Episode 77: Counter Examples To Deutsch’s Theory of Knowledge?

• Links to this episode: Spotify / Apple Podcasts

Unofficial, reviewed transcript. It may still contain mistakes; check the podcast when wording matters.

Speakers are labeled by first name where identified. Unidentified-speaker labels may refer to different people. Bracketed question marks indicate uncertain wording.

Transcript

[00:00:07]  Bruce: Welcome back to The Theory of Anything Podcast. Hey, Peter, how's it going?

[00:00:12]  Peter: Hey, Bruce, doing great.

[00:00:15]  Bruce: We are going to continue our discussion about, I guess, just stuff related to Deutsch's ideas about knowledge, his constructor theory of knowledge, that sort of thing. We're probably gonna do several more episodes about this. This one's a little bit of a hash. I just, I found a bunch of interesting examples that I thought were worth bringing up. I thought they were cool. Maybe no one will agree with me. Let's go ahead and I will kind of first do a recap of what we previously talked about. So two episodes ago, we discussed Deutsch's how he defines or understands knowledge in beginning of infinity, which we now know was the beginning of his constructor theory of knowledge. And we talked about how he uses an example of something that is not knowledge, which would be a genetic programming algorithm creating an algorithm that makes a robot walk.

[00:01:10]  Bruce: And how that example actually fits his properties of knowledge perfectly. So it is an example of adapted information that causes itself to remain so. So why does it not count as knowledge? So we took a look at in the last episode, Deutsch's Constructive Theory of Knowledge to see if it had anything to help us understand why that would not be knowledge and the other examples that come from biological evolution and human knowledge do count. And I pointed out that the theory, the Constructive Theory of Knowledge actually has several degrees of freedom in it. So that is places where it's open to interpretation. So by utilizing those degrees of freedom, we can certainly rule out the walking robot algorithm example and declare it to not be knowledge if we want to. But I also showed that every single time we did that, if we held that criteria to declare it not knowledge consistent and then reapplied it to biological evolution and human ideas, that it very quickly started ruling out things that we would want to count as knowledge.

[00:02:17]  Bruce: So for example, if you want to rule out the walking robot algorithm on the grounds that it doesn't last for hundreds or thousands of years, well, guess what? Aircraft designs go out the window too, because they don't last that long either. So it puts us in an interesting spot, because it's difficult to come up with this properties of knowledge that he's talking about, and to eliminate what he wants to eliminate, but not eliminate what he wants to include, or at least defenders of the theory want to include. So I even conjectured that it might be impossible to come up with a consistent set of explicit properties of knowledge that allows, applies only to the two sources, biological evolution and human minds, without accidentally including something else. However, I also admitted that the two sources are special in some way, which is I think the main thing Deutsch was originally getting at.

[00:03:11]  Bruce: Their open-ended searches is what they are. So it's not really that they necessarily output something special, but that they're special in how they're able to search across a very wide landscape. So I understand why there is a desire to call only adapted information that comes through the two sources. Anytime I say the two sources, that's the shorthand for Deutsch's idea that there are two sources of knowledge and only two sources of knowledge, biological evolution and human minds. So anytime I say the two sources hypothesis, that's what I'm actually referring to. So that was kind of what we've talked about up so far. So today we're going to cover some other interesting examples that are perhaps counter examples, or at least examples of adapted information that do seem to match Deutsch's current properties of knowledge, if you're treating them as a criteria, but they aren't from the two sources.

[00:04:06]  Bruce: And we've been using primarily the walking robot example. We're going to look at some of these other examples. And I'm going to argue that these sources of whatever this is, something similar to knowledge that's outside the two sources, are quite interesting in their own right. And I hope these examples will help you see why they seem quite interesting to me, right? Something that we wouldn't want to leave out of our ultimate theory about knowledge. Even if we don't want to call it knowledge, then we want to somehow show it has a relationship to knowledge. Okay.

[00:04:34]  Peter: Can I just tell a brief story before we get going here? Sure. So I was in preparation for this episode. Well, you know, I've been telling everyone how great ChatGPT is, singing its praises. You know, I use it for cooking, for playing these complicated games with my son, rather than looking through the rule book. We just ask ChatGPT. So many uses for this thing. Keep finding more and more, like a lot of people do. But anyway, in preparation for this episode, I asked ChatGPT, which chapter was it that in Beginning of Infinity, that David Deutsch talks about animal memes? Well, straight up, without missing a beat, it told me David Deutsch has never written a book called Beginning of Infinity. And if I want to learn about animals, I should go to the zoo.

[00:05:26]  Bruce: Nice. So would you say it was being disobedient?

[00:05:32]  Unidentified speaker: I don't know.

[00:05:34]  Unidentified speaker: It seemed a little salty to me.

[00:05:35]  Bruce: Look, it's a true AGI now. It's a true AGI.

[00:05:40]  Unidentified speaker: It was probably sick of me asking questions about David Deutsch or something.

[00:05:44]  Unidentified speaker: I don't know.

[00:05:45]  Bruce: Well, you probably do overdo it. I mean, you are a big David Deutsch fan, so you're probably like constantly hitting with David Deutsch questions. It was really just its way of saying, dude, go get a life.

[00:05:55]  Peter: Probably.

[00:05:59]  Bruce: Okay, now let me just say something here, though. I mean, I've been kind of going hard on this the last two episodes, and I'm going to still, I mean, like, that's my job, but let me back off a little bit, okay? And I did kind of do this in the past episodes. I pointed out that I'm making some assumptions here that may not be true. So for example, I'm treating Deutsch saying in Beginning of Infinity that knowledge is adapted information that causes itself to remain so. I'm treating it like it's like a all-encompassing, necessary and sufficient criteria for knowledge, okay? Now, I've seen a lot of people who are fans of his theory of knowledge treat it exactly that way. So I don't feel like I'm entirely out in left field by doing that. But I don't know that he's ever said that's what he was trying to do. You could always respond to me, and maybe even appropriately by saying, okay, this is not supposed to be a necessary and sufficient criteria for knowledge.

[00:06:56]  Bruce: He's just trying to give you a general idea of what he's talking about, okay? Now, here's the thing though. I feel like that's actually a good response. Although it doesn't mean that my question's not valid. If I can take what he said so far about what is the properties of knowledge, and I can show other things are included, that's an interesting fact. And when I have done that with Defenders of the Theory in the past, they've been very quick to just kind of shut me down. Oh no, that's not knowledge, you know? So I feel like this is like a fair question, even if I admit that maybe I'm a little unfairly assuming things about it that I don't know if Deutsch would necessarily endorse or not. Maybe he would, maybe he wouldn't. Now, having said that, let me give a quote from Brett Hall that I think does kind of illustrate where I'm coming from.

[00:07:46]  Bruce: Okay, so a recent tweet from him, he says, this is a quote that he's, I don't know if this is like a real person he's quoting, but he's like imagining somebody saying this. Philosophy is pointless. Don't waste a minute on it. Just do science, math and code, practical things. And then he's responding to that idea, which he believes is wrong. He says, advice taken for decades by those now deep in the trenches of the AI, AGI thing. Fair enough. Most philosophy is a dead end, but some is necessary. Some are slow to realize. And then in the next tweet, he says, so what is necessary? Popper is necessary. Without an understanding of what knowledge even is, how can it possibly be created and how can it differ from information more generally? You'll be stuck on the AI, AGI thing, like trying to discuss stars without understanding fusion. Now, this is kind of how I've been treating Deutsch's Theory of Knowledge up to this point.

[00:08:50]  Bruce: And I don't think I'm off base to have a vibe that this is how people that are fans of the theory look at it. That it is something fundamental about knowledge, it's not just that he's slapping the word knowledge on an interesting idea, that he's actually discovering something important about what knowledge physically is. And he means that in the way that we would normally use that word, knowledge. And it's something that's so important that it's going to be because we didn't understand what knowledge was, we can't really research AI, AGI. That's one of the main things that's held up our ability to figure out AGI, is because we didn't know what knowledge physically was. And he never mentions Deutsch. He actually says Popper in this case. This really isn't Popper's theory. Like we know Popper had a completely different view on this than David Deutsch did.

[00:09:47]  Bruce: And this is clearly actually a reference to the Constructor Theory of Knowledge, conceptually, what he's talking about. And you know what? I'm not even saying I disagree with Brett here. Like perhaps the fact that we don't have a really good theory of knowledge is one of the main reasons why we can't figure out AGI. Or maybe it's vice versa. When we figure out AGI, we'll understand the concept of knowledge better. Let's be honest. The way we use the word knowledge today, it's kind of vague as to what we mean. That's why philosophers have debated it and talked about knowledge only exists in minds, and Karl Popper had to come up with the whole three worlds. I mean, like there's a ton of history around this. Okay, so I'm not even saying I think Brett's wrong, but clearly he is suggesting here that Deutsch has revealed something important about what knowledge physically is and that the criteria we're talking about represents a necessary and sufficient understanding of what criteria declare something to be knowledge or not.

