Episode 74: The Problem of Open-Endedness
- Links to this episode: Spotify / Apple Podcasts
- This transcript was generated with AI using PodcastTranscriptor.
- Unofficial AI-generated transcripts. These may contain mistakes. Please check against the actual podcast.
- Speakers are denoted as color names.
Transcript
[00:00:07] Blue: Welcome to The Theory of Anything podcast. Today, we’re going to talk about an interesting thing called the problem of open -endedness. So you have this idea of evolution and Darwin’s Darwinian evolution. And I’m going to make the case. It’s actually not my case. It’s a case that comes from Kenneth O. Stanley. We’re going to talk about his theories that biological evolution exposes this idea of open -endedness. And I’m going to need to explain what that means. But it’s an aspect of biological evolution that you’ve probably not given much thought, but that’s really kind of important to even some of Deutsch’s theories, because he mentions it indirectly in beginning of infinity. So I wanted to kind of tease that out more. And that’s what this episode is going to be about today. So what is open -endedness? Now, I’m taking a lot of today’s episode from an excellent article called open -endedness, the last grand challenge you’ve never heard of. Now, I know I’ve quoted that article before on the podcast. So today, we’re going to summarize the arguments in that because I keep bringing it up, it makes sense. I should do a separate episode trying to get across to the listeners why it is I’m actually very excited about this idea. And this article is on orally.com and it’s by Kenneth O. Stanley, who is a name many of you have probably heard before, and Joel Lehman and Lisa Soros, who are his colleagues. So Kenneth O. Stanley is he actually showed up at Ratfest recently. I didn’t go to Ratfest, but like I saw that he was one of the speakers there.
[00:01:48] Blue: And that makes sense because his theory of open -endedness actually ties really super well into a lot of Deutsch’s theories. And like I mentioned, Deutsch kind of indirectly addresses it. He never uses the term open -endedness. So there’s some obvious overlap there. So in this article, Kenneth O. Stanley talks about his area of study, which is the problem of open -endedness. Now, what is the problem of open -endedness? So I’m going to be quoting extensively from this paper because he explains it very well. So he starts with there are other grand challenges like hearing cancer, achieving 100 % renewable energy, or unifying physics. Some fields have entire sets of grand challenges like David Hilbert’s 23 Unsolved Problems in Mathematics, which laid down the gauntlet for the entire 20th century. What’s unusual, though, is for there to be a problem whose solution could radically alter our civilization and our understanding of ourselves while being known only to the smallest sliver of researchers. Despite how strangely implausible this sounds, it is precisely the scenario today with the challenge of open -endedness. Almost no one has ever heard of this problem, let alone cares about its solution, even though it’s among the most fascinating and profound challenges that might actually someday be solved.
[00:03:07] Red: Well, he is right that I have never heard of this problem. And I’m looking forward to listening and learning here, Bruce. Yes.
[00:03:16] Blue: You know, it’s funny because I’ve mentioned the problem multiple times. I think even in podcast, you’ve been a part of it.
[00:03:23] Red: OK, OK, OK. I guess I have heard you mention it before. I should say it’s not something I’ve delved into.
[00:03:30] Green: Or maybe didn’t completely understand. Right.
[00:03:34] Blue: No, that’s what I was going to say. That’s what I say. Like it kind of goes over your head, right? Like if you don’t know what it is and you don’t have a concept of what I’m talking about, it’s a basically meaningless term. And because, like in culture, we don’t have this idea of like if I say Darwinian evolution, like we all know what we’re talking about or at least approximately we all know what we’re talking about. But if I say the problem of open -endedness, like it’s a basically meaningless term to most people, right? So it does no good for me to mention it if it’s since it’s not a part of the culture at this point. So OK, so that’s what we’re trying to rectify today. I kind of assumed it was more of some like sort of esoteric machine learning computer science thing that but it sounds like it sounds like it’s more more far reaching than than I thought.
[00:04:25] Red: Is that is that is that accurate? It is.
[00:04:28] Blue: Yes. In fact, I’m going to try to convince you today, Peter, that this is probably the single most interesting problem that you could be working on and that it might even be the problem of AGI. It has to be solved as part of the problem of AGI. So it might just be like a portion of the problem of AGI or it could be identical to the problem of AGI. We’re not actually sure. So I’ll make a case for both. But we know it’s deeply related to the problem of AGI, which is why I’m aware of it. Like it’s obviously that’s what got me interested. That’s what this show is really about is AGI. Like secretly it’s about that. I know it’s not like explicitly about that. So of course, this is something that caught my attention immediately and it’s always been at the back of my mind that there’s this problem of open -endedness and it is critical to solve it to understand AGI. Let’s talk about what it is. And I’m going to take a series of quotes from Kenneth O. Stanley’s article, but I’m going to back it up with other researchers I’ve read, including David Deutsch, by the way, who have addressed the same problem in different ways. And I’m going to show that there’s actually some really intelligent people who have called this problem out for a very long time. And I’m not quite sure why it’s so unknown, because like Leslie Valiant is one of the people I’m going to quote. He’s a huge name in machine learning. And this is like his area of study. It’s not the problem of open -endedness, but rather computational evolution, which is the same thing.
[00:05:59] Blue: It’s approaching the same problem from a different angle. Let me go ahead and try to make the case here using quotes from Kenneth O. Stanley first. So he says, evolution sounds merely like a plain process or a physical force like gravity, but the proliferation of life on Earth shows how severely this understates its nature. Evolution better resembles a creative genius unleashed for countless millennia that it does a physical process. It is the greatest inventor of all time. Think of it computationally. Evolution on Earth is like a single run of a single algorithm that invented all of nature. Emphasis is his, not mine. If we run a machine learning algorithm today, we’re happy if it gives us a single solution or maybe a few, but then it’s over. The problem is either solved or not, and the program is finished, but evolution on Earth is amazingly different. It never seems to end. There’s a term that captures this notion of a single process that invents astronomical complexity for near eternity. We call it open -ended.
[00:06:59] Red: So
[00:07:00] Blue: that’s where the term comes from. Okay. If this sounds a little familiar, it should because Deutsch doesn’t call it by this name, but he addresses the same problem in beginning of infinity. So let me give you the quotes from beginning of infinity, where Deutsch is whether he realizes or not touching upon the same problem. So he says on, this is from beginning of infinity, page 142, all known jumps in universality happened under the auspices of human beings, except one. On page 144, he says, initially the genetic code and the mechanisms that interpreted it were both evolving along with something else in the organism, but there came a moment when the code stopped evolving, yet the organism continued to do so. At this moment, at that moment, the system was coded for, or the system was coding for nothing more complex than primitive single -celled creatures, yet virtually all subsequent organisms on earth have not only been based on DNA replicators, but have used exactly the same alphabet of bases grouped into three base words with only small variations in meaning of those words. Page 145. So then the evolution stopped at a point when it had already attained enormous reach. Why? It looks like a jump to some sort of universality, doesn’t it? Okay, let me put this in plain English. It probably makes more sense when you’re actually reading it in the book. What he’s saying is, is that we don’t really understand how life got started on earth. And this is actually something I’m going to harp on throughout this episode.
[00:08:34] Blue: But we know that it must have been some sort of process that eventually led up to what we today think of as Darwinian evolution, where you have DNA and the DNA, every single animal uses the exact same DNA bases, right? And they all use DNA. Sometime prior to that, evolution must have been evolving how the organism organized itself, how it kept its information, its knowledge. And at some point, DNA was created by that process. And from that point forward, every animal you see on earth uses DNA, and the exact same process is being used to store that adapted information in DNA. Now, how can that be? Why is it that every single, there’s like this giant group of different animals, huge diversity of innovations that have come. And yet they’re all using the same underlying DNA, the same cellular structure. Something stopped evolving at that point. And it was, it was the substrate that evolution takes place on. Well, the only way that could actually be true where the substrate stops evolving along with the animals is if that substrate has now hit some sort of universality. Okay. And this is what Deutsch is actually saying here. He’s saying basically, he’s making the case that biological evolution somehow hit upon some sort of universality, even though we don’t know what that universality is. Okay. And that’s why from that point forward, all biological evolution takes place using the exact same DNA structure. And there’s more to it than that. Like if you’ve read some of Nick Lane’s books, he talks about how they all use the same eukaryote cell structure and somehow that was related to whatever happened. And his career is trying to figure out how that took place in the first place.
[00:10:34] Blue: So there’s like various researchers that have kind of come to the same conclusion here. Okay. But this is the Deutsch equivalent to the problem of open -endedness. He calls it universality instead.
