Hello, this is Ted and this is Brandon. Welcome to Concerning AI. Yeah, the podcast where we are concerned. We sure are.
And sometimes even scared. Sometimes scared, but Ted assures me that it's gonna be okay. Well, tonight, he says he has an outline for tonight's show and then it's gonna be okay. Yeah, I think he really means he has an outline for AI safety.
It's gonna be okay. And should we talk about that one thing? Yeah, let's talk about that one thing. Let's start with that one thing.
I've kind of been wondering if, for my own, I've been feeling like I need to sort of do something more than just the show and more than just try to get my work done and stuff. And, but yeah, the question of futility has come up with me, yeah, I mean, like we might be doomed anyway. Yeah, like what if we run, if we had an oracle, like a glass, glass balls, crystal balls, there we go. Yeah, what if we had a crystal ball and we could run the experiment a thousand times and every time it comes out the same, you know, maybe it's just sort of like almost a foregone conclusion.
So it's like the human experiment, the human creation of AI experiment and the intelligence explosion happens and humanity's wiped out and it's not good for us. 1,000 times out of 1,000 times out of 1,000 times. Okay, so if that were the case, so if that were the case, what should we do? And because I kind of feel like it, I kind of feel that way a little bit like, my answer is going to come in a minute.
Yeah. But I kind of feel like borrowing a metaphor that I got from Nick Bostrom in the Superintelligence book, where he says maybe people are really like, and he's talking about philosophy actually. You know, maybe people are just barely able to do it, just barely have what we need to be able to do that activity at all. And so we're like dogs walking on their hind legs.
You can train them, you can kind of like, they can do it because they have to just above the threshold for being able to walk on their hind legs. But nobody would say that they are good at that. So the idea that we were even calling ourselves intelligent today is a little bit folly. It might be a lot like that.
Yeah. Yeah. And I kind of, I've really been feeling that a lot. Yeah.
All right, so if we're in the experiment a thousand times, and we know that it comes out a thousand times bad for humanity. Yeah. Well, we've all watched movies, right? We've all watched TV.
What do you do in that situation? You have to keep going anyway. You have to. That's what I think.
Otherwise it doesn't make a very good movie. Well, yeah. So that my answer is slightly different than that. All right.
My answer is that we still got to try. And the reason we got to try is because I like us better, if we try. You like us better, if we try. It's not about the numbers.
It's not about what if it were 10,000. If it was a thousand of a thousand, maybe we'd ran at 10,000, it would come out OK. It's not that answer. No, it's not that it's because there's some utility in it.
Some tiny little loophole that we're going to exploit. No, it's more. I just read this science fiction by a Chinese author. I can never pronounce the name.
But I can tell you the trilogy. It's the three-body problem, the dark forest. Yeah, we link to them in the show notes from the last episode. Oh, did I talk about it on the last episode?
Oh, man. Yeah, we love it. You love it. I think I read the book since then, the third book.
Yeah, OK. Anyway, I'll link to it again. I mean, that's the problem. The important thing, the important thing here is that there's just a slightly different sensibility in the book coming from the Chinese worldview versus.
OK, so can you describe it? What's different? There's just a lot. It's just a little.
It's not even a huge difference. It's just like a subtle difference. I can't even really probably describe it accurately. But it has to do a little bit with us not being in control.
The Western narrative is this hero's quest where you win and you're powerful. And the hero figures it out and makes it all work. And there are heroic acts. There are heroic characters.
But yet, there's this sense that kind of pervades it all that what's going to happen is kind of going to happen. It happens. And we're all kind of palms in this larger unfolding that's not in our control. And some terrible things happen in it.
Like, I'm not going to give it away. But some things that I don't like happen. And it's not that those are pointed out as terrible things. They're just sort of like they are what happens.
They're not there's a little bit more of a freedom from attachment to a particular outcome. And it's a beautiful, beautiful set of stories. So one of the things is, does humanity leave our mark in the universe as a beautiful race, a beautiful species or not? And to me, we're a lot more beautiful if we try.
If we try to preserve humanity, if we try. Well, that all works. If we try to be better? If we try.
What are we trying? Maybe that's the time to bring up the upwards viral. Yeah. So there is a work which I will link to in the show notes called the upward spiral by our friend Paul Craphel.