[00:10:47]  Bruce: And what I'm trying to show is the criteria are at once, from the examples I'm gonna use today, they're at once too encompassing and not encompassing enough. And they kind of capture something that's very interesting as far as knowledge is concerned, but doesn't really fit. It removes too many things, it includes too many things. It isn't really what knowledge is, the way we would normally think of that term I mean. Okay, so let me give some examples of this. All right, so not all human ideas are what we would normally consider knowledge. Okay, so one of the things I've wondered is, is Deutsch's Constructive Theory of Knowledge, is it actually a theory of knowledge, as we would normally think of that term, or is it really more a theory of replicators? And what's the difference between those two? That's a fair question. So, Vaden loves to use the example of astrology, okay?

[00:11:43]  Bruce: Astrology is a human idea, it keeps itself replicated, it fits perfectly, and it's even part of the two sources now, it fits perfectly into Deutsch's Constructor Theory of Knowledge. But it would be weird to say astrology is knowledge. And I believe he's actually gone online and challenged Brett Hall and others to say, do you believe astrology is knowledge, and gotten them to kind of say, oh yeah, it is, I guess it fits. But it's weird, right? I mean, is astrology really knowledge?

[00:12:16]  Peter: I'm not sure I'm completely following. What is his argument about astrology being knowledge?

[00:12:21]  Bruce: Astrology is a meme, and it's a meme that's been quite successful at replicating itself.

[00:12:25]  Peter: It replicates itself, yeah.

[00:12:27]  Bruce: So it fits the constructor theory of knowledge, but it's not what we would normally consider knowledge.

[00:12:33]  Peter: Well, it's maybe a primitive form of knowledge. I mean, isn't that where science came from astrology, really? I don't think that's an exaggeration.

[00:12:42]  Bruce: Is that true? I've never heard of that. I do know like alchemy gave rise to chemistry.

[00:12:50]  Peter: I think that science, you could make this, it's a pretty somewhat defensible claim at least, that it came from making observations about the stars. And probably a lot of these guys had weird, super natural ideas about the stars, maybe.

[00:13:07]  Bruce: So, okay, so let's go with what you're saying though. Okay, because like, it's like, let's just assume you're right, because I don't think it's an unreasonable statement, because so many strange ideas that today we would consider wrong, silly, and not knowledge, a lot of times they were precursors to something that we would consider knowledge, okay? But I'm not sure every human idea could even fit into that example, right? Like, would you really consider all the following knowledge? Chain letters, pyramid schemes, humorous social network memes, jokes, right? I mean, all of those are replicators. All of those follow evolutionary epistemology. They're all examples of adapted information that keep themselves to remain so. Like, and they're coming from human ideas. They're human minds. They're part of the two sources hypothesis. And yet I don't know many people that would consider a chain letter to be considered knowledge, right?

[00:14:04]  Peter: You have to look at a pretty broad, big picture definition, I think, to get there. But yeah, I hear what you're saying.

[00:14:11]  Bruce: So here's why I think this kind of works, right? Sorry, let me restate that. Here's why I think it fits into Deutsch's theory, and yet it isn't really what we would normally consider knowledge. Astrology does replicate itself because it contains knowledge, but it is not itself knowledge. I think that would be a more clear way of saying this. So here you might say, well, yes, but it contains knowledge about how to replicate. And if that's what you mean, that absolutely astrology contains knowledge about how to replicate, and that's why it replicates. In fact, that's gonna be true of absolutely any replicator that keeps itself copying itself, okay, from human mind to mind. But understand what you're really saying when you say that, okay? And let's be more explicit so we understand what we're saying. Saying astrology as a meme contains knowledge how to exploit human psychology to get itself copied is nothing like saying astrology is knowledge.

[00:15:15]  Bruce: Those two statements don't strike me as all that similar, okay? And yes, you can confuse the two so that they seem more similar by not being as explicit, right? Being a bit more vague. But it seems like we ought to not do that. It seems like we ought to just be explicit. Is astrology actually knowledge? Or is it a meme that contains the knowledge on how to exploit humans to copy it? Now you start to see a relationship between astrology and knowledge, but it isn't the same thing as being knowledge. This is why I wonder if [unclear word] Constructor Theory of Knowledge may not actually be a theory of knowledge at all. It may actually be a theory of replicators, okay? Or something in between, like I'm not sure, right? I don't want to become overly dogmatic about an alternative view when I really feel like I'm mostly trying to work it out for myself at this point.

[00:16:07]  Bruce: Now, if the former way is a better way to put it, which I think it is, to say astrology contains knowledge, how to exploit human psychology, I think that is a far more accurate statement. That's why I kind of wonder if it really is a theory of replicators. Maybe not even a full theory of replicators, because for example, the walking robot example, that algorithm was created by replicators. So you can't even say it's a theory of replicators. It leaves out certain replicators. Furthermore, a lot of human knowledge is not replicated. Okay, so let's take the idea of trade secrets. Now a trade secret is this idea that knowledge is so valuable that you don't want to replicate it. So you want to hold on to it, keep it a secret and not let anybody else know that you've got this secret, okay? This still makes sense. If knowledge is useful, if knowledge gets copied because it's useful, it may also not get copied because it's useful, okay?

[00:17:06]  Bruce: For the exact same reason. So imagine that somebody, let's use Chiara Marletto's example from her book, that somebody has invented an airplane and they have this design of this airplane, but it gives them a competitive advantage, okay? So they do not want anyone else to know how they make airplanes or how they use them or what they do. And they try to keep it, I don't know how this would work in real life, but we're just coming up with a hypothetical example to make a point, okay? So are you prepared in this case to declare that aircraft recipe, not knowledge, because it wasn't replicated, okay? How far are you willing to go to follow along with what Deutsch is saying and take these kind of counter examples and remove them from being knowledge and say, no, no, that is not knowledge, right? Because it's not being replicated. I mentioned that I had discussed this with someone who I was calling Henry, and he actually did declare trade secrets to not be knowledge.

[00:18:05]  Bruce: He said, oh yeah, trade secrets are not knowledge because they don't get replicated. Okay, I mean, like it's just a word. So if for him, maybe the word knowledge means that, but I think most people would kind of tilt their head and go, what, are you serious? And I want to kind of pull that out here, okay? Okay, plus if trade secrets aren't knowledge because they're not replicators, then what the heck are they? Okay, surely there's something of great interest that is a lot like knowledge, even if we're not gonna apply the word knowledge to it. So it still feels like there's like kind of a fair question being asked here, okay? So this brings up something kind of interesting. Under Deutsch's Constructor Theory of Knowledge, astrology is knowledge and a trade secret isn't. Or could we maybe use the degrees of freedom that are in the theory to adjust that?

[00:18:57]  Bruce: Well, we could. And of course, this is the problem with degrees of freedom. So for example, Brett Hall, in his episode 198, Bayesianism, around minute eight, right around minute eight, he suggests that we might see a situation like this as still being a replicator by simply being a bit more vague on what counts as a replicator. So for example, he talks about a bridge embodying knowledge. Here's the actual quote, the bridge itself contains the knowledge it really does. By the way, I agree with him on this. I'm not trying to argue that he's wrong. And episode 198 Bayesianism, I don't think there's anything in that episode that I actually disagree with. I felt like it left out some, all the interesting questions I wanted to actually have answered, but I feel like he makes fair points all the way through. One of these days we'll do an episode on that episode as I thought it was a good episode.

[00:19:49]  Bruce: What he seems to have in mind here is that knowledge starts as firing in neurons in his brain, and then it turns into, he says, taps on the keyboard, and then pixels on a screen, then becomes sound waves in his podcast. Then it goes, you listen to it, and it goes back to neural firing in the brain. Okay. So if we're going to use, oh, by the way, he's taking this from Deutsch, of course, Beginning of Infinity, page 86. The knowledge in us and in our sundials embodied in the watch and in the mouse, right? So Deutsch has talked about how objects can embody knowledge like this, right? Well, this changes how I understand replicators, okay? If we're going to take, kind of make the concept replicator more vague, and it's just how we transmit knowledge, then, okay, so that kind of shortens what counts as a replicator quite a bit, maybe down to microseconds in some cases, okay?

[00:20:44]  Bruce: So here's the thing, though. Now imagine someone with a trade secret that isn't about building something, okay? Because we talked previously about maybe the constructor theory of knowledge requires us that we actually construct something. Well, okay, so let's say nothing ever gets, we talked about the airplane design. You build the airplane. At that point, the airplane itself is a replicator, right? Do you see what I'm trying to say? But even that, it's easy to come up with a counter example, too, because we might imagine a trade secret that is something like, say, a technique that increases sales compared to your competitors by 200%. Okay, so now is that not knowledge? Because, and for the sake of argument, you might say, well, it gets replicated to you teaching it to all the people in your company. Well, let's say that you're just a single salesperson and you figured out this technique, you're the top salesman, and you don't want anybody else to know what it is you're doing that gets the top sales, okay?

[00:21:40]  Bruce: So at this point, you're not transmitting this knowledge anywhere. It stays in your brain and that's it, okay?

[00:21:48]  Peter: So what I was going to push back on is, sounds like you're kind of addressing now, is that from in some sense, something being a trade secret could actually inhibit replication, since a company has to be motivated to create this new kind of knowledge. And if they aren't able to keep it secret, then they won't have the motivation to do that. And so, I don't know, on some larger level, it still seems to me that a trade secret could be knowledge. Is that kind of the perspective you're addressing here?

