Episode 39: Byrne’s Methodology for Discovering Animal Insight (part 3)

• 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

[00:00:10]  Bruce: Welcome to The Theory of Anything Podcast. Hello, everybody.

[00:00:13]  Unidentified speaker: Hi, Bruce. Hi, Bruce. Hi, Bruce.

[00:00:16]  Bruce: We've got everybody today, Cameo, Sadia, and Tracy. We're continuing talking about animal intelligence today, and today we're gonna talk about Richard Byrne's methodology in trying to determine which animals are intelligent. I'll have to explain what I even mean by which animals are intelligent, because that's so many loaded terms right there. And so I'm kind of excited about today because I really am impressed with how Richard Byrne goes about this. He is, I don't think he knows anything about Karl Popper, but Karl Popper would be proud. Good example of how, if you're a good scientist, you are a Popperian, you may just not know it. Richard Byrne, who just, as a reminder from the last couple episodes, I discovered Richard Byrne through David Deutsch's book, The Beginning of Infinity. David Deutsch used Richard Byrne's studies to show that animals didn't understand anything and really emphasized their lack of explanatory knowledge.

[00:01:08]  Bruce: When I actually read Richard Byrne's books, I discovered that Richard Byrne interprets his own theories very differently than David Deutsch does and started to become interested in what Richard Byrne's point of view was. So Richard Byrne is, he's specifically, like his whole career is that he wants to figure out when animals started to think. You know, he believes that it was sometime before, maybe not too long before humans. Now, what do we mean by think? Richard Byrne does not mean human level explanations. So we're not disagreeing with the central thesis of Deutsch's position that animals—his, the most important part of Deutsch's position is animals don't have explanatory knowledge like humans, human level explanatory knowledge. Byrne does not disagree with that. But Byrne does believe that a few animals do have the ability to think, and he's trying to figure out which ones do.

[00:02:03]  Bruce: And this is a difficult thing to come up with because it's possible to get false positives because of genetic pre-programming that looks intelligent to us, but in fact is just automatic behavior. And also because, as we've discussed the last two episodes, animals have trial and error learning. They can actually create knowledge on the fly, knowledge in the sense of adapted information, not necessarily in other senses of the term. This is in and of itself a kind of intelligence. So trial and error learning is a form of intelligence and, you know, almost all animals have it. I don't know which animals don't have it. I'm not sure single-celled animals have it. It may be that certainly animals with a brain stem and above all do classical conditioning or trial and error learning, but I'm not sure how many before brain stems do it, if any. Now, I've been saying single-cell animals don't do trial and error learning, but it occurred to me after the last episode that in episode 21, Evolution Outside the Genome, we talked about Michael Levens or Levines—I don't know how to pronounce his name—work where he created some bots out of frog cells and that these cells learned new exploratory behavior.

[00:03:17]  Bruce: They would start doing things that frog cells don't do, and they would explore their environment and they would look around and things like that, trying to find food sources. So I don't know if that would count as trial and error learning or not. That—maybe, maybe not. So let's just say that I don't know for sure that single cell animals don't do trial and error learning, but I'm assuming they don't. For our purposes, we're going to assume they don't, that it was later animals that developed that ability and therefore would be considered somewhat intelligent, whereas a single cell animal would not be.

[00:03:53]  Sadia: Just to kind of just be a little fair, I guess, to be fair to Deutsch, like he might say that those frog cells, they are actually, what they're doing has actually been like, we have actually programmed to do that. In the sense that what they're doing is just like with our programs, we have evolutionary programs too. But having said that, I guess, yeah, I mean, I guess we do understand that there is a type of learning at the genetic level, like we're not, I mean, if we don't restrict ourselves to a single organism, but overall, you know, overall, the evolution of an organism is sort of, you know, evolution of life is a sort of learning, I guess.

[00:04:37]  Bruce: Yes.

[00:04:37]  Sadia: But you're saying, but you're specifically pointing out here that it's not just at the genetic level, right? That's the point or?

[00:04:45]  Bruce: So we know that we know that most animals that we think of, which would not be single celled animals, do trial and error learning. We know that. That was what the last two episodes were about, particularly the last episode. And that is a kind of intelligence and is learning that's entirely outside of the genome. I don't know. I'm not claiming that single celled animals do that. In fact, I would have thought they didn't. The only type of learning that I thought they were capable of was habituation. That's certainly the classical kind of learning that a single celled organism does. But some of Michael Levin's work does question that a little. But like you said, it could be that maybe the cells had ancient programming in there that was, in fact, just genetic automatic behavior that to us looks like exploratory behavior, trial and error type behavior.