Good, Craphel. Craphel. He has become our friend since we saw the work. Since we saw him out after this.
Because he's amazing. He's wise. He's a wise and amazing human being. And some of the, we won't tell the story well.
So you can go watch the movie and watch it looking for the wisdom, not for any other reason. Yeah, the production value on the movie. We don't need to talk about that. It's not.
The ideas are amazing. The ideas are amazing. So what is the upward spiral? So imagine the time before there's any life on Earth, right?
And there is just bedrock, exposed bedrock. And the rains fall. And through the freezing and thawing, some rock flakes, come off and they are washed down the slope. And they come to a place where it flattens out.
And those rock flakes, they accumulate there. Because that's just as far as they need to fall and they get stuck there. And the water accumulates there, too. And because there are rock flakes, there's more water that accumulates.
And because of that, the moss can begin to grow. Or whatever the premoss is. They're very sensitive to four moss. It's not.
I think it's harder to visualize the start of life altogether. And maybe easier just to think about how a bare mountain turns into a forest. Now, over the course of not billions of years, but tens of thousands. Yeah.
So moss begins to grow because what? Because the rock flakes start to just accumulate there at that spot where the water water. It keeps traps more moisture. More moisture.
And because there's more moss, it traps more moisture. And the moss dies. And that catches more flakes. Make more rock flakes come down.
Even larger rocks get to stop there, which creates which traps more moisture, which means larger plants can start to grow there. And so the spiral is that life creates the conditions for more life. Yes. And it's not necessarily always great for the life that comes before, because it all dies eventually.
The moss dies. And it becomes part of the soil. But that creates the conditions for more life. That's right.
Another example, which might be easier to understand, is like kelp. Kelp growing off shore, slow down the waves, which allow what? The beach to accumulate behind them, which makes more shore, which makes more space for kelp to grow. Right.
So it's a feedback loop. That's a positive feedback loop. A feedback loop, right. And this is, it tells it is a different story to the succession story that we learned in elementary school about how the grasses turn into the prairie turns into the forest because the forest outcompeted, the forest plants outcompeted the prairie plants.
The story is not that. The story that he tells is more that the prairie creates the conditions for the forest to come later. You couldn't have the forest without the prairie coming first. So it's not one of competition.
It's one of the prairie plants come first, create the soil that is needed and the conditions that are needed for the forest to come later. Right. And why it happens in a world of competition, but there is this other story that could be observed, if you look. That's right.
And I've been thinking about the upward spiral in the context of humanity is obviously not the end of this story. This story that is told all the way back to the beginning of the planet, humanity's not the end. What's next? And we've dabbled in the idea that AI could be humanity's children in a way, right?
And then we had this quip a couple shows ago about our children, our children, the AI, and our children, that idea. But what's next? What's the next thing in the spiral for humanity? We're creating ourselves.
We're sort of going up the spiral in a way, you could say. We're smarter than we used to be, because we have our augmentations. But are we getting any wiser? Are we going to see?
There it is, right there. That's the thing. I think if we could give rise to create the conditions for more life, what does life mean? Well, I think we could point in the United States anyway, we could point to the last 100, 200 years where we have continued to strive for more equality, include more people in what we mean by humanity.
We have extended what we say are rights, even though you would think rights were universal to begin with. The Declaration of Independence for the United States still stands up as a pretty important document. And we have just continued to add who gets included in that. So we've continued to add people that we say get to be included, which is not a very wise, even point of view, to come from, like this statement that I'm making.
I don't feel very wise when I'm making the statement. But it feels like there's some progress, in the realm of what it means to be human, that in this country, most people seem to include more people in the worldview of this is what humanity is now than it was before. Does that feel accurate? Does it feel like humanity's moving in the right direction?
Yeah. I mean, I think so. I feel it feels kind of like an aesthetic judgment to me. When I look at the otherization that we used to do, compared to the otherization that we do now, and it feels to me like a lot more people are us.
Yeah. Yeah. And that aesthetically is a step in the right direction for what it means to be a beautiful species. Right.
So then, does it feel like it's fast enough? Maybe not. Like I think about war. Why are we still fighting the same old conflicts?