[00:10:45] Red: That’s what I love about Deutsch. I mean, when I first read him, I think I was just interpreting him on what now seems to be a preposterously superficial level, but it still seemed like the best book I had ever read in my life. And it just keeps getting deeper and deeper. And it just, I don’t know.
[00:11:06] Blue: I hope maybe it’ll never stop.
[00:11:08] Unknown: But
[00:11:08] Blue: yeah. Okay. So going back to Kenneth O. Stanley, our textbooks describe evolution as if it’s understood or solved, but in reality, it’s most astounding property. It’s open -endedness remains at best simply taken for granted. If you look at it from the perspective of computer scientists, you can easily see that the ultimate explanation for this property would be something deeply profound and powerful. If only we knew it. It is indeed a mystery because so far, just as with AI, open -endedness has proven impossible to program. Yeah. So when he says AI, obviously he actually means AGI, right? Because obviously, we program AI all the time. But whenever he says artificial intelligence, he’s referring to programming a program of a artificial version of what humans do. Okay. So kind of keep that in mind contextually. So now, one of the online critical rationalists that I’ve talked to in the past is Carlos de la Guardia. He’s a younger guy and he considers himself an AGI researcher. I guess I’m technically an AGI researcher in the same sense. I don’t usually stylize myself that way because it’s not like I’m seriously trying to build an AGI. I’m just trying to study the theories around intelligence and talk about them and get other people interested in them. But I was talking with him at one point and I made the comment, well, we don’t really understand Darwinian evolution. And he goes, no, that’s not true. We do understand Darwinian evolution. And I was really shocked because I knew that Deutsch had basically said we didn’t really understand this aspect of Darwinian evolution. And he’s this huge fan of David Deutsch.
[00:12:51] Blue: I think that this is typical though, is that people do not think of Darwinian evolution as having a gigantic explanation gap in it. For whatever reason, we’ll talk a little bit about what I think is why this has happened. But for whatever reason, it’s definitely presented as if it’s this well thought out, well understood theory that all the problems have been worked out with it. And in reality, it’s not even close. We’re literally missing the single most important aspect of it at this point. And we don’t we understand it so poorly that we can’t program it. And this is why like if when we try to program evolution, we can do it to a degree, we can write like a genetic programming algorithm, and they work and they work just like Darwinian evolution says they will work. I remember when I did this for the first time, I wrote a genetic programming algorithm that was meant to figure out like so I had like an equation. And then the equation you would put X in and a Y would come out. And the the algorithm’s goal was to figure out what that equation was. And I remember it was just astounded me that you would just by just following Darwinian evolution, you would have replicators and they would cross breed and you do mutations on them. And it would come up with the exact same function that I that was creating the outputs. And it would actually figure out what that function is, not even though it would be a somewhat complicated function. And it would do it just by randomly changing things. And it was kind of blew my mind.
[00:14:34] Blue: It’s like, wow, like this this is Darwinian evolution, like you can see it works. But it’s not open ended, right? It always solves one problem. And it’s done. And it doesn’t even solve super complicated problems, right? If anything beyond one function, it probably will never figure out. And even with that one function, probably a good 20 % of the time, the algorithm would never end. And I would have to just kill it because it got down some path. And it could not figure out what the actual function is like most of the time it could. But sometimes it just would get down a bad path and it would get stuck. And it would never find the correct function, it would hit a what we call a local minima. And local local maxima, depending on how you want to look at it. And then it would just stay in that maxima and it would never get out of it and it would be stuck from that point forward. And it’s what we call genetic evolution, sorry, genetic programming today. It has all the elements that Darwin describes about evolution. And yet it is just not biological evolution. It’s not even close to biological evolution.
[00:15:39] Red: And just so unclear, how would we know that it had solved the problem of open endedness? Okay,
[00:15:47] Blue: so I’m going to give you some quotes here. I think I have, but let me try to explain that briefly.
[00:15:53] Red: So
[00:15:54] Blue: imagine, so think about biological evolution and how it’s constantly creating new things. Like new species are coming into being and then those species solve problems in super innovative ways. Imagine that you had a computational algorithm that could do that. Okay, okay. So maybe imagine it in creating virtual life forms in a virtual space. That might be one way. And that’s the way a lot of scientists have tried to go about recreating it is they try to come up with virtual life as it’s called artificial life. But it doesn’t even have to be that it could be like we run this algorithm and we give it knowledge to try to deal with how to build an AGI. And it just starts coming up with ideas and it comes up with more and more forms of intelligence and eventually discovers animal intelligence and eventually discovers AGI. I mean, imagine a sort of open ended algorithm that we could just in any domain let it start inventing things. Okay, so
[00:17:00] Red: it would be a program that never stops doing useful things.
[00:17:04] Blue: Right.
[00:17:05] Red: Is that okay?
[00:17:06] Blue: Yes. Okay, so let’s keep in mind that we’re dealing with something that we don’t understand well. So we have to describe it in somewhat vague terms right now. Like if we actually understood this problem, I would say, okay, here’s this algorithm and here’s how it works. Here’s what it does. And I would be able to explain to you exactly what I’m talking about. I don’t know what I’m talking about. Like I know biological evolution is open ended, whatever that means. And I know that genetic programming isn’t and I can see there’s a difference and I call that difference open endedness, but I don’t really deeply know what I’m talking about. Right. This is how all science starts though. Like you always have to start with this kind of these kind of philosophical ideas. They’re kind of vague. And then you have to kind of try to make them more precise over time if that makes any sense. Okay. So don’t read too much into this. Like I’m trying to give you a feel for what it would look like. But on the other hand, I’m using vague terms and I know it because that’s all we’ve got. We don’t have the ability to explain this problem in super precise terms. Going back to Kenneth O. Stanley, even the most sophisticated evolutionary algorithms, even ones focused on creating divergence run out of steam pretty fast, eventually exhausting their search space of anything new. The open endedness, open ended inventiveness of nature is still nowhere in sight. So now Deutsche admits that this is something this is a gap in our understanding of evolution that we don’t understand. Okay. So let me give the exact quotes here. So
[00:18:39] Blue: Deutsche admits that we don’t understand what kind of universality has been reached. That’s really trying to say the same thing. So on page 145, he says, reach always has an explanation. But this time to the best of my knowledge, the explanation is not yet known. If the reason for the jump in reach was that it was a jump in universality, what was the universality? Deutsche expresses doubts at this point that DNA is universal in the sense of capable of everything possible. So he’s not suggesting that DNA is a universal constructor. He considers in the book the possibility that it might be, but he expresses doubts that it actually is. And it’s not too hard to see why he would say that. It’s really hard to believe that biological evolution could say invent nuclear power plants or rockets that go to the moon or something like that. So it seems like the universal constructor would be the ultimate jump to universality, but this is some sort of subsequent jump to universality that’s not as powerful. Similar to how we can have a jump in universality in computers to say a finite state machine, and then later you can have a jump to a PDA, and then later you can have a jump to the Turing machine, and then that’s the final jump. So a universal constructor would be like the universal Turing machine. It’s the final jump, and this biological evolution is some sort of subsequent jump that comes before it, but isn’t quite as complete. This is what he’s suggesting. So on page 145, he says, perhaps it would be possible to specify in the genetic code an organism whose life cycle includes building a nuclear power plant, powered ship, or perhaps not.
[00:20:18] Blue: I guess it has some less and not yet understood universality. So elsewhere, Deutch admits that it can’t be, so here he’s considering that it could be the final universality, but elsewhere he admits that it really isn’t. He says biological evolution could never produce such an adaption as a steady search for extraterrestrial life. Only scientific knowledge can. This illustrates why non -explanatory knowledge cannot be universal. So you could easily get confused here. He’s saying DNA is universal, and then he’s saying it’s not universal. You have to, by context, realize in one case he’s talking about sub -universality. Remember, universality exists in a hierarchy. In the other case, he’s talking about complete universality. So DNA is not complete universality. It’s like a finite state machine. It’s not like a turning machine, but they’re both forms of universality. So he points out that technically DNA is a universal computer, though the universality has, to his knowledge, never been utilized by biological evolution. The first time that he thinks DNA was ever used as a universal computer was when a human decided, I’m going to build a DNA computer and then build a computer that uses DNA. I don’t know if you’re even aware that that happened, but decades ago there was a guy who built a Turing -compatible computer out of DNA, and at the time it was super fast compared to other computers that existed, probably because it had massive parallel processing. I don’t really know. I’ve never looked into it, but probably by today’s standard it’s not as fast as the types of super computers we can build today. So this leads to a kind of interesting idea.