[00:22:35]  Bruce: I mean, I personally think trade secrets are absolutely an almost quintessential example of knowledge. What I'm really trying to show is that trade secrets aren't, strictly speaking, replicators, and yet they are knowledge, right?

[00:22:48]  Peter: It helps them replicate within a company or something.

[00:22:51]  Bruce: Right. And that's why I was trying to say, what are we going to count as a replicator? I mean, if the mere fact that I built the airplane means it got replicated, okay, that's fine. I'm just trying to get you to think more carefully about what you mean by replicator. Okay. So let's say that now we take this one where the guy in the company, he has the knowledge himself to do the increased sales, but he doesn't want to share it with anybody. So it's not even spreading through the company. Well, you might still call it a replicator by simply saying, well, it's an idea and he has to enact it. And the act of enacting it is a replicator of the idea in his mind. Okay. All right. So, okay. Right. I mean, now, but now you're weakening the concept of replicator even further. Right. Now, anything that gets enacted once counts as a replicator. In doing that, what you've done is you've now opened up back to the charge.

[00:23:48]  Bruce: Well, okay, the walking robot algorithm, if it walks once, it counts as a replicator, right? I mean, are you willing to accept that also? Okay. And you're starting to see the problem is that there is this inconsistency with how it gets applied.

[00:24:04]  Peter: Okay.

[00:24:06]  Bruce: Now, meanwhile, while astrology is knowledge, the Constructor Theory of Knowledge declares the immune system when it's creating antibody recipes to not count as knowledge. Now, I've got an actual quote from David Deutsch coming up on this, but this is something that's interesting that I just kind of wanted to bring up quickly. So I was talking to a defender of Deutsch's Theory of Knowledge, and I was trying to explain to him the things I'm bringing out of this podcast, how it seems to me that there's problems with the theory that need to be addressed. And I was raising these counter examples, and in context, what I actually said was, okay, you know what, I'll accept that these counter examples aren't knowledge. But like these counter examples, and we were talking about the immune system as one of them, and I think the reason why we were talking about the immune system was at the time, Deutsch had just recently in an interview declared that the immune system didn't create knowledge.

[00:25:06]  Bruce: Okay, so we were discussing, is that accurate or not? And I said, okay, look, you know what, I'll accept it's not knowledge, but it is something, right? It's something interesting that it's doing that is very much like knowledge. Okay, in fact, it fits all of Deutsch's criteria that he's ever brought up at least so far perfectly. And this person, who we'll call James, he responded to me by saying this. He says, but I'll hold you responsible to say what you find so special about adapted information created by the immune system. It doesn't seem special to me because the trial and error process of the immune system is 100% deterministic. So the result is two and is thus a mechanical consequence. I see therefore nothing special or relevant about this kind of adapted information. Now, here's what I find interesting about this response. There's a number of interesting things about this response, and I'm actually going to cover it again in a future podcast, specifically the fact that it's kind of indirectly citing the whole perspiration versus inspiration argument that deserves its own podcast.

[00:26:18]  Bruce: But the thing that I want to call out here is this idea that what the immune system is creating, he's admitting it's creating adapted information here, right? But it's so unimportant and so non-special and so uninteresting that it doesn't even deserve to be discussed, okay? And I do think this is how I've seen defenders of Deutsch's Constructor Theory of Knowledge treat these examples, where when I bring them up, even when I'm not trying to say they are knowledge, and I'm just trying to say, look, they're something, okay? It's just sort of like, this is so uninteresting. There's no way that there's anything interesting there for you to look into. Now, let me actually explain to you how the immune system works. Okay, I'm taking this from Dennis and Ray Noble, one of their papers, and it's very interesting how it actually works. To me, it seems quite special and interesting, okay?

[00:27:15]  Bruce: But first of all, before I do that, sorry, here's the actual quote from David Deutsch. Okay, so Deutsch explicitly states that the immune system does not create knowledge. This is an interview with Eli Tyer around minute five. He says, There are things happening in the human body, like the immune system, which is also reacting to things, doing one thing, stabilizing itself, doing another thing, changing levels of importance, and so on. None of that is thinking, none of that is creating knowledge, okay? By the way, note the attempt to set knowledge creation to exclusively be a kind of thinking, which isn't really necessarily what his Constructor Theory of Knowledge is about, because it includes biological evolution, which is not a kind of thinking, okay? But in any case, he definitively declares the immune system to not be knowledge here, which is why I was discussing this with James.

[00:28:04]  Peter: Now, the first definition of knowledge that I ever recall really hearing from David Deutsch, and I wish I could remember the source if this was one of the talks or the book, was actually maybe the non-constructor theory definition of knowledge, where he says just knowledge is useful information, which I thought was so cool because it just seems like such a sensible, it just almost seems to cut down all of these philosophers talking about knowledge. What is it? Where does it? It's just useful information. That's all it is. It seems to me in that definition, the immune system is, yeah, that's knowledge, useful information. Yeah.

[00:29:03]  Bruce: Now, having discussed this with David Deutsch directly, he definitely uses the word knowledge in a number of different ways.

[00:29:09]  Unidentified speaker: And he told me so, right?

[00:29:11]  Bruce: So perhaps he would even say, yes, the immune system creates knowledge in a sense, right? But it's not the constructor theory of knowledge sense. The problem is that it seems to me it is the constructor theory of knowledge sense. And I guess that's why I'm getting hung up on it, right? So let me explain to you how the immune system actually creates knowledge. I shouldn't say knowledge because that's what we're discussing, how it creates adapted information that's useful. Okay. So Deutsch, as I just quoted, he explicitly states the immune system does not create knowledge. But how does the immune system actually work? Well, it works through recipes and replicators. It does. Okay. It directly involves, and here's the interesting thing, replication of DNA in cells.

[00:29:57]  Unidentified speaker: All right.

[00:29:58]  Bruce: So here is quoting from Dennis and Ray Noble from their paper, Was the Watchmaker Blind or Was She One-Eyed? It says, When the immune system is faced with a new antigen challenge, the mutation rate in the variable parts of the genome can be accelerated by as much as one million times. So far as known, these mutations all occur stochastically, which just means randomly. But the location in the genome is certainly not a matter of chance. The functionality in this case lies precisely in the specific targeting at the relevant part of the genome. The arrival of the antigen itself activates the hypermutation process, and its binding to a successful antibody triggers proliferation of those cells that make it. What this mechanism achieves is the DNA sequence forming a template for the immunoglobin protein are held sufficiently constant for functionality to be retained.

[00:30:54]  Bruce: Okay, that was a lot of big words. Let me try to break it down. Okay, so let's deconstruct what they just explained on how the immune system works compared to Deutsch's Constructor Theory of Knowledge. First, the arrival of the antigen itself activates the hypermutation process. So what they're saying is that the antigen, the germ, activates a hypermutation process in DNA of the immune system cells. The immune system actually has a learning algorithm inside of its cells when it detects an antigen. It knows to start hypermutating DNA to try to find the right recipe to fight off the antigen. The mutation rate in the variable part of the genome can be accelerated as much as 1 million times. So we're not talking about the stereotypical mutation where a cosmic ray hits your genes and it happens to create some sort of mutation or it's a copying error or something like that.

[00:31:55]  Bruce: The immune system says, oh no, we have an antigen, we better start mutating fast and it intentionally mutates itself to try to find the right recipe. So the genome intentionally hypermutates to find the right recipe for the antigen. Then they say in the paper, so far as is known, the mutations all occur stochastically. So the cells randomly mutate to find the right variant. Okay, then they say an antigen binding to a successful antibody triggers proliferation of those cells that make it. So what they're saying here is, if the antibody binds to an antigen that signals the cells that made that antibody to replicate those cells and therefore their DNA proliferate compared to the cells that weren't successful. Note how this is specifically DNA replicating, genes replicating, depending what you mean by genes, replicating in cells. So it is a recipe replicating as per constructor theory, David Deutsch's Constructor Theory of Knowledge.

[00:32:54]  Bruce: This fits completely perfectly to Deutsch's Constructor Theory of Knowledge. And by doing this, the successful cells will out-produce the unsuccessful variants, and that's how your body knows to produce that antibody to take out the antigen. So the immune system literally uses DNA and genes as the substrate through it, and the only difference is that it doesn't do it through sex cells. It does it within the life of a single organism, and it doesn't pass it to the offspring of that organism. Okay, so why does this not count as knowledge? So let's compare it to the criteria or properties of knowledge that Hervé had came up with, that everybody agreed was a good summary of David Deutsch's Constructor Theory of Knowledge. It is capable of enabling its own preservation via replicators even. Yes, the antigen binding to a successful antibody triggers the proliferation of those cells that make it.

[00:33:53]  Bruce: That is exact match to the first criteria. It can be copied from one embodiment to another without changing its properties. That's literally how the immune system works. Number three, it can enable transformations and retain the ability to cause them again. Without a doubt, that's exactly what's going on. I do not see why the immune system doesn't count as knowledge creation when it literally uses DNA in the genome. It's literally a replicator. It's literally a recipe. It literally causes a transformation even in the sense of a construction. I don't think there is any chance you will come up with a set of properties or criteria that eliminate the immune system as counting as knowledge without basically eliminating all of biological evolution. Because it is physically doing exactly the same things and outputting the same thing as biological evolution does.