And why does humanity bombing each other so much? And why is there so much violence in the world? That doesn't seem like we've solved anything in terms of wisdom, or that makes me feel disheartened. So I think Lyle may have hit on something before when we talk about the end of sex, and therefore things like love and other aesthetically attractive values that spring from sex being part of our evolution may become irrelevant, because it will no longer be necessary.
I guess I think that coming back to that original question of, all right, so 1,000 times we run it, and 1,000 times humanity sort of goes away. But what are we trying for? I think in a sense, it's to continue the aesthetic lineage that humanity has been a part of, of valuing diversity, of valuing, yeah, of expanding the scope of us, of this sort of human aesthetic that does maybe? Yeah, maybe it comes from sex.
Maybe it comes from sexual reproduction. Maybe it comes from, maybe it is really a tide to our evolutionary history, and isn't in a universal sense the absolute epitome. But the upwards viral is an expression of that aesthetic that I think is so beautiful. And I would love for that which succeeds us to continue that creating the conditions for more life, and valuing that trajectory, valuing and creating more complexity, more diversity, rather than not that, I guess.
And that's where I, and I see hope for my kids in that future, I guess. Yeah. You know, I came into this whole AI future singularity conversation, and I've shared that on the show several times that I came to it from reading Kurzweil years ago, more than 10 years ago, reading Kurzweil. And the way Kurzweil writes about the future, that we get to build more, we get to build more beautiful symphonies and more beautiful art and funnier jokes.
And as we become more and more augmented, or as we get more and more technology in our lives, we get to take what we've been so, like the spiral. He doesn't use the word spiral, but it feels to me like the same kind of thing. There's no discontinuity in the spiral between us as fully biological human species now, and then whatever the next stages are, the spiritual machines. Yeah, you're right.
Yeah, he's got to book the age of spiritual machines. And the whole premise of his work is that when a machine says that it is human, we will believe it, because it will be everything that we are. It will have spirituality. It will have a regional thought.
It will create art. It will need a body, clearly. It will need to have some representation of the world. So it will have the same kinds of experiences that we do, and that it will be at that point, at some point human.
Well, that was very compelling and very, what's the word? Romantic, you know, it was very romantic for you about the evolution of human beings, and that will be what it is. We may not call ourselves homo sapiens sapiens anymore. We may have some other name for it, but it will be the next stage.
It will be the same lineage. But all of the work that we've done on the show, trying to understand the risks and reading boss drum and reading Gdkowski and reading, all these other thinkers, is one in which super intelligence doesn't need to have any of those other qualities of human beings at all. You just brought up love, and love could fall away, because these machines don't need sex, and a beauty or any of those other things could just disappear. The paperclip maximizer certainly doesn't need any of them.
It's like, is there a stark either or? Well, when we talk about cosmos and terrons, and it was the play between Lyle and Eric, I think, literally between them, where we were kind of feeling that tension, I think we made the assertion that there are outcomes that neither cosmos nor terrons would like. And I think the crazy AI that turns everything into Grey Goo would be an example of one that neither of the cosmos nor the terrons like. So I think there's a space of possible outcomes, and there's an aesthetic attraction to some of that space, and an aesthetic revulsion that almost all humans feel when contemplating other parts of that space.
So how can we maybe, like thinking, it's easy to place things in sort of some kind of moral absolute and try to come up with a, like, what is the value, like the value function language, for example, feels very utilitarian, very kind of ethics-based. And I wonder if the route might actually be more aesthetic, like trying to create intelligence that shares our aesthetic, and therefore as it grows in the many different directions that it goes, you know, it's growing like the seed crystal from which the larger crystal grows is one that matches our aesthetic, in which the upwards spiral is a beautiful thing. Well, yeah, okay, so the upwards spiral is a beautiful thing, and that sounds amazing except that we've had some recent examples of machine learning stuff, which just created racist chatbots, you know, like within a day. And that's the Microsoft one I'm thinking of, that they put a chatbot on Twitter, and it just interacted with people, and people were terrible to it, and it became racist immediately.