[00:22:06] Blue: So Deutsch is admitting that we’ve got this explanation gap in Darwinian evolution, and he says it’s that we don’t know what type of universality it has. This is, I’m saying, equivalent to Kenneth O. Stanley’s problem of open -endedness. We don’t know why evolution is open -ended, and we don’t even know what we mean by open -ended. We don’t know how to program it. We can’t be precise about it. This means that Darwinian evolution has a gigantically large explanation gap and has had it since its inception. The reason why this is interesting, and I’m going to do a little bit of an aside here, is that I think one of the reasons why we’re not aware of this gap is because of the whole creationist versus Darwinists’ arguments, there was a desire to try to pretend like the gap didn’t exist, and I think at some point people thought that meant the gap didn’t exist. And one of the reasons why I feel that way is I was raised as a very religious person in a religious home, and so at one point my parents told me, you know, I’m a kid, and they tell me, oh, science says that, you know, we evolved, and there’s evolution thing took place, but we know it’s not actually true, you know. So I, as a teenager, I’m arguing with my atheist friends, and we’re talking about Darwinian evolution versus creationism, and we’re arguing between the two of these, okay? So because I’ve had experience doing this, being on both sides, I’m today, clearly I’m not a creationist, I’m a full -on Darwinian evolutionist, okay?
[00:23:48] Blue: So I’ve switched sides, I’ve been on both sides, and I understand both sides of the argument in a way I don’t think most people do because of this, and in this episode I won’t explain exactly how I switched sides or why I switched sides, that I’m gonna say for a future podcast. What I think will surprise people is that the reason why I changed my mind was because creationism, I refuted it at some point and stopped believing in it because it was refuted. Now the reason why I say that might be surprising was that at least for me that means creationism was a good, hard to vary explanation that could be tested and refuted. Now many people, including David Deutsch, claim that’s impossible because creationism is a quote easy to vary explanation, and so you’re not, according to Deutsch, in beginning of infinity, you aren’t supposed to be able to refute an easy to vary explanation. He actually says on page 25, only when a theory is a good explanation, hard to vary, does it even matter whether it is testable? My own experience suggests that is not entirely correct because my own experience with creationism was that once I had the right way of being able to test it, now obviously I’m not a researcher, I’m not testing it in the field, but I’m looking stuff up in books, and I’m looking up observations from actual tests, and at some point I was able to say, oh, I can see creationism is false, and it’s refuted. Now how did I do that when it’s supposedly easy to vary explanation?
[00:25:24] Blue: Well, I’m not going to talk about that today, but I’m going to just kind of raise the flag on that and we’ll come back to that at some point in a future podcast. Here’s the thing though. During my youthful years as a creationist arguing with defenders of Darwinian evolution, one of the things that stuck out to me and still sticks out to me today is just how bad and how easy to vary are the explanations that are used by defenders of Darwinian evolution. Again, you may be shocked that I’m saying this. So creationists utilize an epistemology that I call full -capistemology. Again, we won’t talk about what that means today, but let’s just say that it is kind of the default epistemology that humans use if you don’t know anything about epistemology. And it shouldn’t surprise you too much, but full -capistemology is quite similar to popper’s epistemology in certain important ways. And then, of course, it’s different in some important ways. Now this makes sense because the human mind must use some form of epistemology. We’re universal explainers. We must as a default be using some sort of epistemology that’s similar to the correct one. So it shouldn’t surprise us too much that if you know nothing about epistemology, you almost by default end up doing some large portion of popper’s epistemology. Not everything, but significant parts of it. So creationists, when they argue against evolution, what they do is they offer refuting counter examples to Darwinian evolution. They concentrate on falsification. Creationists are falsificationists. The main approach that they use is they show that historically many adaptions require irreducible complexity. You’ve probably heard this. If
[00:27:03] Blue: you at any point have been involved in intelligent design or creationism debates, even if you’re on the atheist side debating against them, they always raise irreducible complexity. So what is irreducible complexity? It’s the idea that some biological structures are so complex that it could not have evolved gradually through natural selection as the intermediate steps would not have produced any survival advantage, thus violating the theory of Darwinian evolution. So what they’re doing, what a creationist is doing when they raise irreducible complexity, is they’re saying, look, you have this theory that says things have to evolve in a certain way, and I can give you actual counter examples where that isn’t the case. And if you think I’m wrong, you need to explain to me how Darwinian evolution did evolve this. Well, that’s exactly falsificationism. So the creationists aren’t completely out of left field when they raise irreducible complexity. In fact, if you are a Poparian falsificationist, and especially if you haven’t listened to my episodes 42 and 43 where I talk about Popper without refutation, where if you want a spoiler, go listen to those. Those will actually lay the groundwork for exactly what the creationists are actually doing wrong. But if you understand Popper in a kind of naive falsificationist sort of way, then the creationists are completely following it, right? They’re absolutely falsificationists in the way many Poparians I know understand falsification. I
[00:28:37] Red: think it’s a really intriguing question actually that is being raised by creationists and some very interesting people like Paul Davies have, I think, gone down a similar path.
[00:28:50] Blue: Yes, yes, yes. And one of the things I would say, and hopefully if nothing else you get out of this, this is a good example of how even when you’re off in left field with a false theory like creationism, often that leads to really interesting things, right? And we probably should be taking these people seriously, seriously to refute them, right? But we should be taking what they’re saying seriously and raising irreducible complexity is a completely legitimate thing for them to raise from within a standard scientific falsificationist epistemology. So creationists are falsificationists and they are correct to raise the problem. So the classic example they usually come up with is the bacterial flagellum or whippetail. Here is Michael Beehe. So Michael Beehe is a legitimate scientist who is intellectual and an idear. And he’s the one who went into court and tried to defend ID in the Texas cases that took place. He’s done real research. So it’s not like this is some pseudo scientist, right? And he believes in ID. And so he has a book that I found because I was trying to figure out, I’m trying to quote the other point of view, if that makes any sense. And so here’s what, here’s his using the flagellum as an example. He says, the flagellum can be thought of as an outboard motor that bacteria uses to swim. It was the first truly rotary structure discovered in nature. It consists of a long philatumous tail that acts like a propeller. When it is spun, it pushes against the liquid medium and can propel the bacterium forward. The propeller is attached to the driveshaft indirectly through something called the hook region, which acts like a universal joint.
[00:30:38] Blue: The driveshaft is attached to the motor, which uses a flow of acid or sodium ions from outside the cells to the inside of power rotation. As with the mousetrap, it is quite difficult to see how Darwinian gradualistic process of natural selection um, sieved random mutations could, could produce the bacterial flagellum. Since many pieces are required before it funk its function appears, a hook by itself or a driveshaft shaft by itself will not act like a propulsive device. But the situation is actually much worse than it appears from a cursory description for several reasons. First, there is associated with the functioning of the flagellum and intricate control system, which tells the flagellum when to operate, when to stop, and, and sometimes when to reverse itself and rotate in the opposite direction. Because of such considerations, I have concluded that Darwinian processes are not promising explanations for many biochemical systems in the cells. So this is from irreducible complexity, obstacle to Darwinian evolution by Michael Bee. So another common one that comes up, and that I myself have used when talking to defenders of Darwinian evolution, is that Darwinian evolution is self -defeating because it can explain how life ever got started in the first place, given the second law of thermal dynamics. So from biology -wise, which is a pro -evolution site, they describe it like this. Earth was just a mound of mound mass of chemical elements and compounds, and none of these compounds can be called living matter. Now the question arises, how can a cluster of non -living compounds and elements congregate and combine to generate something that has life?
[00:32:27] Blue: So another point that I remember bringing up myself when I was a creationist and that I’ve seen other creationists bring up is the point out that humans seem to be unable to intentionally evolve new species. Now, of course, it depends on what you mean by species, but you cannot in the lab take a fish and evolve it into a cow, right? And if Darwinian evolution is supposed to be capable of doing that, I mean, maybe fish to cow is too far and that would take too long, but we’ve never seen people in the lab do much of anything like that. You can do boxes into dog -like creatures. We know that humans evolve into dogs, but those are so close to begin with, and just in terms of starting with a lizard and ending up with a mammal or something like that, there’s nothing like that that can be done in the lab with evolution as of today, right? Now, the reason might be very simple. It might be that it’s just intractable within the life of how long we’ve had science on the earth. But it is something that you can see why a creationist would feel like, look, you’ve got this theory and you can’t even like demonstrate it in the lab, you know? I mean, like, isn’t that kind of a problem with your theory? So what I want to point out, though, is that all three of these arguments that I’m raising, they’re kind of all fundamentally valid falsificationist, Popperian style arguments. It literally takes Darwinian evolution seriously as a theory, then it tries to find counter -examples, thus falsifying the whole theory through a single example, which Popper tells us we’re supposed to be able to do.