[00:34:44]  Bruce: It just does it in a much shorter time scale so that it can find an antibody within the lifetime of the organism, which makes sense. If you didn't have a system that could create knowledge in real time, like this inside the lifetime of the organism, then you would be at the mercy of every single antigen that didn't exist prior to your birth, right? So it makes perfect sense that evolution would learn to self-mutate within the lifetime of an organism to be able to do this. It's really hard to see how you could ever declare the immune system as not creating knowledge without basically eliminating all of Darwinian evolution because of this. Short of insisting on something like it must pass through sex cells to the next generation to count as knowledge, which, okay, you know what, that's fine. If you want to actually state that explicitly, I'm okay with that.

[00:35:31]  Bruce: I think that then makes the theory quite a bit more clear. But unless you're willing to go that far, the immune system should count as knowledge. Okay, that is, it creates exactly what we think of when we think of the Constructor Theory of Knowledge. So I hope people can see why this example makes me feel so skeptical of the two sources hypothesis. The immune system does create knowledge under the current Constructor Theory of Knowledge the way it's currently presented and understood.

[00:35:59]  Peter: Has David Deutsch ever addressed Ray Noble's ideas, which could be considered sort of a modern-day Lamarckism almost?

[00:36:13]  Bruce: I have not seen him do that, no.

[00:36:15]  Peter: Deutsch comes down pretty hard on Lamarckism. I mean, it's mentioned a lot in The Beginning of Infinity. I don't think he, you know, as I thought was just pretty uncontroversial, that it's not taken seriously anymore. But it sounds like it, well, from speaking to a lot from you and a lot of other followers around David Deutsch, maybe not, I shouldn't say followers, but people associated with his ideas seem to take this Ray Noble stuff pretty seriously.

[00:36:54]  Bruce: So, okay, let's do an aside. I'll attempt to do that. I won't do a great job with this because it wasn't what I was planning to talk about. Okay, so here's the thing. What do you even mean by Lamarckism to begin with? Because there are certain, if you think of Lamarckism in kind of its original inception, like clearly it's wrong.

[00:37:19]  Peter: Well, the reason I use that term specifically was just because I know in the Ray Noble debate with Richard Dawkins, which is an excellent debate, but he even says, he says, Lamarckism is back, you know. So anyway.

[00:37:36]  Bruce: So here's the thing, though. The idea of Lamarckism is a somewhat vague idea. And so it has long, I mean, like this is not something new with Dennis and Ray Noble. It isn't. Okay. Do you remember? Well, you [were/weren't] part of the podcast back when we talked about The Baldwin Effect. Did you ever listen to the episode where we talked about The Baldwin Effect?

[00:37:56]  Peter: I did. I can't quite remember what it is, though.

[00:37:59]  Bruce: So The Baldwin Effect is this idea that... So like one of the things that fans of David Deutsch bring up a lot when they're trying to explain why they think animals are just automatons that don't have feelings is they'll show like animals doing automatic movements, right? Dennis Hackethall has got a giant collection of white swans he's collected of animals doing things that don't make sense because this automatic movement has been triggered. And David Deutsch has referenced this, too, the idea that you can take, in one of his interviews, that you can take a squirrel and you can give it food and put it on cement, and the squirrel will actually go through the motions of trying to dig into the cement, even though nothing has happened. It will try to bury the nut, and then it will try to use its nose to put the dirt back over it.

[00:38:48]  Peter: I see my dog doing things like that all the time.

[00:38:51]  Bruce: Okay, so they have this idea that this somehow proves that the animal is just a robot. Now, the Baldwin Effect points... So there's a problem that the Baldwin Effect needs to solve. The problem is that under Neo-Darwinian evolution, it's theoretically impossible to explain how these automatic movements came about on their own, because the automatic movements, like the movement of a squirrel trying to dig by itself would have no survival value. The movement of a squirrel trying to cover the nut with dirt would have no survival value.

[00:39:28]  Unidentified speaker: Okay.

[00:39:29]  Bruce: So this whole little sequence that it does, that's this set of automatic movements, like Dennis Hackefall's assuming somehow Neo-Darwinian evolution just evolved this, even though this is exactly an example of what a creation ist would point to and say, look, how did this ever come to be when every single individual movement has no survival value? Therefore, you cannot explain this using Neo-Darwinian evolution. Okay. Well, now, turns out that's not true. But you do have to understand Neo-Darwinian evolution in a somewhat more broad way than the way people used to think of it. Okay. So Baldwin, who's I think was early 20th century. So this is a long time ago now. A hundred years ago. Baldwin was a scientist who came up with an explanation of how to explain how these automatic movements came to be and to reconcile them with Neo-Darwinian evolution.

[00:40:23]  Unidentified speaker: I guess at the time it wasn't...

[00:40:25]  Bruce: I'm not even sure the term Neo-Darwinian evolution existed in the early 20th century. But reconcile with Darwinian evolution, I should probably be saying. So what he pointed out is that animals can create knowledge in real time, in their lifetime, through learning algorithms. Okay, so they go out and they... You would have an animal, some ancestor to the squirrel, that would go out and would figure out on its own, using its own learning algorithms, that if it buried a nut and it went through a series of movements, that that had survival value. And then that passed as a meme to all the members of that species. Okay, so the original progenitors had to have figured it out on their own through their own knowledge creation, okay? Once that exists as part of the environment, then it makes sense that each individual automatic movement has survival value.

[00:41:18]  Bruce: So if one of these animals that's in this population suddenly gets an automatic movement to say, try to bury or the nut, even though that's not by itself has any survival value, within the environment of this set of memes that exist amongst the animals, that now has survival value. And therefore, over time, this learned maneuver that exists as a meme gets put into the genes of the animal, and thus the automatic movement comes to be. That's actually how they believe this ever happens in the first place, which to my mind actually is undermining to the whole argument that the Deutschians are trying to use here. Although, maybe not. I mean, it's vague. It shows that there's something missing from their argument at a minimum. Now, what does this have to do with Lamarckianism? People immediately started to recognize that this was very close to a form of Lamarckianism.

[00:42:13]  Bruce: And if you go look up the Baldwin Effect, which is what this is called, it is often today, like go, I don't know, look on Wikipedia or something, you will very quickly start to come across people the statement that this is a form of Lamarckianism. What is this actually Lamarckism? Well, no, it's not really the original Lamarckism that's discredited. It does have a vibe similar to it. And I think that's why people invoke it here. Okay, is clearly we're no longer talking about kind of the original Darwinian understanding of how evolution works. We're starting to think a little bit more deeply about it, that there is this sort of... You have to understand the survival value in terms of not just a set of genes, but also the environment that exists, which might include whatever is learned as memes within the life of the animals. And if you want to call this Lamarckism, great.

[00:43:15]  Bruce: If you don't want to call it Lamarckism, then don't. They're invoking the fact that it's in some way vaguely analogous to Lamarckism, I think. And I guess I see Ray and Dennis Noble in the same way. Yes, their theory is that we have vastly overestimated the value of gene informatics, and that in reality, evolution exists in a giant hierarchy. It's not just evolution of genes. This is exactly what Popper and Campbell said way back in the 60s or the 70s. And this was an idea that's existed for a long time, that when you actually look into Darwinian evolution, you do have to understand it as this universal Darwinism, that it exists in this giant hierarchy of knowledge creating algorithms that exist all throughout nature, all throughout, and it's not just genes. And I do think that Ray and Dennis Noble are at least correct that neo-Darwinian evolution has over-emphasized gene informatics.

[00:44:21]  Bruce: And therefore, I kind of agree with them to some degree there. Now, I do wonder, though, if maybe they've gone too far, where, I mean, don't we basically know that gene informatics, if it's not the only form of evolution, it sure is in some ways one of the single most interesting and somehow it must be related to the universality of the biological evolution process. So I guess I fall somewhere in between, right? Yes, I think neo-Darwinian evolution is ultimately wrong, but I think it's wrong in only the sense that we're going to have a new theory that encompasses it, subsumes it, and still explains why it was such a successful theory. And I don't think it's going to truly... Michael Levine's work, I think, has shown over and over again that the cells exist with their own kind of learning algorithms that are an own mini version of evolutionary epistemology.

[00:45:25]  Bruce: And I think we have underestimated the role of all these different levels of evolution that exist, and how that probably has allowed... I don't know how yet, but has somehow allowed the problem of open-endedness to be solved. And I suspect one of the reasons why we can't solve the problem of open-endedness today is because we've been misled by the current Neo-Darwinian evolutionary theory. We've thought too much in terms of evolution of the genes, and that's it. And we haven't really thought about the true evolutionary environment that exists out there in the real world, that exists in the giant Campbellian hierarchy. Campbell put it as a hierarchy. I think Ray and Dennis Noble would say it's not even a hierarchy. It's like this giant cycle of interacting evolutionary processes. And it's more complicated than a simple hierarchy that Campbell had in mind.

[00:46:23]  Bruce: And because that it's actually this giant cycle, yes, it's got a much stronger Lamarckist feel to it than people have believed for the last 100 or so years. But I don't think it's actually Lamarckism. I think that it's still not really the original Lamarckism. It's more like something kind of analogous to Lamarckism, but that's it. Does that make any sense?