[00:05:36]  Bruce: It's really hard to tell at this point. So I'm going to start with the assumption that single celled animals don't do trial and error learning. And therefore, we can consider them non-intelligent even in that sense. Does that make sense? However, trial and error learning, even though it is a form of intelligence, is not the type of intelligence Byrne is looking for. He would consider trial and error learning to be unintelligent the way he would use the term. In fact, he really doesn't use the word intelligent or understanding at all. That is kind of what he's looking for. His book is called The Thinking Ape, he's trying to figure out the evolutionary origins of intelligence. That's the subtitle. But when you actually look at how he goes about it, he dispenses with terms, popular terms like thinking and intelligence pretty quickly. And he gets way more specific about what he's looking for.

[00:06:29]  Bruce: And you would have to do that, right? I mean, because the word intelligence is so vague and could include trial and error learning, then like all animals are intelligent, and that's not really what he's looking for. He's trying to figure out when they develop something more meaningful than that. So he's upfront, discounting trial and error learning as the kind of intelligence that he's looking for, even though that may count as intelligence in the layman sense. It's not the sense he's looking for. Does that make sense?

[00:06:58]  Sadia: Yeah.

[00:06:59]  Bruce: So then he looks into, OK, what kinds of intelligence, what does the word intelligent mean to other people? And he points out that there's been various scientists who have attempted to define intelligence in various ways. So he looks at some of their definitions of intelligence, and then he says, OK, can animals do this or not? So one of the ones he gives an example of is the learning set. So a learning set, I'll describe that in just a second. It requires seeing the logical connection between different problems and generalizing. So as Byrne puts this from his book, The Thinking Ape, generalizing across the details of individual problems, realizing that the rule is the same. To notice such rules, animals must characterize the rewarded stimuli in abstract terms. And then he goes on to say, this is page 151 and 152, learning sets therefore test the ability of the animal to make somewhat abstract generalizations.

[00:07:53]  Bruce: So he gives the example of an experiment that's been done with a lot of different kinds of animals called picking the odd one out. So in this case, they're able to be trained to get a reward by picking the odd one out. But can they do it on a novel set of items? They're not just simply trained on a specific set of items, but can they generalize to a novel set of items? And he says, according to experiments, many animals can do a learning set, all the way down to rats and squirrels, he says. So we talked about, David Deutsch gave the example of squirrel automatic behavior and suggesting that they don't really understand anything. They can actually do learning sets, right? They can abstract in this really kind of very vague level of abstraction, certainly not a deep level of abstraction, but they can do something like this. Now, here's the thing that's interesting, though, they're not necessarily great at it.

[00:08:47]  Bruce: And when you get to like primates, the primates get better, are better than other animals at it. And in fact, the closer the primate is to a human in the evolutionary tree, the better they are at learning sets. It's a graduated improvement, according to Byrne, with like apes, you know, chimpanzees, which are the closest to humans being the best at it. And then like lemurs being somewhere down there, better than a rat or a squirrel, but, you know, nowhere near the level of a chimpanzee. Animals do have at least that much ability to abstract, you know, all the way down to like rats and squirrels. This isn't the kind of intelligence he's looking for either, though. OK, so when we talk about, you know, when did animals, which animals are intelligent, he's not including this concept of intelligence either. He also talks about another possible definition that scientists have used, the ability to concatenate desperate [wording unclear] facts.

[00:09:39]  Bruce: So he asked the question like this, he says, can an animal concatenate, this is a quote from page 152, concatenate knowledge learned in two previous circumstances in order to deal with a third novel circumstance. Now, this, this sounds like when I put it like that, this sounds pretty convincing. Oh, yeah, that would be that would really require intelligence. But he points out that rats easily qualify because, in fact, any, any animal that can do classical conditioning easily qualifies because if you really think about it, a rat, you give it some food, it gets sick, so it tries to avoid that food in the future. You're never giving it the exact same food, right? It's it has to somehow generalize its knowledge from one one point in time to another point in time. If that's all you're looking for, then hey, that's classical conditioning. All animals, you know, when I say all again, I mean, from brain stems on up, are able to do that.