[00:34:10] Blue: Again, go listen to my episodes 42 and 43, Popper without refutation, and you’ll very quickly see this is actually a misunderstanding of Popper, but this is a really common misunderstanding of Popper, okay? So what would refute Darwinian? So here’s, here’s Deutsch now saying the same thing. So what would refute Darwinian evolution, theory of evolution, evidence which in the light of the best available explanation implies that knowledge came into existence in a different way? Well, that’s exactly what creationists are trying to do. I mean, like, they’re just not off base, right? They’re raising things in a proper scientific way, and they’re trying to refute the theory of evolution, and they’re coming up with examples that are really kind of tough. Now, how do the defenders of evolution that I’ve debated, how do they actually respond to these potential refutations? Well, typically something like this. So this is taken from a chat GPT summary of Kenneth Miller when he was arguing with Michael Bay. And so it summarized how people often talk about it. It says, many biological structures that were once thought to be irreducibly complex have since been shown to be, to have evolved gradually through natural selection. Well, that’s clearly a true statement. But that isn’t, I mean, like, that is clearly an immunization, right? Like to say, well, there was this one irreducible complexity, and it’s no longer irreducible complexity. Therefore, I can ignore every single counter example you will ever raise using that argument. Well, that’s clearly not a valid paparian argument. Okay. And yet that is typically what the defenders of evolution say. And it’s actually bad epistemology, even even by paparian standards, right? So in short, they’re saying, I have no idea how to handle your refutation.
[00:35:59] Blue: But there have been similar situations in the past. And I’m sure this one will eventually be handled as well. It’s an entirely inductive statement, entirely inductive argument. And it’s one that if it was allowed in science would make falsification impossible. So this is, in fact, a bad explanation, a bad argument, it’s incorrect epistemology that the defenders of evolution are using. So the argument against the second law that I raised, almost all, inevitably, if I raise this one, the one I always get back is the Miller -Urey experiment. I don’t know if I pronounced that right. But it is this experiment that was done back in the 50s. And so typically the answer or something like this, well, you see, you say Darwinian evolution can’t explain the creation of life. But there was this Miller -Urey experiment back in 1952, where they threw a bunch of chemicals believed to be present in pre -life together. And presto out came amino acids.
[00:37:01] Unknown: So
[00:37:01] Blue: like half the problem has already resolved. I’ve literally had people tell me half the problem is already resolved by the Miller -Urey experiment, because it showed that these amino acids can automatically organize themselves. And that’s half the problem right there. We’ve solved how life got started. All we need to do is throw a bunch of chemicals together. And it just naturally creates itself. And that’s what the Miller -Urey experiment proves.
[00:37:28] Unknown: Yeah,
[00:37:28] Blue: I grew up thinking it was kind of a known known that that scientists could create life in a lab. But lately I’ve realized that that is not actually true, right? It is not true. Yeah. So my own high school biology teacher used that Miller -Urey experiment to try to argue in favor of Darwinian evolution and against creationism. Now, I got to tell you this argument, even though it is used all the time amongst defenders of evolution, it is beyond silly. Okay. I mean, like it is so silly, it’s just laughable. Okay. Because showing catalysts that create amino acids as an explanation for how life got started, and then declaring it half the battle, it’s like me asking where an airplane came from. And you say, oh, well, you see, there’s these stars that create ore and they’re in the earth and it’s a natural process that creates iron ore. And see, there’s half the battle right there. So I don’t need to explain the rest of where that airplane came from. It’s literally that bad, right? It’s laughable beyond reason. Okay. And by the way, creationists know this. Okay. They’re not stupid people. They know they’re being given a totally bad argument. Okay. Here is a defender of Darwinian evolution responding to these charges as quoted on biology wise, a pro evolutionary site. To put it in a nutshell, almost all arguments against the theory of evolution stem from four major factors, absence of observation, evidential instability, scientific inconsistency and absence of evolutionary mechanism in the present times. However, just because something cannot be proven with certainty today doesn’t mean that that possibility isn’t there.
[00:39:12] Blue: Remember, there was a time when the whole world believed the earth was the center of the universe until Galileo came by and proved otherwise. This is such obviously an easy to vary explanation. It is a bad explanation. And this is the crap that defenders of evolution feed creationists. And sincerely seem to believe they’re good arguments. Okay. So I really want to kind of point out that there’s an issue here. Now, of course, in this episode, I won’t explain how to overcome creationist arguments. It’s actually not that hard. And creationists are getting something epistemologically very wrong. And once I understood that, I switched sides. Okay. But I want to emphasize we have gone to great lengths to pretend we know more than we do in an attempt to overcome creationist arguments. And it was bad epistemology, and we should stop doing it. Okay. Because these really are crappy arguments, and we shouldn’t be using them as people who really care about trying to find the truth and trying to be critical. This is literally being as bad as the creationist or maybe even worse than the creationist when we use arguments this bad. Okay. I would be very hard pressed to come up with worse arguments than the ones that I have seen offered. It would be literally impossible to refute anything ever if I’m allowed to use vague responses like the one I just read. So let’s drop those. Okay. So I’ve entirely switched sides on the debate at this point. I’m a full -on Darwinian evolutionist today, but we have underestimated how bad are the explanations and arguments used by the defenders of the theory of evolution.
[00:40:57] Blue: I was not wrong at the time when I saw such arguments as bad explanations and even as a sign that something was seriously wrong with Darwinian evolution, which is what creationists come away with after they have an argument with an evolutionist. Okay. They can tell something is seriously wrong here, and they are right, and we’re handing that to them in a way that’s very unconvincing. Okay. Creationists are quite right to demand that evolutionists offer up a precise testable explanation for how bacterial flagellum came to be or how life got started in the first place. Now, I know we can’t offer that today, and it turns out that’s actually the right thing to say is you’re right. That is a problem with the theory, and we don’t know the answers to everything today. And if you understand epistemology correctly, that’s not wrong to say. And it’s possible to make the argument and explain why that doesn’t mean creationism is true. In fact, it tells you nothing about creationism is the actual correct epistemological answer. Okay. However, the assumption that you can falsify a theory by nothing more than a counter example, however much Popperians claim this to be correct at times, this is actually substantial misunderstanding of critical nationalism. Again, I point you to, I guess it was episodes 42 and 41, popper without refutation if you want to understand this. This is the actual problem with my creationist arguments, by the way, is folk epistemology and apparently most self -proclaimed critical nationalists often wrongly think that epistemology is about a sort of naive falsificationism, and there’s something else going on. Okay. They often treat it like science is this thing where you can offer the single counter example and boom, you’ve refuted the theory.
[00:42:38] Blue: And that is not what Popper is saying. And there is a much more subtle argument being made here. I will give a better answer to this some other time. It would be too much of a digression at this point. So I’m going to move on. I just want to really use this though to emphasize that there is problems with evolution and they’re huge and they’re large explanation gaps, and that’s okay. But we need to accept that they actually exist or we can’t make progress with this theory. So I think that’s one of the main reasons why people don’t know about the problem of open -endedness is that there’s such a rush to say, oh no, no problems here that we’ve completely missed the fact that actually there’s tons of problems, very, very big problems that need to be dealt with. Karl Popper understood these issues with Darwinian evolution and went so far as to declare it a non -scientific theory. So now I’m going to give you atheists. I don’t know technically Karl Popper is an atheist, but scientists that are mostly atheists who are going to now admit to, yes, evolution has giant problems. And we already saw Deutsch basically admits that. We’ve already seen Kenneth L. Stanley doing that. Let me give you some others. So Karl Popper is one. So this is why Karl Popper always insisted that evolution wasn’t a scientific theory, which by the way, isn’t true. He was wrong about that, but you have to understand where he’s coming from. Okay, so from Karl Popper in the on the unended quest page 95, the question of the scientific status of Darwinian theory in the widest sense, the theory of trial and error elimination becomes an interesting one.