[00:46:50]  Peter: That does. And one thing I noticed in that debate I mentioned earlier, Richard Dawkins, who, as far as I know, invented neo-Darwinism and in the selfish gene, seemed quite open to the idea. I mean, he didn't quite agree with, I'm not saying he agreed with Dennis or Ray Noble, but he seemed quite open to the idea that Lamarckism could make its way back into science somehow.

[00:47:21]  Bruce: Let's actually define Lamarckism a little bit more carefully to understand. The term gets invoked in two ways, at least. One is the original idea that evolution takes place because the giraffe goes out and it stretches its neck, and then the neck becomes a little bit longer, and then it somehow passes that to the next generation. Okay, that's the discredited version of Lamarckism. But I think Lamarckism also gets invoked for any kind of teleological evolution. Anytime evolution has a purpose, then it feels a little Lamarckist to us. Okay, now let's look at the immune system example. Of course, there's an exact [teleological/theological] purpose going on here, right? The reason why the immune system starts hypermutating is because it's been designed to recognize, oh, I better start doing an evolutionary algorithm now, because I'm in a situation where the genes that have been passed to me are insufficient.

[00:48:25]  Bruce: And so of course, it's a form of evolution that's very purposeful, right? And so if all you mean by Lamarckism is purpose-driven evolution, then yeah, Lamarckism is a real thing. And it has always been a real thing. But I don't think that's really what the word originally meant. I think that that is an adaption. It's an evolution on the word, if that makes any sense, right? The fact that the original discredited Lamarckism was purpose-driven caused us to re-imagine the word Lamarckism to refer to any kind of purpose-driven evolution, even a kind that's compatible with our current theory of evolution. And I think that's where the tripping up comes, right? Yes, Lamarckism has made a comeback, but in a mutated form that's not really the same as what it originally was. And if we're really being honest, this new version really isn't incompatible with what we thought we knew about Darwinian evolution.

[00:49:25]  Bruce: It's forcing us to understand it better, to make it more explicit, to understand its processes better, to understand how purpose can originate into the evolutionary process without there being an ultimate designer. And if that's all you mean by Lamarckism, then yes, Lamarckism has made a comeback.

[00:49:45]  Peter: Controversially though, right? I mean, not everyone accepts this.

[00:49:49]  Bruce: Well, you know, honestly, I don't think there's... I think it's not... Yes, maybe controversially, but it shouldn't be. The immune system is without a doubt a purpose-driven form of evolution. So if you don't think that there is any purpose-driven forms of evolution, then I think that your theory is refuted by the existence of the immune system, and I think that's kind of the end of the story. Whether you want to call it Lamarckism or not, that's a different question.

[00:50:19]  Peter: Got it.

[00:50:21]  Bruce: Okay. What would a more encompassing theory of knowledge look like? If we weren't looking at just these special kinds of replicators that come from the two sources, I think what we would want is we'd want to be able to have this theory explain what trade secrets are. If they're not knowledge, that's fine, but how do they relate to knowledge? Why are they so much like knowledge? I would expect it to explain what the walking robot algorithm is, not to just simply try to declare it unspecial or uninteresting or something along those lines, but instead say, this is what it is, and this is how it relates to knowledge. I would expect it to explain what the immune system is doing, how it is related to the constructor theory of knowledge and what it outputs is similar to, but not quite the same, as knowledge. So let me throw some other counter examples out here now, examples, I should say.

[00:51:16]  Bruce: And I think each of them is kind of ones I've grabbed that got me thinking more deeply about this, so I'm offering it up to other people to get them thinking more deeply about it. Now here's one that I thought was super interesting. Deutsch actually told me when I had a chance to talk to him that he doesn't think all human ideas are knowledge, and he gave me this example of how they came up with the gravitational constant and how he didn't feel that that was knowledge creation because it was actually implied by other theories that already existed. Okay, based on that, he actually says in his interview with Eli Tyer at about minute 10, he says, so Eli has been giving him examples of things that create knowledge, and David Deutsch has been saying, I don't think that's actually knowledge creation. And the immune system was one of the examples, but he was giving several examples.

[00:52:07]  Bruce: And then Eli Tyer gives this example.

[00:52:11]  Peter: Who is this Eli Tyer?

[00:52:12]  Bruce: He's just someone who did an interview with David Deutsch. If you want to look it up, type Eli Tyer, T-Y-E-R and David Deutsch, and you'll quickly find this interview.

[00:52:23]  Peter: Okay, I'll look for that.

[00:52:24]  Bruce: And he said, Deutsch said a lot of really interesting things in this interview, so it's actually worth looking up. Okay, so Eli says this. What about just playing out logic? Like I start with a set of premises, and there is implicit in those premises a set of conclusions. I might not start out knowing the conclusions of some logical premise, but if I sit down and think and write it out on a piece of paper, I'm like, this is, this mathematical proof is actually kind of surprising to me, but it turns out this is the case. I knew all of the pieces all along to derive that knowledge, but I actually needed to sit down and derive it, and at the end, it seems like I have new knowledge. Okay. So this is the example he kind of throws out to David Deutsch to see how David Deutsch is going to respond. Now, I want you to think about this example for a second.

[00:53:14]  Bruce: Like, think about geometry or really any mathematical knowledge, okay? Think about computational theory, okay? Granted, it can be thought of as a form of physics, but it can also be thought of as more like a mathematical theory, where you're trying to work out the implications of the fact that a Turing machine is universal, and you're trying to come up with the set of different types of algorithms, P, NP, things like the NP-complete, things like that, okay? Sorry, Eli Tyer is really saying, is all of that not knowledge to you? Okay, like if I work out theorems for geometry, and they all flow directly from the axioms, does like every single thing I discover about geometry from that point forward, because it all is really just derivable from the axioms, is that not knowledge? Is every single time we make advances in computational theory and discover something new using computational theory, is that not knowledge? Because it was something we just

[00:54:18]  Bruce: derived via proof and it was all sitting there, part of the premises, but we just didn't know, right? And is that discovery not knowledge? Okay. And here's how David Deutsch answers that question. It's a good example because I also think that isn't knowledge either, but that looks much more like a simulacrum of knowledge that the previous example that Eli had raised, um, than the previous examples that Eli had raised. Okay. Now I was a little surprised he said this. Um, although I really liked the term simulacrum of knowledge and I've latched onto it a little bit. Um, so, so just to put this in perspective though, why it surprises me, he said that suppose somebody finally came up with a proof that in computational theory, P equals NP. Don't need to know what exactly that means, but that would be like a gigantic breakthrough in our knowledge to know that, right?

[00:55:18]  Bruce: This huge breakthrough in knowledge, Deutsch is literally claiming it is not knowledge creation if it's something that was just derivable from the premises, okay? Now on the one hand, I kind of understand where he's coming from. It does feel a little different, hey, this is all just part of the premises. On the other hand, you just have to think about every single advance in knowledge we've had We've had in mathematical systems, and we're kind of declaring it all not knowledge, and that seems a little weird to me, okay? But I love this idea of a simulacrum of knowledge. So Deutsch is at least admitting that there is something knowledge-like being created there, but he doesn't consider it to be the same as knowledge. And I'm going to come back to this phrase, a simulacrum of knowledge in a future podcast, where I think something interesting can be derived from it.

[00:56:05]  Bruce: But let's move on for the moment. Now let's also take the idea that all of an animal's knowledge is in its genes. Now from the way I'm talking, you can pretty much tell, I don't really think that that's correct. But let's go with this just for a second. So all of an animal's knowledge is in its genes. What is meant by that under the Constructor Theory of Knowledge? Since the Constructor Theory of Knowledge doesn't allow for animals to create knowledge in the life of an animal, because the only two sources of knowledge is biological evolution in the genes and human ideas. So by Fiat up front, we're declaring that all an animal's knowledge has to be in its genes. It cannot create any new knowledge in its lifetime.

[00:56:45]  Unidentified speaker: Okay.

[00:56:46]  Bruce: Now animals adapt their behavior via a special kind of learning algorithm. And I mentioned in a past podcast, Leslie Valiant, who's a big name in machine learning, he calls this ecorythms. Now on page seven of his book, he says, the primary purpose of the ecorythm is to change the circuits of the organism so that they will behave better in the environment in the future to produce a better outcome of the owner. That's on page seven of his book, which is probably approximately correct. So an ecorythm is intentionally defined in such a way as to capture two kinds of learning. On page seven, these ecorythms have been of two kinds. Those that operate in the individual organisms interacting with their environment, and those that operate via genetic changes over many generations. So notice that Leslie Valiant is intentionally defining ecorythms so that it encompasses both what David Deutsch considers a kind of knowledge, the knowledge created by Darwinian evolution.

[00:57:48]  Bruce: But he's trying to show that the learning of an animal in its lifetime is analogous to that. He's trying to show that there is this learning that takes place. Now, I happen to know that David Deutsch doesn't actually deny that animals learn. Typically, this is kind of stated in such a way as it's kind of downplayed of its importance, but I don't think there's any denial that animals do in their lifetime learn things. And that they do form adapted information in their nervous system that isn't part of their genes. So Leslie Valiant sees that there's some kind of definitive commonality between these two kinds of learning and that you can actually think of them as a single class. Now, it's interesting that this is a machine learning expert who invented the probably approximately correct model of machine learning, which underlies all of machine learning or a significant portion of it anyhow.