[00:10:31]  Bruce: So that's not really that impressive either, if that's all you're looking for. And so that's not what he's looking for in terms of intelligence either. However, he does go on to say that he would actually, as a side note, consider, though, that this is actually somewhat impressive. We kind of downplay classical conditioning. We treat it like it's some mechanical rule, but we don't know how to really do it with AI. It's like really hard to do something like classical conditioning. Certainly, we can't do it with a level of generality that an animal can do it, right? We would require either thousands of human labeled examples through supervised learning, or we would have to have a human program the world space into it for a reinforcement learner. We have no idea how to make a robot that can do dog level classical conditioning with, and you don't get told in advance exactly which problem it's going to have to solve.

[00:11:21]  Unidentified speaker: Right.

[00:11:22]  Cameo: It's interesting because I think, especially in the artificial intelligence world, sometimes we're dismissive of natural intelligence without really acknowledging just how impressive it is.

[00:11:36]  Bruce: Yes.

[00:11:37]  Unidentified speaker: Yes.

[00:11:38]  Bruce: And you have to actually be in the field of artificial intelligence and be trying to code it to know how little you actually know.

[00:11:46]  Unidentified speaker: Right.

[00:11:47]  Bruce: We take it so much for granted that a dog can do classical conditioning or trial and error learning, and that that's unintelligent, when in fact we don't have a clue how to go program either of those.

[00:11:58]  Unidentified speaker: Right.

[00:11:59]  Cameo: So I think it also illustrates some of the problem with the words that we're using to describe the emerging technology that is things like machine learning.

[00:12:09]  Bruce: Yeah.

[00:12:10]  Cameo: Because the phrase artificial intelligence implies a level of sophistication that isn't there. Like if you tell people, like this isn't even close to as simplistic as what your dog does when he realizes that you get the leash, he should respond by getting excited because you're going to walk, right? Like if you said there is no ML in the world that can do that on the fly with new sets of circumstances without being programmed to do them, I think a lot of people would be surprised.

[00:12:45]  Bruce: Yes, I agree. And I think from having talked with a lot of kind of Popperians, Deutschians on the internet, I don't think they even realize this is the case. Right. They kind of downplay animal intelligence as stupid, which is somewhat true.

[00:12:59]  Sadia: I'll be honest with you. I don't understand. Like I would love if there was a way for you to do a show because me not being in that field, you know, I do wonder like what is what what is it that we can’t do there? Like with AI, why is it not possible to have an AI where to certain responses it says, hey, I'm excited, you know, or some sort of response like that?

[00:13:22]  Bruce: We can. It just would be very specific, right? So first of all, I did do shows on this. So you can go back and you look at the reinforcement learning episode, for example, where I've actually given the mathematics behind how reinforcement learning works in machine, machine intelligence, machine learning, I should say. You have to program in the world space. So like if I know that what I need is for the dog to respond excitedly because it's going to get its leash. Yes, RL or supervised learning could do it. But if I don't know in advance, that's what the rewards are going to be based on. There's just no way to get the machine learning to do it, right? I mean, it's dogs are on the fly programming the world space from to use the terms of reinforcement learning. Right. And we don't know how to do that.

[00:14:13]  Unidentified speaker: Right.

[00:14:13]  Cameo: Because you have a little organism that in one situation sees you getting the leash and gets excited because that means it's going on on a walk. And then in the next minute sees you opening the fridge and realizes you get meat out of the fridge when you open the fridge.

[00:14:28]  Sadia: Oh, so we're talking about, you know, semantics, like meaning here, right? It seems like dogs do understand meaning to a certain extent.

[00:14:37]  Cameo: No, I don't think we're talking about it from a semantic standpoint. We're talking about it from a as new things are presented to the dog, it has the ability to learn a new reaction set based on that new learning. Whereas for a machine learning program, you would need to go in and tell it about the new situation and train it to respond to the new situation versus it being a machine that can do it itself.

[00:15:04]  Bruce: So it isn't necessarily about meaning. And this is actually an important point. Like we don't know how dogs do trial and error learning, but there's no particular reason to believe that it's about meaning. It could be for all we know, right? But what it really is, is that a dog can be introduced to a leash and it can figure out what a leash is conceptually as an abstraction. And you can pull in a different leash that looks different. It will still know it's a leash. It somehow abstracts the concept of a leash. And if I knew I needed to abstract specifically the concept of a leash, I could do that via supervised learning. But the dog, no one knew that leashes were going to be invented, right? And the dog had to have had some intelligence that existed evolutionarily before leashes. And so it has been able to learn, and it's not just leashes, refrigerators, right?