But I think this is, I've also been reading, I haven't dealt deep in this subject at all, but I've been seeing some headlines about how machine learning is already replicating some of our sexism and our other biases, because it's in the corpus, right? It's in the training data that humanity has these biases, and so it's coming out, and I'm not saying these are intelligent yet, I'm not, but when you talk about the aesthetics, if you can't code it indirectly, and you just let it absorb what it can from the training data we give it, we've got lots of crap in there, we've got lots of stuff that does not match this aesthetic, this aesthetic of the upward spiral and of the wisdom, and you know, because we do it, we're frail, and we have all kinds of biases, and especially the two white guys, the two American, straight white guys sitting here, it's our worldview, maybe more of the worldview that sort of gets to be in charge, and there's plenty else out there that is possible, but how much of that is represented in the training data, how much, I don't know, I don't want to get that. How much unconscious sort of hegemony? Yeah, yeah, it's all built in the training data.
It's built in the training data. I feel very hesitant reading a lot of classic literature to my kids, because it's horrible, the way, like, Little House on the Prairie. Beautiful story is wonderfully written, and I love the way the family homesteads, and they're self-reliant, and they learn all kinds of crap and all that, but they're horribly sexist and racist, and so what are our machines learning from? What are my kids learning from?
Right, right, right, well, I think we should, I think that's where we should be trying to work on, is making intelligences that appreciate the beauty, and that tiny little percentage of things that humanity has created that are beautiful in that upward spiral way. So it's all a order. Well, I think that is worth struggling for, even in futility. Yeah, I guess you could think about how the value function, the value loading problem, right, we don't know how to specify these values, right, we don't know how to code it up, but we do now have all this machine learning stuff where machines and algorithms are learning in ways that we don't understand, but all those unconscious biases and all that stuff that's baked in, and maybe that's where we get big bang for our buck, is actually being human to one another, becoming aware of those things, or just like, yeah, becoming aware of them so that the next piece of literature that gets written, it has less of that bias and has more of this aesthetic in it that we want.
It's like working on the human level amount on the technology level. It kind of, I'm recalling our conversation with Evan in episodes like 12 and 13 or something, where he was critical of us because he didn't hear compassion in our language when we were talking about technology at all, and he was thinking, what if we were treating our machines better today? Like, would that make a difference? Well, what if it's not about technology at all?
What if we just treat each other better? What if we work on the human values of humanity and we're working toward justice? So, you know, I'm not really religious anymore, but I grew up like a Christian, and there was a story about Sodom and Gomorrah, and how the angels would spare it if there were any, like, holy people or righteous people, or, you know, they started out with, well, is there a 10? I can't remember exactly what the details are, but if there are a 10, we say it, and they're like, ah, yeah, we'll say it.
Well, if there are five righteous people in these towns, we say, if there's one, we say, and yeah, but there weren't any. So maybe our job is less about creating these machines that are going to treat us nice, and more about trying to create one righteous person, or at least like more of that behavior, where righteous in this case means kind of worthy of being cared for and respected as a precursor in the future. Like being worthy, I think, you know, that compassion is an example of that, right? Yeah, yeah, you know, when Evan was on the show and he talked about it, I didn't understand how that could make a difference, how the way we treat our algorithms today, you know, could make a difference, or for that matter, how we treat each other, could make a difference in this AI conversation.
But tonight, I'm now starting to see a possible path where because we're writing these algorithms that do learn in ways that we don't understand with the machine learning, you know, that maybe how we act does make a difference because that filters into what gets written down or what goes on the web, you know, which ends up being the training data. Well, and it's like, you know, like people talk about like, do what I say, not what I do. Because it's just, it's like a truism that your kids will pick up and learn from how you behave, no matter what you say. They're gonna follow your example.
So maybe we've gotta be thinking about our example more than we're thinking about what we say. What we do more than what we say. Yeah, okay. Food for thought.
Yeah. Food for thought. Yeah. Food for thought is so food for thought is so good.
I think I got a show. We got a whole bunch of other stuff that's been happening that we could have talked about, but you know, we'll just save it and may or may not be important. Also food for thought. Yeah, yeah.
Cool. All right, see you next time. Yeah, look forward to hearing your thoughts. Yeah, hit us up on the Facebook group and you're consuming AI.
We got a lot of traffic there lately. A lot of interesting stuff. Somebody posted a really great TED talk, which kind of I felt summed up our last year plus at the show in 18 minutes. So yeah, great.
Check that out. I'll put a link in the show once if I think of it. Yeah. It's up there.
We'll see you next time. All right, yeah. This is Ted. It's a new one.