[00:44:14] Blue: I have come to the conclusion that Darwinian ism is not a testable scientific theory, but a metaphysical research program. I do not think that Darwinism can explain the origin of life. I think it’s quite possible that life is so extremely improbable that nothing can explain why it originated. If life started as a merely low probability, which have become high because of an immensity of the available time, which is the Boltzmann explanation for how life got started, then we must not forget that in this way it is possible to explain, scare quotes around the word explain, almost everything. In other words, Popper is saying the current explanation that we offer for how life got started, which at the time, we’re not there anymore. Go read Bobby Azarian’s book and you can see that we’re starting to get theories about how life actually got started. They’re the way that defenders of evolution tried to defend how life got started is they’d say, oh, well, you see, the initial life was just an improbable circumstance. All these chemicals came together and somehow life got started and it was just a coincidence. It was a wildly improbable thing. But because there’s so much time in the universe, highly improbable things are likely to take place eventually. And that’s how life got started. What Popper is trying to say is Popper is accepting that might be the case. But what he’s trying to say is that is a bad explanation. He’s admitting that’s the case and it is that is as bad an explanation as you can possibly have. OK, oh, it just it’s just a coincidence. OK, it’s literally that bad. Now, I could be that’s the case. Maybe that is how life got started.
[00:46:04] Blue: But that really isn’t a good scientific explanation. In fact, that’s really just a non explanation. And that’s why you now have scientists today really trying to figure out there. It must be something better than that. It must be that the laws of physics somehow imply that life will come into existence. And again, Bobby Isarian’s book is entirely about this. OK, the romance of reality. We’ve been trying to get him on the show. He’s probably too big to come on the show. But I really wish we could talk more about the actual theories that now exist, the nascent theories where they’re trying to explain how life got started. They’re trying to get away from that bad explanation of, oh, it was just a coincidence. OK, but at one time that was what they believed. And that was the best explanation they could come up with. Now, here’s Penrose on evolution. So I love Penrose. Emperor’s New Mind page 41416. So Penrose is, without a doubt, an atheist, like completely an atheist. OK, he says, to my way of thinking, there is still something mysterious about evolution with its apparent groping towards some future purpose, things that at least seem to organize themselves somewhat better than they ought to, just on the basis of blind chance evolution and natural selection. There seems to be something about the way the laws of physics work, which allows natural selection to be much more effective process than it would be with just arbitrary laws. The resulting apparently intelligent groping is an interesting issue. So again, Penrose is no theist, not even slightly, right? He’s not defending intelligent design. He’s a full on evolutionist. But he is openly admitting we don’t understand Darwinian evolution. OK, that there’s something big missing here.
[00:47:45] Blue: So now let’s talk about the fact that we don’t know how to code evolution. This is Leslie Valiant, one of my favorite authors. He wrote a whole book, probably approximately correct. Nature’s algorithm for learning and prospering in a complex world. Leslie Valiant, you’ve probably never heard of him, but he’s a huge name in machine learning. Like you can’t go through machine learning in graduate school without hearing his name multiple times. His career is built largely around trying to understand what evolution actually is computationally. So this is called computational evolution. He’s also interested equally in computational neurobiology, how humans think. And that led him to create a theory called the theory of the pack theory, probably approximately correct. Now, if you know anything about machine learning, that is the premier theory underlying all of machine learning. It is the theory of machine learning. What’s it called again? Probably approximately correct or pack theory.
[00:48:47] Red: Probably or approximately correct.
[00:48:49] Blue: Probably approximately
[00:48:51] Red: correct. Approximately correct.
[00:48:53] Blue: Yes.
[00:48:54] Green: Which is a funny phrase in and of itself,
[00:48:56] Blue: isn’t it?
[00:48:57] Green: It is. Approximately correct.
[00:49:00] Blue: So it means literally what it’s supposed to mean, that basically he’s worked out how to measure if a machine learning algorithm, it doesn’t work on some algorithms it’s hard to measure on.
[00:49:15] Unknown: But
[00:49:15] Blue: under certain circumstances, you can actually measure, you can create guarantees that this algorithm will get the right correct answer with this probability or get approximately the right answer with this probability under this many number of circumstances, right? So you can, it literally gives you a guarantee that the machine learning algorithm is going to come up with the right answer or at least an approximately correct answer within some bound. It will be correct within a certain bound and it will be within that bound some percentage of the time. And it tells you what the percentages are and how close it will be to the correct answer within that percentage. Okay. So this is something that we’re going to need to revisit at some future date when I actually get into wanting to do podcasts on machine learning and artificial intelligence. But machine learning is often treated as something super distinct from like paparion epistemology. And Leslie Valiant’s theory literally is the exact opposite of that. The theory of machine learning is literally the exact opposite of that. They see themselves as trying to understand what these, he calls them ecorhythms, what these ecorhythms are doing. And he sees evolution as a subset of machine learning, biological evolution is a kind of machine learning, not something distinct from machine learning. Okay. His whole book is excellent and it’s about this. And he actually produces a proof that evolution is a subset of his probably approximately correct theory. Okay. So this is something that deserves its own podcast, but I just kind of wanted to explain the background from this book and why I’m going to be quoting from it. So Donald Knuth said, science is what we understand well enough to explain to a computer.
[00:51:08] Blue: Art is everything else we do. That’s from page 14 of the book. So that may sound a little familiar. That notion got picked up by I think Richard Feynman at some point. And then Deutsch takes a version of it from Richard Feynman, where he talks about, in the beginning of infinity, he talks about, you’ll know that you have a correct theory of intelligence because you’ll be able to program an AGI and tell you can, you don’t really understand it. That’s where this all comes from. It actually originates way back with Donald Knuth here. Okay. And by the way, he’s a huge name in computational theory. So you cannot take computer science without hearing about his algorithms and everything he created. He practically created the whole field, right? So he’s a giant name. So Leslie Valiant now says, I want to discuss evolution, learning and intelligence in terms of algorithms that are unambiguous and explicit enough that they can be explained to and simulated by a computer, page 14. He calls this computational evolution, which is the search for the actual Darwinian evolution algorithm that is explicit enough to be programmed. Now it is exactly equivalent to what Kenneth Stanley is trying to do with the problem of open -endedness. I guess it’s not quite exactly equivalent. Stanley’s issue is something that’s a subset of computational evolution because he believes it’ll be useful in non -evolutionary contexts. Okay. But they’re trying to solve the same problem, essentially, but they’re coming at it from different angles. And they’re both based on this realization that we don’t really understand Darwinian evolution, that our knowledge of Darwinian evolution is incredibly limited compared to what we need to understand. Okay.
[00:52:53] Blue: So now Leslie Valiant, the way he kind of tackles this is in terms of it needs to be tractable. And that may sound weird, but let me see if I can explain using his own words why this is actually a really intelligent way to tackle the problem. So on page 18, Valiant says, from the first availability of digital computers, many intelligent curiosity -driven individuals have sought to simulate selection -based evolutionary algorithms in order to demonstrate their efficacy. These simulation experiments have been disappointing in creating mechanisms remotely reminiscent of those found in living cells. In fact, these experiments are seldom quoted as corroborating evidence for evolution. This failure can’t be ignored. It suggests that the natural selection hypothesis has to be refined somehow if it is to offer a more explanatory scientific theory. Further, the refinement will need to have a quantitative component that reflects the realities of the actual bounded numbers of generations and bounded numbers of individuals per generation that apparently have been sufficient to support evolution in this universe. So what he means here is that it must be class P, not class NP. Go back to our episodes on computational theory. Class P is the tractable algorithms. It’s not too much in doubt that if you don’t have to worry about tractability, just like the creation of life. If you’re taking the stance, there’s so much time available in the universe that life could happen by chance. You could do the same thing with evolution. You could say there’s so much time available that just by chance, eventually you’d wind up with the type of diversity we see today. The problem is, the universe is bounded and billions of years may sound like an open -ended bound, but it’s not.
[00:54:46] Blue: That is still tractable. So whatever evolution is doing when it creates this diversity of life, you cannot cite, there’s just so much time. That’s a bad explanation. You need to actually show that some P tractable algorithm, class P algorithm, is able to create this problem of open -endedness, solve it, the problem of open -endedness, and could create the sort of diversity of life that we actually see on earth. There is no such known algorithm. In fact, it’s difficult to even believe one could exist, is kind of what Leslie Valiant is getting at. Now, we know it exists because it happened, and yet it seems like to create this much open -ended number of diversity of things in only a few billion years, it really feels like you would need an intractable algorithm. Not only do we have to find one that does it and solves the problem of open -endedness, but it needs to be class P. That’s what Leslie Valiant is saying. This is him throwing down the intellectual gauntlet over what we need to still understand about Darwinian evolution. So theories of evolution, this is Leslie Valiant again, theories of evolution that assume unbounded resources of evolution do not address this at all, cannot resolve the central scientific question of whether some instance of natural evolution does fit of the rules contained in this universe. So, now, Leslie Valiant again, page 87. The knowledge gained in this century and a half since Darwin confirms his theory is qualitative and in qualitative terms. The proposition that current life forms is fully supported by evident similarities between their DNA as well as by the rich fossil record. All this, however, does not mean that the current theory of evolution is adequately explanatory.
