agree. What did you think?

Where did you think Arjun was going with that? Well, I don't know. I think the human beings appear to me to make meaning out of everything. We make up meanings for everything that don't exist without us, without us making them up.

So we make up reasons why things happened that may or may not. Yeah, we make up reasons. Causal chains. Yeah, I'm walking through the train station.

And somebody has an expression on their face, and I make it mean that they don't like me. Or I look at my fly to make sure it's not open. And they may not have anything to do with me. And maybe they're thinking about their own day.

We make things up. So all right. Well, Arjun also talks about the idea of language on the episode, apparently. I don't remember this, of course.

But it was concluded that language was an important step. And Arjun says, I think that is not entirely true. When a dog barks, that is also language. When a bee dances around to tell its hive members where the flowers can be found, that is also language.

But what is missing is language to explain things in a logical way. Logic and a language to store logic is a very important step in consciousness because logic helps us understand the environment much better and with richer details. Well, I think in light of the wizard hat, this is really interesting. Because if we can communicate emotions directly without having to go through words, which are such a blunt instrument, we may have a different feeling about.

The important language is. Exactly. Now, maybe we had to have language in order to get to the step of the wizard hat. But when does the wizard hat get installed?

Does it get installed upon birth in the future? Right? When we have the wizard hat? Do babies get the wizard hat before?

What about the wizard hat before? They've learned language before birth. Oh, before birth. Yeah, like when it just grow.

And you're blood stream. And it's all depending on what the mechanism of the sub-semit mechanism is. That's right. Maybe it comes right through the umbilical cord.

Yeah. Just comes in the blood stream from your mother since your mother has this stuff already in her. Yeah. Whew.

All right. So then, you know, there's a bunch more stuff here. And I'm sure it would be fascinating to have a whole other episode on the rest of the audience email. But I don't remember the episode enough to continue.

Yeah. Let's move on. Yeah. Yeah.

So, Arjun, thank you. I loved it. Not a comment about you at all. Just my attention span really is what that's about.

Right. We've been a group behind. The Brad, the Trandon, the super intelligence that comes together when we put our heads together. That's right.

Does not have any longer attention span than this constituent part, too, apparently. That's right. All right. So moving on to Norbert's voice memo.

Norbert had a different approach. And I'm going to let him tell us about his approach. Here we go. Whoa.

It's in Brandon. Here's Norbert. I've got a bunch of comments to make about the latest podcast. And my thinking is to record them as individual sentences or ideas.

And then you can respond to each one or whatever. I don't know. But so I'll make a bunch of recordings and I'll number them. OK.

Thanks, Arjun. You know what? You know what? I used to have a website called ThoughtChunks.

Yeah. And it was a wiki. And it was all about trying to distill individual ideas down to their bare essence. And then you'd make tours through them with connective sentences.

That's basically what an essay would be. Yeah. A tour through the thought chunk. So this is, yeah, I like it.

Let's see how it goes. Yeah. So we'll play some of these. And we'll probably respond to some and not to others.

And here we go. First of all, on the most general level, I think your approach is way too abstract. Your argumentation tries to cover all bases with one or just a few principles. And that makes it unpractical, in my opinion.

I don't have anything more specific to say about that. But that's just the most general idea. It's a pretty sweeping criticism. It gets right on the show.

I kind of like it. Yeah. If you want to make that shorter on the next batch of these, you can just say, you suck. Or I think that's probably true.

Yeah. All right. Let's go on to the next one. Another more to say about that.

I like Joyce's comments with respect to utility. And I also feel that the utility function is a central concern to the thinking of the podcast. And utility is from the field of economics. And a lot of the important ideas discussed are also from the field of economics.

And you have avoided these economic terms, I feel. And since there is a long tradition in economics about goals and utility and a lot of these terms, I see no reason to ignore this long tradition, which goes over several centuries already. And I want to dive into that a little bit. So let me see if I can summarize that.

Yeah. Yeah. Economics is a tradition with a long and glorious history. Stop dissing it.

Sounds good. All right. Let's see what else over here. Let's see what else to say here.

Yep. First and foremost, let's stop talking about paper clips. There's no reason to maximize paper clips. Can we all agree on that?

In the field of economics, what we want to maximize is profit. And well, there's also a maximization of utility. But in real world terms, the maximization of profit is a very clear, short, medium, and even long-term goal. Yeah, boy.

I'm having a hard time connecting the profit motive to the AI. The reason we talk about paper clips is because we have to give our AI a problem that it can understand. And maximizing profit is too abstract. It can accuse us of being too abstract.

But I think profit is too abstract in the short term. Yeah, it doesn't know anything about money or anything else. So we have to give it toy problems. And maximizing paper clips is a toy problem, which is why we keep talking about it.

If a toy problem could be dangerous, we have a big problem. Because that's where we always start with. We always start with toy problems. We want to maximize profits, but we've got to start with something that is actually within our capabilities.

An example of this is whenever you're making a business plan, any size of operation, any size of concern has the top line and the bottom line that they're trying to maximize. But it's too abstract even for large companies, even for startups. So you always end up choosing instrumental goals that are things that you can actually definitely achieve, and that you can definitely influence that you believe will move you in that direction. And those become the thing you focus on in the short term.