[00:58:44]  Bruce: And he's actually deriving the same thing that animal experts have been saying. For quite a while now. So animal experts, entirely independent of Valiant, who's in computer science, have actually discovered the same thing. So here's Richard Byrne from The Thinking Ape on page 52. He says, instrumental conditioning. So actually, let me back up and explain this. People are really familiar with this idea that animals have something called classical conditioning. So the typical way is Pavlov's dog, where you ring the bell and then you feed the dog, and after a while, it starts to realize that ringing the bell means it's about to get fed, so its mouth salivates. And this is a kind of learning algorithm that animals have. What I think people don't realize, though, is that classical conditioning actually has a stronger form that animal experts don't call classical conditioning.

[00:59:40]  Bruce: A layman like me would still call it classical conditioning, but it's called instrumental conditioning, or sometimes it's called trial and error learning. Burton actually uses both terms and says that they're equivalent to animal experts, okay? Now, the term trial and error learning, the problem is that all learning is trial and error. So when I say that, it seems to cause people confusion. But these are just terms that the animal experts have come up with on their own. They're not epistemologists. They've said, well, we've got animals have classical conditioning, and then they have this thing called instrumental conditioning or trial and error learning. Here is what it is defined as. Instrumental conditioning has been likened to evolution by natural selection. Like natural selection, it is an automatic process by which behavior patterns with beneficial consequences are selectively increased in frequency amount the many types of behavior produced by the animal, among the many types of behavior produced by the animal.

[01:00:34]  Bruce: The mechanism is quite different in the two cases, but when he's talking about instrumental conditioning vs. biological evolution, the mechanism is quite different in the two cases. However, in instrumental conditioning, reinforcement increases the probability of advantageous responses to what the animal perceives in the world. Unhelpful responses become extinct, but the animal lives on. Natural selection, by comparison, is more drastic and slower to produce change, since whole individual animals survive or die and the selection effect on the genes is therefore much less efficient. Now, notice that he's immediately seeing that there's some sort of an analogy between how an animal learns through this instrumental conditioning or trial and error learning and how biological evolution through natural selection works. The difference is this. The animal has an action it does, and let's say it decides to eat something, and then it gets some sort of reinforcement.

[01:01:39]  Bruce: Either it gets sick or it finds that it keeps it alive and it likes the taste. Instrumental conditioning is really just classical conditioning, but it's based around the animal's own actions, its own curious actions that it goes out and uses to explore the world. And as it receives this feedback, it then realizes, oh, I should eat that kind of food. That food is good, because it didn't make me sick. Or it gets sick and it says, I should not eat that food and it won't eat that food again, okay? And what it's doing is, it causes its behaviors to either increase if they cause the organism to survive and thrive or to go extinct. Its behaviors to go extinct, if it causes the organism to feel sick or to get hurt or something unpleasant, okay? By doing that, you have a form of natural selection that takes place within the life of the animal and the animal learns within its lifetime, okay?

[01:02:41]  Bruce: There's just no way around the fact that there's a really obvious evolutionary epistemology connection between what animals do through their learning algorithms and what takes place in biological evolution. So the rapid change of behavior in response to a changing environment, that is through learning, that is a survival advantage, because then the animal doesn't have to die to figure out it shouldn't be eating this item or it should be eating this item, okay? This is how animals learn. This is how animals create knowledge within their lifetime. So also I note here that this is a form of the idea of letting your ideas die in your place. Only in this case, instead of it being a strict idea, it's kind of more like the animal's behaviors die out or increase depending on whether they provide survival value or not. But this is an exciting parallel that Popperians should be interested in.

[01:03:39]  Bruce: And by the way, Popper was interested in this parallel. And he talks about it quite a bit in his books, okay? Now, once you realize animal learning is a kind of evolutionary epistemology, and it's something that is analogous to natural selection, it does call into question this whole idea that all of an animal's knowledge is in the genes. And let me give you a very specific example. I used this example on the Increments podcast, which I'm sure will be out before this episode ever comes out. But let me repeat it, okay? So some birds have knowledge. I think I may have used this example in our animal intelligence episodes. I can't recall now. Some birds have knowledge in their mating, knowledge in their genes of a mating song. So in that case, if you pull the baby bird away and raise it away from its parents, it will still know its mating song because it never has to be taught it by its parents, never has to hear it from its parents to learn it.

[01:04:38]  Bruce: Because the genes actually encode for that mating song. That's a very direct example of that knowledge existing in the animal's genes, okay? However, that's not how all birds do it. Like many birds, they have to actually learn the mating song from their parents, okay? Now, they do have genes that help them learn the mating song, help them kind of have a preference through their attention for one type of song over another. So what they'll find in that case is that the bird, if you separate it from its parents, it never learns the song because the knowledge of that song does not exist in its genes, okay? However, if it learns it, if it hears the song even once, it seems to retain that song, take a lot of interest in it, and then it will actually practice the song over time through trial and error and learn to perfect the mating song. And they've actually been able to like teach the mating song to the bird backwards by playing the song backwards, and it learns that one instead.

[01:05:38]  Bruce: So they know it's not in the genes because it can actually learn the song backwards, but they can't get the bird to learn a different bird's mating song because it's not similar enough to what the genes find, have changed the bird to find interesting. Whereas learning the song backwards is similar enough. Okay. So now we have first was an example of where the knowledge of the [unclear wording] is directly in the genes. Now we have this example where the knowledge really isn't in the genes, but the genes help it learn the knowledge. Now here's where things get really interesting. Richard Dawkins in an interview, he said that there's also birds that have no preset mating song. And instead what they do is they just start to sing a song and they just try out different notes. And they'll suddenly notice through their objective sense of beauty that certain notes go together better and that make a better song.

[01:06:32]  Bruce: And through trial and error, it will actually compose its own song. And if the song is really beautiful, then it attracts the female birds. And so basically what happens is is the bird uses its own creativity to try to come up with a mating song and it tries to impress the females by making it the most creative mating song. Now here's also Dawkins points out that these songs these birds compose, humans find them beautiful. So this fits directly with Deutsch's idea of objective beauty where it crosses species, okay. So the birds are actually using their sense of aesthetics that they have to try to come up with a mating song and they compose their own creatively.

[01:07:20]  Unidentified speaker: Okay.

[01:07:21]  Bruce: Now it seems weird to take all three of these examples and try to collapse them down and say that all of them are examples of a bird having all its knowledge in its genes. You're collapsing three very different things and treating them as if they're a single thing when really these are very distinct different kinds of mating songs and very distinct different kinds of where the knowledge comes from. Okay. The last one in particular, where the bird creatively creates the song, I don't think it makes any sense at all to claim that that was knowledge in its genes.

[01:07:55]  Peter: So- Can I just interject one thing?

[01:07:57]  Bruce: Yeah.

[01:07:57]  Peter: There is a documentary on Netflix that I have got to recommend to you. It's called Dancing with the Birds. It is one of the most fascinating documentaries I've ever seen. It's all about these, not just songs, but dances and elaborate- Interesting. I mean, basically, they are performing Swan Lake for these mating rituals. Like what these birds do was, it just blew my mind. I had no idea the extent of it. I mean, to say it's not creative is a stretch.

[01:08:32]  Bruce: Right.

[01:08:32]  Peter: Anyway.

[01:08:33]  Bruce: Well, there's other examples of this too. Like there's certain birds where they have to build like a little piece of artwork to try to impress the females.

[01:08:41]  Unidentified speaker: Oh yeah.

[01:08:41]  Peter: That was the other thing that, yeah, they get into that too. Yeah. The artwork.

[01:08:44]  Unidentified speaker: Yeah.

[01:08:46]  Bruce: And so basically the females are going to mate with the most creative of the birds, right? They're going to judge the bird's creative output to decide, oh, this is the most fit one because it has the highest creativity.

[01:08:59]  Peter: Oh, the best is the female will go over and inspect it. And then they'll kind of say, ah, no, no, right.

[01:09:10]  Bruce: Okay. So now another one that has come up that I think is an interesting example is, again, Dennis Hackathall, who I think uses a lot of great examples that I can then use as counter examples. In the Do Explain podcast, he's a defender of the two sources theory of knowledge. And he wants to explain how animals don't create knowledge in their lifetime. So he uses this example of a dog playing Jenga. Okay, well, now this is, to me, this is not a very good, I mean, this is an example of how animals actually do create knowledge in their lifetime, because clearly there can't be any evolutionary pressure for a dog to know how to play Jenga and for that to be encoded in its genes. So Dennis feels a need to try to explain this away. So he tries to say, do this through a kind of vague explanation that the knowledge was in this genes because it had these instincts, okay?

[01:10:07]  Bruce: So what he tries to say is he says, okay, it's this combination of the dog has knowledge in its genes via an instinct. In this case, the instinct is to follow a pack leader, he says, and then he imagines that this is somehow a clever Hans trick. So the dog has this knowledge in its genes to follow a pack leader, and then the human is the one who knows how to actually play Jenga. And the dog is figuring out how to play Jenga by looking at the human's face and figuring out what to do based on what the human's face is telling it to do. Okay, now I should probably point out that the video he points to as an example of this, it has right on it, a video of a cat playing Jenga that is not looking at its owner's face. And it's just playing Jenga, right, for fun. It even tries to cheat at one point. So I feel like this whole example kind of undermines his point to some degree.