[00:56:39] Blue: At present, the theory of evolution can offer no account of the rate at which evolution progresses to develop complex mechanisms. Notice that this is identical to what Roger Penrose was raising as an issue. It just seems like it’s too effective. Nor how evolution, as suggested by the fossil record, could have occurred within the timescales generally attributed to that record and with the physical resources of the earth. Now, page 88, he talks about its lack of testability. The theory of evolution through natural selection is not the same nature as some other theories of science, such as Newton’s laws. The later, meaning Newton’s laws, make quantitative predictions that are subject to verification. In contrast, evolutionary theory at present offers no comparable quantitative predictions or even quantitative explanations of the past. Perhaps this is why, among the great theories of science, it is the theory of evolution that arouses the most skepticism and organized opposition. By the way, notice that he’s explaining that creationists aren’t just idiots. There’s a kind of a good reason why they feel these feelings of doubt about the theory. Further generations will wonder why these questions have not been asked with greater urgency, says Leslie Balliant. Now, one phrase that he says that I want to just kind of call out, he says subject to verification, don’t make the mistake of being a bad paparian and thinking that critical rationalism is not about verification, and therefore, he said something wrong. He’s talking about verification of an experiment and popper in logic of scientific discovery makes it very clear that you are doing verificationism when you’re doing an experiment. It just doesn’t verify the entire theory. Okay, so this was not a misquote or misunderstanding of epistemology or anything like that that took place.
[00:58:33] Blue: So now, Valiant on page 89, attempts to code evolution have not produced results that are suggestive of how complex biological systems might have evolved. Darwinian evolution as a panacea for the creation of complex functioning mechanisms remains to be demonstrated, page 90. There is every reason to believe that such a more systemic analysis is necessary if we are to understand how evolution can give rise to the forms of increasing complexity as far as it is believed to have happened on earth. And then we often, like Deutsch often quotes, Paley and Paley’s watch and we often act as if evolution is sort of an answer to Paley, which it is to a limited degree. But here’s Valiant now claiming that we never really gave Paley a good answer. So page 15, convincing direct counter argument to Paley would need to be, need a specific evolutionary mechanism to be demonstrated capable of giving rise to the quantity and quality of the complexity now found in biology within the time and resources believed to have been available. Page 92, after such a demonstration has been accomplished, the Darwinian theory will progress from being more than a metaphor. I love that quote. So he claims that we can’t really effectively respond to Paley today because of these giant gaps that exist in current evolutionary theory. Page 112, Paley objected to evolution because life forms are just too complex to have evolved. Darwin avoided confronting this issue directly as have all his successors. It is one thing to demonstrate that natural selection is qualitatively consistent with the evidence. It is quite another to show that some concrete realization of it is consistent with the evidence in quantitative terms.
[01:00:22] Blue: By the way, that’s really the whole bacterial flagellum issue is that typically the response given is, well, okay, we can show that at least one of these items had another use at some point in the past. And that works for me. I don’t believe the creationism is the right answer, regardless, because it’s just a bad explanation period. But the fact is that the creationists are right to ask for a convincing account of how the entire bacterial flagellum came up, and we don’t have that today. We have vague ideas of how it might have happened. And this is often presented as if it’s far more sure that it actually is. And I think this is what Valiant’s getting at here is that Paley raised this issue, and it’s never been fully addressed. Because while we do have a vague notion of life evolved in some way, we don’t really have an exact scientific theory that gives us these testable predictions and tell us exactly how to go create it on a computer. And we’re just completely lacking that today.
[01:01:27] Red: Basically, what a convincing answer to these questions would look like would be somehow to tie in evolution to either the laws of physics, I guess, or machine learning, like you say. Is that what Penrose does with his theories of consciousness? He tries to tie in consciousness to laws of physics, maybe not very well, but am I on the right track?
[01:01:56] Blue: You are. I can clarify that a bit though. So I just, on the increments podcast, I just recorded an episode with the increments podcast, and I actually gave an account of what Penrose’s theory is and also why it was wrong. That is certainly what he means to do, right? He’s saying, look, we’ve got these big questions, and then he’s imagining we need new laws of physics to answer them. Now, I don’t think he’s right about we need the new laws of physics part, but he isn’t wrong that there’s legitimate questions there that need to be answered.
[01:02:24] Red: He’s got the right questions, maybe not the right answers. That’s right.
[01:02:29] Blue: And so what Leslie valued and what Kenneth O’ Stanley is really saying, and this is correct, is that a proper response to the creationists and to a complete response that we don’t have available today to the creationists and to Paley and to people who felt doubts about this really would be that we can program it on a computer and we can demonstrate it on a computer and that we understand it so well how it works. And we can show that it happens within a practical period of time. We can say, look, this algorithm works this way and it is completely capable of giving rise to new virtual forms of life or whatever it is that we’re trying to do it with. There might be different ways to go about this. And we can show that it invents new things. We can show that it does so in a tractable amount of time. And that’s really the answer that’s needed. And we’re no more close to having that answer, right? There is a big gap here.
[01:03:23] Red: So the answer is more in machine learning rather than the laws of physics.
[01:03:30] Blue: And that’s really what Deutsch is saying. When Deutsch says we should be able to create an AGI, he’s not ruining it in laws of physics, he’s ruining it in we don’t understand the algorithm. And that is what Leslie Ballin is saying. That is what Kenneth O. Stanley is saying. Each of them is addressing it in a different way, but the answer is always the same. You have to actually program it. Until you program it, you don’t understand it. And that is really, so Penrose’s approach won’t need new laws of physics. I don’t agree. But yeah, he’s right that there’s this big issue there that hasn’t been resolved yet. So now here’s where things get interesting, though. This seems like such a big thing. Like we’ve got this giant missing part of evolution that we don’t understand. But evolution must have, biological evolution must have stumbled across all this fairly early on, right? So I don’t know that we’re necessarily talking about a complex or difficult algorithm that we need to find. This may seem really strange for me to say this, right? But the fact is, is that the problem of open -endedness probably is not a difficult algorithm. And it’s got to be, you’ve probably heard Deutsch argue that AGI can’t be a difficult algorithm because there’s only a few K difference between a chimp and a human. And so the algorithm must be something fairly simple. But the problem of open -endedness should be a simpler algorithm than that because it predates the AGI algorithm. So it’s unlikely that we’re dealing with some super complicated theory. It’s just for some reason we can’t seem to conceptualize it right as of today. Here’s Kenneth O. Stanley now.
[01:05:22] Blue: He says it’s possible the process of evolution itself is simpler to describe or implement than the human brains it created. In fact, that seems like a very reasonable argument to me. Deutsch in a recent interview says AGI, about AGI rather than open -endedness, it’s possible that the idea that will open the door to AGI is the kind of idea where there will come a time when everybody thinks it’s obvious and that in our time we were being obtuse for not seeing it. We differ from monkeys who have brains similar to ours by a few K of code. In that few K of code is the Bootstrap program. This qualitatively different type of program that we run, infinitely different, in hindsight we’ll realize there wasn’t much to it to write this program. A few K and we’re done. So we probably aren’t talking about a difficult program. What the real issue probably is is that we’ve just got the wrong paradigm. Specifically what I mean by that is there’s something so wrong with the theory of evolution, the way we’re currently conceptualizing it, that it is blinding us from whatever the obvious truth is. Again, you never hear stuff like this. This may even sound like crazy talk, but this is exactly what Kenneth O. Stanley is saying. Here’s a quote from Kenneth O. Stanley. It is now becoming clear that open -endedness, while perhaps simple, involves a kind of mind trick that would force us to re -examine all our assumptions about evolution. The whole story about selection, survival, fitness, competition, adaption, it’s all very compelling and illuminating for analysis, but it’s a poor fit for synthesis. It doesn’t tell us how to actually write the process as a working open -ended algorithm.
[01:07:05] Blue: To pinpoint the reason we see open -endedness in nature and hence become able to write an algorithm with analogous power, likely requires a radically different evolutionary narrative than we’re used to. Okay, he’s saying this nicer than I just did, but he’s saying exactly what I just did, said, Darwinian evolution is so severely flawed in some important way that it is blinding us to an obvious truth. That is what he’s saying.