Like monthly active users, for example, for talking about a web startup, it could be a whole bunch of other goals. So that's an example of if even if your company is trying to maximize profits, when you go to create your learning system, it would be very unlikely that maximizing profits is the goal that you would build into that thing. Again, because it's just too abstract, but it might be that your goal is to maximize. Like you just said, the number of signups that you get for a website.

Or maybe it's how much engagement each of the people who signs up on your website does. I'm sticking with the website thing. Or maybe it's the number of referrals that you get per person who signs up. And each of those is too abstract.

So then you bring up, you do reductionist. Yeah, I think if we think of AI like a tool, it's even easier to think about. If I'm going to make a new hammer, then my metric is not profit. My metric is how many nails do it drive per minute or per hour or whatever.

I'm going to look at the quality of the work that the tool does, on the task that I'm giving it to do. That's what I'm going to measure. And yes, I'm building a new tool so that I can make more profit. But that's an outcome of driving more nails more efficiently.

Yeah, I think we're going on and on about this. But you definitely did stir us up. OK, let's go on. Well played, Norbert, well played.

It's important to note that computer algorithms, programs, et cetera, are entirely indifferent to what the outcome for humans might be. So whether the outcome is zero or plus 1 billion or minus 1 billion, the computers couldn't care less. I think this is central. This is central to why there is an existential threat from AI to humanity, right?

I mean, that sums it up. Yeah, they have the thing that they care about in an abstract way, caring not in the human sense but their utility function, whatever it is that they are optimizing toward, the algorithm is optimizing for. That's it. That's all there is.

Yep. All right, so Brandon, you told us about the books you're reading. And I just picked up a book, which I can't remember the name of, but it was written in 2005. And I will put the link in the show notes for what it's called.

2005, so pretty old, right? 12 years old. And it is about the evolution of humanity. It's about how we are choosing our own evolution right now.

So it feels like it's talking about all the things that are going to lead up to neural link, but it doesn't know that yet, right? It's a certain 12 years ago. It's talking about telekinetic monkeys that they had these monkeys implanted with whatever sensors. And the monkeys could figure out how to move the robot arm.

And so I'm reading it, even though it's old, I'm reading it because I'm fascinated by how we used to think about this stuff not too long ago, 12 years ago. And I'm not that far into it. So I'll have more to report back on it. I just find it fascinating to think of the progression of not only evolution, but then human technological advancements.

And it's very difficult to understand how far away or how close any of this stuff could be because we're all going to wait for our flying cars. Well, you know what really helps with this sort of thing is to go back and watch a movie from not too long ago that was awesome, like a science fiction movie that was awesome. And an example of that is I went back and I watched part of the Matrix. Just I don't know, I was on Amazon Prime or something and I saw it there.

And I was like, oh, I wonder how that's held up. It's held up pretty well. But whoa, the technology that they're showing is cutting edge with their flip phones. And it is amazing what we take for granted now.

Yeah. One of the things in this book is this was pretty iPhone, right? So it started to talk about how we are already sideboards in 2005 before we had touch screens before we had touch interfaces. I was talking about the Blackberry.

And how did it take? My seven-year-old has been just really wanting a camera, really, really wanting a camera. And we have one. So we've let him start using it.

It is cutting edge. It is totally cutting edge from 2005-ish. So it's a Canon digital L. So it's not cutting edge.

But every time he's trying to work with that thing, he's always touching the screen. Trying to make the screen work. No, it's not a touch screen. It's not a touch screen.

You've got to push these little buttons. Yeah. Yeah. Any screen that you can't touch is broken.

It's broken. Yeah. My kids with the laptop, they're just boom, boom. Yeah.

Every time they're trying to touch the screen. Yeah. All right. Well, let's wrap it up.

Thank you everybody for your feedback. We'd love to hear your voices. You can now start to get more of a feeling of what we like to hear on the show. So send us more of that.

InsertingAI at gmail.com is our email address. Just a type of voice memo. It worked better than email because when you email us, you tend to go on and on to the honest. All of you with the internet.

It's so true. But you know that you're recording this for playing on the show. You're going to edit yourself. You're going to keep it shorter.

So please, go ahead and do that. Save us the work. It makes it a lot easier. And we'd love to hear that.

Also, there's lots of activity in our Facebook group. It's concerning AI. You just search for that on Facebook. I'll put the link in the show notes for that too.

Lots of interesting articles that people are posting, not necessarily related to the existential threat of AI, but a lot of it related to what's happening currently in AI. So there's a lot of interesting stuff going on there. We'd love to have your contributions there. And any last words before we sign off, Brandon?

Yeah, this is coming out of left field, but I think we should try to pick an event and go to it. Like maybe meet some folks face-to-face. And by an event, I mean, there's been a number of workshops and other kind of concerning AI conference type events. Wouldn't it be fun to go to one of those and connect with some people in real life?

So if you're going to any of those and you have a suggestion for where we could go and meet you, let us know. Let us know by sending us a voice memo or dropping it in the Facebook group. I think that would be really fun. So cool, I get it.

All right, so this is it, signing off. And Brandon also signing off with my click-on. Awesome, keep your Wuffy up. Keep your Wuffy up.