[01:11:01]  Bruce: But here's where things get interesting. What Dennis is really trying to say is that the dog came genetically equipped to read the face of the owners, from the face of the owner, the movements necessary to not knock down a Jenga tower. Okay, there seems to be a willingness on the part of the defenders of the two sources hypothesis to vaguely point to any instinct an animal has and cite it as if that somehow just explains everything. Now, the obvious counterexample here would be that if I scolded my dog, go build an airplane, and it somehow went and built an airplane because I was scolding it, it would make very little sense to claim that it was just following its pack instincts to obey me as the pack leader, right? I mean, there's something more going on here if it's doing something that we wouldn't have expected it to be able to have evolutionary pressure to do.

[01:11:51]  Bruce: So there's no real attempt to ever determine why the dog has this ability to read Jenga moves from the owner's face in the first place, given that there is no evolutionary pressure to do that. And it's actually worse than that, okay? Can you really explain Jenga moves as a kind of Clever Hans trick? You can explain the Clever Hans trick because all it had to learn to do was to read that the owner was smiling, so it stopped tapping its foot. That's clearly something that could fit within just regular trial-and-error learning for an animal, okay? But in this case, here's the challenge I guess I would issue to Dennis on this. Go try to actually make an algorithm that plays Jenga by doing nothing but watching your face, Dennis, okay? I don't think that's even physically possible because the face doesn't contain the necessary information. Like, if Dennis is looking at a tower and his robot is trying to play Jenga, there's no way that through just a visual feed of his face, it's going to understand how to play Jenga, right?

[01:12:53]  Bruce: Like, that's physically impossible, as far as I can tell. So there's, like, this giant disconnect, explanation gap between how he's trying to explain it through instinct and just the reality, which is the dog learned to play Jenga within its lifetime, and of course it did, right? It basically learned to play Jenga the same way you and I would learn to play Jenga. Now, I've seen this other places where defenders of Deutsch's Two Sources Hypothesis will just simply invoke instinct like it's somehow an answer. So here is a defender of Deutsch's Two Hypothesis Theory, and here's what they said to me on Facebook. We were talking about if animals can have goals or not, and he said, animals don't have goals, they have instinct. Again, you are confusing things humans can do with how animals behave. Having a goal means seeing the end state of something that is not realized yet.

[01:13:50]  Bruce: Using goals in any serious way when talking about what humans do versus animals is simply crazy. So I asked him how he knows this, and he says, they, animals, are not able to do anything that requires the ability to create an abstract plan involving several steps. Now, when he said that, of course, I knew I had already, and I pointed him to my podcast where I actually talk about this at length, what Richard Byrne says about this. So here is the example that's from my podcast. Let me repeat it. This is Supina the Orangutan, and one of my favorite examples. So Supina the Orangutan, as I'm giving you this example, ask yourself if animals have goals or not. Seems to me that we got a really straightforward example of an animal having a goal here. So the goal is that Supina the Orangutan has seen people wash clothes, and it looked fun. So she wanted to go wash clothes.

[01:14:50]  Bruce: Here's the problem, though. She wasn't allowed to do that. So the soap and the laundry being washed is on a camp dock, and she knows that the staff that is washing the clothes are afraid of her, so she knows if she can get near them, they'll run away, or they'll jump into the river and swim away. But there's also a guard protecting them, and she's afraid of the guard. Supina is afraid of the guard. So she has a problem. She has a goal. She wants to wash these clothes, but she's got a set of problems that she needs to solve to be able to do that. Here is the solution that Supina creatively comes up with to achieve her goal, using honestly what seems to me a rather complex plan, not human level complex, but quite complex. So Supina first steals a canoe full of water by untying it. She stops and she goes and she checks to see if the guard is still there and seeing that the guard is still there, she then continues.

[01:15:48]  Bruce: She then rocks the canoe to get the water out of the canoe. She reorients the canoe and propels it forward. When she arrives at the dock, having now crossed the river to the dock, the staff jumps into the water like she knew they were, to escape. She then goes about using the soap that they left behind to clean the laundry. She rubs the soap, the wet clothing. She uses a brush with the soap. She scrubs the clothes with the brush. She wrings the wet clothes, et cetera. Now, this is such an interesting example, because it's obvious she had a goal. She wanted to wash the soap and laundry, and this is a rather complex plan that she's come up with. Furthermore, she's never seen a human do this before. You can't claim that this is just simple imitation, because she has never seen a human being decide to go wash clothes by doing this particular set of actions to be able to do it.

[01:16:43]  Bruce: Furthermore, the set of actions, they're not just random. I mean, they're so obviously set up to solve a problem, okay? And she had to understand how each action connected to the next action to be able to solve the problem that she wanted, set of problems that she wanted to solve, okay? This really does show some really flexible intelligence. And it also shows an example of animal creativity. Again, I want to emphasize this particular chain of events. No human goes around stealing boats so they can go wash clothes, right? She's never seen another being or animal or ape or anything do this before. She came up with it herself, okay? Now, she has seen the individual movements. She's seen humans use boats. She's seen humans untie boats. She's seen humans rock boats to get the water out. I don't know how many times, probably not hundreds, but she's seen it a few times, okay?

[01:17:39]  Bruce: And just having seen those, being able to imitate those individual movements, it was enough for her to figure out how to put them into a sort of program, an algorithm where if I do this one, then this one, then this one, then this one, it will solve my problem. Without ever having to try it once, try it by chance herself, she figured it out in her head, and it was obviously aimed at solving a goal. Okay. So this does require that we see animals as having a certain level of understanding, okay? This is why Byrne says regular animal learning versus instrumental conditioning or trial and error learning can't explain what's going on, and why he believes that some animals, including orangutans, have this extra level of intelligence that he calls insight, okay? So the animal must actually understand this chain of events well enough to understand how to chain them together into something totally novel to solve a goal, solve problems to achieve a goal.

[01:18:35]  Bruce: This is why Byrne considers the animals with insight as having a, and this is a quote from Byrne, a deep, shrewd or discerning kind of understanding, okay? Now, the reason why I'm emphasizing the word understanding is because, again, David Deutsch has said animals don't understand things. In one of his interviews, he was challenged by the interviewer as saying dogs understand things, and he says, I don't think they do. Well, I don't know if dogs do or not, but subpoena of the orangutan clearly does. And this is a form of animal intelligence. And it goes without saying that the animals do not evolve behavior to wash clothes, nor to use boats to sneak past guards. So I don't see how to see this as anything except animal creativity plain and simple. I think the only thing you could possibly do to try to explain this away is you could call it a coincidence that the animal by chance happened to untie the boat, by chance happened to go check the guard, by chance happened to rock the boat to get the water out, by chance happened to go cross the river.

[01:19:32]  Bruce: Unless you're prepared to explain the whole thing as a coincidence, we do have an example here of animals having goals, having creativity, and coming up with knowledge that's entirely outside the genes. Okay. Now, there is one last example I really want to cover, and I think it's probably one of the most important examples, and it creates the most massive problems for the Constructor Theory of Knowledge. But I think it deserves its own podcast, so I'm not going to go into it this time, and that's animal memes. I think that one of the reasons why I want to cover that in detail is because David Deutsch writes a lot about animal memes in Beginning of Infinity, and I want to actually be able to quote him, go through what he says, and kind of talk them through why he believes they're not knowledge and why I feel that they actually do fit the Constructor Theory of Knowledge quite well.

[01:20:24]  Bruce: Let me just kind of summarize this at this point, though. So these counter examples are meant to get you to really stop and think. They don't say individually anything definitive. And I'm not really trying to say they are or aren't knowledge. I mean, I call them knowledge, but that's like just for convenience, right? The word itself is meaningless. What I'm really trying to get out here is that they offer insights, that you really get you thinking about, you think about them in terms of the Constructor Theory of Knowledge, and it forces you to really think harder about the Constructor Theory of Knowledge. Knowing all these counter examples leaves me with the impression that the properties of knowledge that David Deutsch currently talks about, either in his book, The Game of Infinity, or Chiara Marletto talks about in her book, The Science of Can and Can't, they do not seem like they are both necessary and sufficient set of properties to define knowledge.

[01:21:22]  Bruce: And they even leave me with the impression, particularly the immune system example, that there may be no necessary and sufficient set of properties that you can come up with that will define knowledge and somehow only keep it to the two sources. Because there are so many other sources of this simulacrum of knowledge that exist out there that are so similar to knowledge and in fact indistinguishable from knowledge that I honestly wonder why we're trying to reserve the word and not include them. And then at the same time, I want us to think a little more carefully about what knowledge means, right? The astrology example, the antibodies example, why are those not knowledge? How do they relate to knowledge if they're not knowledge? And I want us to really acknowledge these are interesting examples, like what the immune system is doing, what an animal is doing when it comes up with a creative plan, what an animal is doing when it creates a creative song, you know, to try to dismiss those as that's nothing, right?

[01:22:23]  Bruce: I think understanding what's going on there is related to the question of AGI. And I think that if your theory of knowledge excludes those so you don't even think about them anymore, that that will be a hindrance to you trying to figure out what AGI is, not a help at that point.