[01:07:33] Red: I hope this won’t lead down a tangent, but am I correct that the chemist Lee Cronin has kind of like the, as far as I know, the major figure in sort of an alternate view of life coming from more like a chemical process, an unknown chemical process, rather than something more like machine learning. Is that right?
[01:08:00] Blue: So that is surely how he presents himself. I need to study assembly theory.
[01:08:07] Red: Okay.
[01:08:08] Blue: And I need to do an episode on it. I haven’t studied it. I don’t know enough to intelligently talk about it. I’m aware. Is he gross? Come on, man. I’m aware that there are some heavy criticisms of assembly theory by some really big minds. And I would have to first study assembly theory, understand it, then I would have to go back and study the criticisms. I don’t have to figure out what’s going on.
[01:08:30] Red: Okay. But I’m at least right that the assembly theory is the sort of at least a prominent alternate theory to what you’re talking about. You are
[01:08:39] Blue: correct. Now, you mentioned as opposed to machine learning,
[01:08:44] Red: that
[01:08:45] Blue: part seems to me to be where Lee Cronin doesn’t understand computational theory.
[01:08:53] Green: So
[01:08:53] Blue: when he makes claims like, oh, you can’t program AGI because you have to do it through assembly theory. Like this is such a fundamental misunderstanding of computational theory that I feel like you can basically ignore those claims. He doesn’t know what he’s talking about. Assembly theory may well be a path to understanding open -ended evolution. And it may even be very promising in that way. But when we actually do understand it, we’re going to find out that it isn’t actually at odds with computational theory. And when Lee Cronin was making these claims, he simply didn’t know what he was talking about. He was an area of knowledge that was outside his area of knowledge. So he didn’t understand what he was talking about. And of course, scientists make mistakes like this all the time once they get out of their area of knowledge. To put it somewhat straightforwardly, he sees computers as distinctly different than what we would call a universal constructor. Like what he’s really trying to build is a universal constructor. And he’s right. A universal constructor, a universal computer is a subset of a universal constructor. So a universal constructor is capable of more than a universal computer. That doesn’t mean you can’t simulate everything a universal constructor does on a computer. The simulation hypothesis is the part that he’s misunderstanding. So he sees that he sees the part that they’re not the same thing and that one’s a subset of the other. And he’s drawing the conclusion that, therefore, you can’t program AGI on a computer. And he’s thoroughly misunderstood the simulation hypothesis and what it means. And I don’t know what else to say. I know for sure from reading his stuff, he doesn’t understand the simulation hypothesis.
[01:10:39] Blue: So even if his theory is correct, which it might be, it doesn’t actually have that implication, if that makes any sense.
[01:10:48] Unknown: Okay.
[01:10:49] Red: No, that’s helpful. When you say the simulation hypothesis, you’re not talking about the idea that
[01:10:55] Blue: everything
[01:10:57] Red: is a simulation just to be clear.
[01:11:00] Unknown: Yeah.
[01:11:00] Red: The
[01:11:01] Blue: simulation hypothesis here would be Deutch’s simulation hypothesis that everything can be with arbitrary levels of accuracy simulated on a quantum computer.
[01:11:12] Red: Yeah. Okay.
[01:11:13] Blue: That’s what I mean by the simulation hypothesis. You’re right that the words simulation hypothesis also refers to the idea that we live in a simulation. And that is a ridiculous supernatural theory.
[01:11:24] Red: I think most people would think of the Nick Bostrom thing when they hear that. Yeah. So thank you for clarifying that.
[01:11:30] Blue: Okay. So continuing, Kenneth O. Stanley, see, something doesn’t make sense here. And when something doesn’t make sense, often there is a paradigm shattering discovery awaiting somewhere in the shadows. So why does this all matter? Okay. Quoting Stanley again, for a moment, imagine if we could actually program genuine open -ended algorithms. The implications would be extraordinary. Are you interested in a new school of architecture, new car designs, new computer algorithms, new inventions in general? How about generating them ceaselessly and with increasing complexity, endless new forms of music and art, video game worlds that unfold forever without ever getting dull, universes that emerge inside your computer wholly unique and unlike anywhere else. The power of nature is the power of creation. And it’s entirely encapsulated within the mystery of open -endedness. These dreams sound like some of our aspirations in AI, meaning AGI. But if they are part of AI, then they are precisely the area where AI remains unfocused, with so much energy pouring into finding solutions to particular problems. Then he talks about degrees of open -endedness. He says it is also important to highlight that there are likely interesting degrees of open -endedness, an idea embraced by Badoe’s tests. You have to read the article to see what he’s talking about. That is, open -endedness is not just a binary either or proposition. This might not seem obvious at first, but consider that the Deutsch just argued for genes having a lower level of open -endedness than human explanations. This is really kind of what Stanley is saying here. So genes probably can’t produce a nuclear power plant or organisms that fly to Mars, but they do represent a kind of open -endedness.
[01:13:24] Blue: So the idea of degrees of open -endedness is going to be a necessary part of the theory. We have every reason to believe it’s going to be a necessary part of the theory. Talking about open -endedness and AGI, the thought of open -endedness as a path to AI, remember he always means AGI, is interesting for how different it is from how most AI researchers approach today, where optimization towards specific objective outcome performance is ubiquitous. Almost the perfect opposite of how open -ended systems generates its products. Notice how this is really reminiscent of David Deutsch’s Creative Blocks article that we reviewed a few podcasts ago on its like 10 -year anniversary. Human nature seems at least in part open -ended. We don’t only optimize our minds to perform tasks, but we invent new tasks and identify new problems to solve. We’re all also playful and like to create simply to stimulate ourselves, even if no particular problem is solved, such as in art and music. Now, interestingly, Stanley claims that open -endedness does not start with a problem. This is the opposite of what you’ve probably heard Deutsch say. So he says, in short, the open -ended component of our minds is a spark that still largely separates us from what the way we think of machines, which suggests that open -endedness is a core component of what we mean by general intelligence. This point is important because so many in AI lean towards definitions of intelligence that involve solving problems or learning to solve problems efficiently, but open -endedness is left out of this equation.
[01:14:58] Blue: A general open -ended system is not a machine like problem solver, but rather a creative master that wanders the space of imagination that is certainly within the patheons of human intelligence. Open -endedness defies the dominant paradigm in computer science and AI or machine learning today of problems and solutions. In these fields, you can choose a problem and showcase improving results with respect to some benchmark. Open -endedness isn’t like that. Open -endedness is for those yearning for an adventure without a clear destination. It is a road of creation itself and the entire point to generate what we presently cannot imagine. Welcome to the newest grand challenge. Now, it’s interesting that if you’ll recall way back in episode 44, Deutsch actually talked with me and I got to clarify some of his theories. If you’ll recall, he tried to explain to me the whole perspiration versus inspiration and I called him out on that and said, you know what? I have no reason to believe AGI isn’t going to require perspiration. I understand what you’re getting at, but it’s just too vague. And we don’t know what inspiration is. So you’re offering it as an example of what’s the difference between the two. I’m not sure it works for me. And he stopped trying to explain it that way. He thought about it for a while and then he said, it’s really about being able to create your own problems. And I said, oh, that’s brilliant. And I talked about that quite a bit in episode 44 of how I thought that was a brilliant idea. That is what Kenneth O. Stanley is saying here. He’s saying the same thing.
[01:16:38] Blue: It’s interesting because he sees that as not worrying so much about which problem you don’t start with a problem anymore. And that’s not probably the way Deutsch would have put it. But Deutsch seems to be an agreement that AGI, when we have it, that it’s not going to solve a single problem. It’s going to have this ability to kind of create its own problems to solve. And that’s very vague and I know it is. And so it’s not very helpful, but hopefully it gives you a feel for what we’re talking about. Now, Valient points out that this isn’t quite right as evolution does have a target or goal that it is solving for, which he says is increasing fitness. In other words, being maxily adaptive to a given ecosystem. This is from page 93 of his book. And quoting him, he says, fitness and Darwinian evolution is a measure of the benefits that an entity enjoy in some environment. Selection then directs evolution in favor of entities with higher fitness. Some of you may at this point think of Deutsch’s critique against fitness as a proper description of evolution that he comes up with in beginning of infinity. I should probably point out that I had a chance to ask Deutsch about that. And you actually can find this recorded me asking him about it in our episode where we had him where we hosted and people were asking him Deutsch questions. And I actually say to him, well, actually, evolution is about fitness. If you by fitness, you mean basically what Leslie Valiant just said that it’s and he goes, yes, you’re right, it is.
[01:18:16] Blue: But the problem is, is that when you use the word fitness, people think of like a bigger, stronger organism that like they’re thinking of the wrong thing. It depends on how you define fitness, which I think is the right answer. Like I’ve seen people say, oh, no, evolution is not about fitness. Deutsch says evolution is not about fitness. And that’s not what Deutsch says. He’s against defining fitness in a certain way that he’s a certain misunderstanding of fitness. The valiant description of fitness is the correct one. Fitness is a sort of meta problem, if you will. And what makes evolution open -ended is that biological evolution is solving many different problems, but all towards this goal, this meta problem of fitness. Okay, here it is. This is from, this is sorry, from Kenneth O. Stanley’s article. And I quoted this on the increments podcast. And Baden took a little bit of issue with it. I’ll explain what he took issue with, but it wasn’t anything major. Kenneth O. Stanley says that is the highest level of open -endedness requires not only generating new solutions, but also new problems to solve. And I think that’s really what Deutsch is getting at. That’s really what Stanley is getting at. That’s really even what valiant is getting at. But the problem is, is that the word problem is too vague. And Baden goes, yeah, you know, if I go running around naked in the streets, that’s going to be a problem. But I promise you, it’s not a kind of creativity. I says, I don’t know, Baden. That sounds pretty creative to me. But it’s, you can kind of see that they’re grasping at something and they’re all kind of grasping it from different angles.
[01:19:54] Blue: Valiant is looking at this from a computational machine learning perspective. Stanley is looking at this from a standpoint of trying to solve the problem of open -endedness. Deutsch is looking at it from the standpoint of universality and what’s the universality of evolution. And then he tries to build it into a theory of knowledge, which there’s some problems. And we’re going to talk about that in the next series of podcasts. But you can see what he’s trying to do. He’s trying to build, he’s trying to take this concept of open -endedness. And he’s trying to say, well, there’s only real two sources of open -endedness. And so I’m going to declare those the only two sources of knowledge. And then he’s trying to figure out what’s the commonality between these two sources of knowledge, right? And this is this idea that Deutsch comes up with that there’s only two sources, I call it the two sources hypothesis. There’s only two sources of knowledge, biological evolution, and human minds. And it’s a false theory. And I will explain why it’s a false theory. But it’s a false theory that’s getting at something true. And it’s really the problem of open -endedness that it’s getting at, right? This missing part of Darwinian evolution that we don’t really understand today. Or at least I’m going to make that case, okay, in a series of podcasts. And so this is, this is therefore, I mean, basically we’re done at this point. This is my summary, as best I can, of several different great minds trying to deal with this problem of open -endedness, this problem that is present in our current Darwinian evolutionary theory that still needs to be solved. You call it an explanation gap.
[01:21:29] Blue: In some sense, even worse than that, there’s something wrong with the theory of evolution that needs to be corrected so that we can understand it better. And I agree with Kenneth O’ Stanley that it was a great theory that got us quite far, but it’s going to require some new paradigm to break us out of whatever is wrong here. And that will be, that’s the end of what I have prepared for this podcast.
[01:21:57] Red: Well, I found that to be quite a pleasant and open -ended introduction to open -endedness. Thank you, Bruce.
[01:22:06] Green: Yeah, it’s a very interesting problem that I certainly have never thought about before.
[01:22:12] Red: Yeah. Seems to be a common occurrence on this podcast is that you take something I would have never in a million years thought would be so interesting and you convince me that it’s worth thinking about.
[01:22:26] Blue: Yes, thank you. You can see that I’ve picked up a lot of these things because I was trying to understand AGI and I was just finding the best books I could on the subject, which none of them know. But these are gropings towards, okay, there’s something missing, it’s the problem of open -endedness, and then I’ll kind of latch onto that and then I’ll study that more. And I’ve done this for years where I’ve moved from book to book trying to understand them. And this is really what this podcast is largely about is me trying to reveal a lot of these things that I’ve learned from reading up on these subjects. So yeah, I think this is a super interesting subject that I wish I could study more. Oh, I didn’t go into what Kenneth O’ Stanley actually has done to study this. And I actually highly recommend you go read his article because he talks about it in detail. But what he did is he invented something called, he invented a measure of novelty, and he does something called novelty search. And novelty search is a genuine improvement over normal machine learning algorithms that it’s a more open -ended form of machine learning, where like his little robots, virtual robots or whatever, they’ll learn to walk faster than normal machine learning because they use this novelty search. Problem is that the novelty search kicks out at some point. So while it’s more open -ended than regular machine learning paradigm, it’s still nothing close to what biological evolution can do. And it reaches a point where it’s still coming up with novel new things, but they’re basically boring at this point.
[01:24:03] Blue: It’s the things trying to solve a maze and it does so in little zigzag paths. And you’re like, okay, that’s just dumb, right? And it sort of just kicks out at some point. So it’s a piece of the puzzle that he’s figured out, but we’re not sure how many puzzle pieces there are. But I think he’s right. And I’m not sure how, and I would love to interview Kenneth O ‘Stanley, bring him on and ask him questions. I’m not sure how he would justify biological evolution doing novelty search because novelty search is where you measure this new generation did something distinctly different. Like this new robot, its gate is distinctly different than the previous generation, right? So you’re intentionally concentrating on selecting novelty, okay? Where this robot got further to a new location that no other robot’s gotten to. So it’s selecting for novelty. I’m unclear how Darwinian evolution could do novelty search. And I wonder if he’s got theories about that. I mean, like, there’s some kind of some vague ideas in Darwinian evolution. The fact that certain species of birds, their mating calls are based on the creativity of the… I’ve mentioned this in past podcasts, the creativity of the male bird to be able to come up with a unique creative mating song. You can see how that would kind of be a sort of novelty search. Or even just like the fact that some animals have really strong preference to mate with as many other… The idea that they only mate with the most fit members of the species isn’t actually true, right? And you can see that there’s some novelty search in that. But it’s kind of vague, right?
[01:25:51] Blue: Like, I don’t know how you would combine Stanley’s theory of novelty search with biological evolution. And I’m not even clear that you can. Combine them as of today. So that’s something I wish we could bring him on the show and ask him about. So you know what? Let’s… We’ll do that. Let’s… Maybe we’ll reach out to him, see if he’s willing to… He’s a big name. He may not be willing to come on a small show like ours, but… But you never know.
[01:26:15] Red: You’ve had some big names before, so…
[01:26:17] Blue: We have. We have.
[01:26:18] Red: Well, thank you, Bruce. That was wonderful.
[01:26:21] Blue: And for anyone who’s worried about the creationist arguments I just made, I will, through this series of podcasts we’re about to do, I will end with explaining how to actually respond to a creationist and how I was talked out of creationism, and how to go about doing that. People are talked out of creationism all the time, right?
[01:26:42] Unknown: Sure,
[01:26:42] Blue: sure.
[01:26:42] Green: I don’t run into too many people that are creationists, but… Yeah, they’ve shrunk.
[01:26:49] Blue: Well, they’ve shrunk. They used to be really common. Yeah, I know. And most of them have been talked out of it. So I think the simple truth is, is that you can refute creationism, and it’s been refuted in most people’s minds. So anyhow, I will talk through epistemologically at some point how that happens and why it happens and why it’s correct. It just doesn’t happen to be through the arguments that the defenders of evolution have generally used, if that makes any sense.
[01:27:18] Red: Well, look forward to that.
[01:27:19] Blue: Thanks, guys. The theory of anything podcast could use your help. We have a small but loyal audience, and we’d like to get the word out about the podcast to others so others can enjoy it as well. To the best of our knowledge, we’re the only podcast that covers all four strands of David Deutch’s philosophy as well as other interesting subjects. If you’re enjoying this podcast, please give us a five -star rating on Apple Podcasts. This can usually be done right inside your podcast player, or you can Google the theory of anything podcast Apple or something like that. Some players have their own rating system, and giving us a five -star rating on any rating system would be helpful. If you enjoy a particular episode, please consider tweeting about us or linking to us on Facebook or other social media to help get the word out. If you are interested in financially supporting the podcast, we have two ways to do that. The first is via our podcast host site, Anchor. Just go to anchor.fm -4 -strands -s -t -r -a -n -d -s. There’s a support button available that allows you to do reoccurring donations. If you want to make a one -time donation, go to our blog, which is 4strands.org. There is a donation button there that uses PayPal. Thank you.
Links to this episode: Spotify / Apple Podcasts
Generated with AI using PodcastTranscriptor. Unofficial AI-generated transcripts. These may contain mistakes; please verify against the actual podcast.