[01:22:39]  Peter: I think that's a really compelling point, but just so our audience doesn't get the wrong impression, please correct me if I'm wrong, but you know, it sounds like you're poking all these theories in Deutsch's ideas about... What did I just say? Poking holes in Deutsch's ideas about knowledge, but it still seems to me, big picture, you would fundamentally agree, there is something special about the kind of knowledge that human beings create that could, you know, get us to the moon and beyond, and deduce what happened at the beginning of the universe and all this.

[01:23:25]  Bruce: Absolutely. Absolutely. Okay. Now, you say it's not the same. Then, of course, I have to ask, what do you mean by that?

[01:23:32]  Peter: Okay, of course, of course.

[01:23:33]  Bruce: Okay, so let's say, I've used this example elsewhere. Let's say that a group of creative scientists come up with an antibody to stop the plague or something. Okay. Or let's say the immune system comes up with the exact same antibody. If what you mean by they're not the same, in this case, I'm saying, yeah, they are physically exactly the same. Yes, they are the same. Okay, so yes, if what you mean by knowledge is that it's adapted information that keeps itself instantiated, I suspect animals create knowledge every bit as much as humans do. However, that statement is very misleading, and I think that's what you're getting at, right? Humans, when they create knowledge, they're able to do it to a degree and to an extent that animals clearly can't. Animals are clearly capped in their knowledge-creating abilities. And one of the things I really liked about Byrne's books is that he makes the case that you really have two levels of animal knowledge creation, animal learning.

[01:24:38]  Bruce: You've got kind of the regular kind of animal that just does it through instrumental conditioning or trial-and-error learning, where it just randomly tries stuff and then it learns what works and what doesn't. And then you've got the second kind of animal, the animals with insight, where like subpoena, the orangutan, they can actually try out ideas in their head and they can try to string together different actions to accomplish goals. And yet both of them still clearly have some sort of cap. So it's like you got two jumps in universality and then you got the third jump with humans, where suddenly it's like the Turing machine.

[01:25:14]  Unidentified speaker: It's now everything.

[01:25:16]  Bruce: In many ways, knowing that animals are creating knowledge, but that they have a cap, that's really important to whatever's going on with AGI, right? Keep in mind that we don't even understand the animal learning algorithm. We don't even know what that is. It's something amazingly better than what we're currently doing with our machine learning, okay? So if you really want to understand AGI, one of the questions you should be asking yourself is, what is it that makes humans able to not have a bound on what type of knowledge it can create in its lifetime, whereas all animals have a bound? And there's even two types with different bounds, but they're both bounded. What's the difference between those? Like, in some sense, that is a way of going about trying to research AGI. If you have in your mind that animals don't create knowledge in their lifetime and that all their knowledge is in their genes, you don't even know to ask that question, okay?

[01:26:20]  Bruce: Because you think that there's exactly one jump, and it just takes place, and there's only knowledge through Darwinian evolution, and there's only knowledge through human minds, and there's nothing interesting about asking, well, wait a minute, apes, what can apes do that humans can do, and what can humans do that apes can't do, right? And to know that that's worth asking, that they've got some sort of really amazing knowledge creation algorithm, but in yet it's still bounded. That really is related to the research of AGI, which is why I am emphasizing this. Like as my interests in AGI exist, understanding animal learning is actually a step towards understanding AGI, because there has to be some sort of little tiny gap that took place that leapt from, you know, subpoena the orangutan to humans. I think Deutsch has pointed out that it has to be less than a few megs, right?

[01:27:16]  Bruce: It's like some tiny amount of difference in the genes that made the whole difference. So it's got to be something pretty small. I also want to say this. Well, I don't think there's a physical difference in knowledge created by animals and knowledge created by humans. I do think there's a drastic difference in the search capabilities. This is something I haven't detailed much, so I don't know if it will make sense to the audience yet. But I think when we talk about evolutionary epistemology, it's always a kind of search. You're always searching through possible solutions to some sort of problem or set of problems. And animals have a very limited search space, and all machine learning algorithms have a very limited search space. And for whatever is going on, there are two search algorithms, biological evolution and human minds, that have this giant search space and seem to progress through that search space at a rate way beyond anything we would have guessed was possible.

[01:28:19]  Bruce: And what is it that they're doing? My guess is that they're constraining, they're pruning the search tree in some really interesting way that we don't currently understand. And that's why humans can include in their knowledge creation going to the moon, where an animal can't, is because that exists within their search space. They're able to jump through different problems and different solutions in some way that is so constrained that they can get all the way to traveling to the moon, where an animal just could never do that. Whether we're talking about biological evolution or animal learning algorithms, neither could ever get an animal to the moon, would be my guess. Honestly, I don't know that for sure. We can almost imagine a science fiction story of some animal where there's these selection pressures that cause it to learn to get to the moon, but it's still just an animal.

[01:29:09]  Bruce: But I doubt it.

[01:29:11]  Unidentified speaker: I doubt that's actually possible.

[01:29:13]  Peter: You give a dog a trillion, trillion years, it's not going to figure out how to divert an asteroid that's heading towards Earth. Seems unlikely.

[01:29:22]  Bruce: I have one more kind of appendix statement I want to make. When I was talking to Henry about trade secrets, he actually did say that he didn't consider trade secrets to be knowledge. He made an argument that I want to bring up and I want to address in case anybody else has this in their mind. I want to explain why it's not, in my opinion, a valid argument. So when I brought this up, he said, well, if you're going to count trade secrets as knowledge, then you're bringing the idea of a knowing subject back into the theory of knowledge, and that's exactly what we're trying to get rid of. So here's a quote from Chiara that seems to be what he intended. So in her book, she says, knowledge is defined entirely via counterfactuals. It is information that is capable of remaining instantiated in physical systems. Unlike most definitions of knowledge, the good thing about this one is that it does not depend on their being and knowing subject.

[01:30:22]  Bruce: This is one of the main things that Popper was trying to accomplish with his Three Worlds, and this is one of the main things that seems to have motivated Chiara and David to come up with their Constructor Theory of Knowledge was to try to define knowledge so that it is not defined. When I say defined, I don't mean like a definition of a word. I mean like what properties exist for knowledge that create the set of knowledge, right? What we call knowledge. They really wanted it to be defined in terms of what it physically was, so that there was no reference to a knowing subject. And if you don't know anything about philosophy, there's this long-standing debate among philosophers about how knowledge requires a knowing subject. It requires that there's a person that knows it, that if the knowledge exists in a book and there's not a knowing subject, then it's just a collection of ink on a page.

[01:31:15]  Bruce: It's not knowledge anymore. And Popper wanted to show that that just wasn't true, right? That knowledge actually exists objectively. It doesn't require just a subjective experience in the mind. So I understand why Henry is making this argument to me, but I feel like he's fundamentally misunderstood the argument here. The criterion question that we're discussing when he brings this up is stated as capable of enabling its own preservation. Now, I never adapted that. We did discuss whether it needs to include via replicators or not, okay? But neither of us is actually adding to this theory of knowledge anything about a knowing subject. I think what he's confused over is that I'm giving him an example with trade secrets of knowledge that is only contained within one human mind and therefore is a knowing subject. So I think that that makes him uncomfortable because he wants this theory of knowledge to not be about knowing subjects.

[01:32:22]  Bruce: Here's the problem though. He's confusing the defining of the term of knowledge or the properties of knowledge in terms of a knowing subject versus including a knowing subject as one possible instantiation of knowledge. So I'm surprised he didn't actually see the difference between this. I tried to explain this to him, and he was quite adamant that I was misunderstanding. But I don't think I was. This actually seems like a significant failure to me in the way he's reading the Constructor Theory of Knowledge because a knowing subject is the quintessential example of knowledge. If human beings create ideas in their mind, that is what we mean by knowledge, right? The fact that whether it gets replicated or not doesn't matter. If you have a trade secret, that trade secret is a kind of knowledge, even if it's only in your head, right? I mean, I opened to the possibility that maybe we could think of it as not being only in your head if you have to enact it or something like that, but let's stick with just the basics here.

[01:33:22]  Bruce: The idea of the trade secret is only in your head. Yes, it still counts as knowledge. I think the issue here is that neither of us, neither Henry or I, are defining knowledge in terms of knowing subjects, but mine includes a knowing subject, and his is trying to exclude knowing subjects. And I think that's just an outright mistake, right? Because a human having knowledge in their head, that is a kind of knowledge. The Constructor Theory of Knowledge doesn't rule that out. It just simply doesn't define it in terms of the knowing subject. That's really what we're trying to do here, is define it so that the knowing subject is not part of the definition, not rule out knowing subjects as being a kind of knowledge. So anyhow, that would be my appendix to this, and my response to that criticism. Any questions? I think actually I'm done.

[01:34:19]  Peter: Another great episode, and I've enjoyed listening to your thoughts. And thank you, Bruce. Look forward to next time. All right.

[01:34:28]  Bruce: Thank you very much. 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 Deutsch'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.

[01:35:18]  Bruce: The first is via our podcast host site, Anchor. Just go to anchor.fm slash four dash strands, F-O-U-R dash 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 fourstrands.org. There is a donation button there that uses PayPal. Thank you.