[00:15:56]  Bruce: You open the refrigerator and now it knows, oh, we went to the refrigerator, that means it's going to get meat. Oh, you're going to the door, you're putting your coat on. That means it's going to get to go on a walk, right? It's abstracting concepts that never got labeled for it in the first place. It's doing it through what we would call, in machine learning terms, unsupervised learning.

[00:16:14]  Unidentified speaker: Of course… [overlapping speech unclear]

[00:16:15]  Sadia: I guess in the case of a machine, we're always supervising it. But in the case of the dog, it's just something naturally within is taking it to the next level. Whereas, you know, like nothing in its evolution ever helped dog recognize a leash, but it's actually now recognizing a leash. I mean, I guess in a way, yeah, we were kind of training them, but still, the recognition has to come from within them, right?

[00:16:40]  Unidentified speaker: That's right.

[00:16:41]  Bruce: They have a little bit of a… [overlapping speech unclear]

[00:16:43]  Sadia: But in that sense, aren't they actually creating meaning then, right? So a leash means something to them.

[00:16:49]  Bruce: If that's what you mean by meaning, then yes. Then yes, they're creating some kind of meaning. I wasn't thinking of meaning in that sense. But yes, I guess in a very loose sense, we could call that meaning.

[00:16:59]  Cameo: You know, not to keep us too long on this side note, but I was thinking about my dogs just the other day because sometimes I haven't even gone to get the leash or fill up the water bottle. But I've decided that I'm going to take the dogs for a walk. And they start getting excited. And I tried to think, like, how can I keep from showing any of the things that might indicate that I'm going to be taking them on a walk? Because once they start getting excited, I find it really irritating. Because they like bounce around and the one dog paws on me. So I've been trying to suppress the indicators that they might use to decide that I'm about to take them on a walk. But I must be doing things I'm not even conscious of or a change in my posture. I tried to change the order I would get my shoes on. And I'm giving them signals that I can't even control.

[00:17:52]  Sadia: Actually, I've noticed the exact same thing with my cat. It's kind of annoying. And sometimes I feel like he overreads me. And then, you know, he'll kind of, but it is weird that how at certain times he'll pick up if I'm kind of pretending that I'm going to my laundry room and looking for something else, whereas actually I'm looking for something he doesn't like, which is to put that flea medication on him. He right away, somehow he just picks up on my language there. How I'm kind of, you know, it gets suspicious. It is weird. Like it's annoying. It is weird that… [overlapping speech unclear]

[00:18:28]  Cameo: And that's a type of learning that artificial intelligence, as we currently define it, isn't capable of.

[00:18:40]  Bruce: So, just to make the distinction clear, if we wanted to train artificial intelligence specifically to know when you were going on a walk, we probably could. But if we hadn't specifically put in some sort of world space, you know, reward system based on going on a walk, then there's no way one of our AIs would figure this out, right? It's just way too narrow to do that. Okay, dogs are doing that. They're doing something that we can't even conceive in terms of AI algorithms right now. This is what I'm trying to emphasize. This is why I say trial error learning, which all animals have, is a kind of intelligence. We should just accept that. It just isn't the kind of intelligence that Byrne's looking for, right? To him, he's still counting that as more or less unintelligent behavior. He accepts that animals are able to do this. He's accepting that that's a kind of abstraction.

[00:19:32]  Bruce: He's not denying any of that. He's just really saying, this is not what I'm looking for. I'm looking for something more impressive than that. OK, so having now acknowledged that animals are intelligent, at least in the sense of trial and error learning, which is very impressive by AI standards, Byrne says, OK, let's talk about abstraction. By the way, he points out that all animals can abstract in another sense. Whenever I say all animals, assume I mean brain stem and up, not literally all animals. Like birds, if you teach them to get a reward for clicking a red button, they will also abstract that to an orange button, but not a white button. They seem to be able to abstract in terms, you know, a human would do that too. We kind of just know orange and red are similar. And so we can abstract that. Well, birds can do that. OK, all animals can do that, in fact.

[00:20:24]  Bruce: So that's another sense in which animals can abstract that's actually kind of impressive. Again, he doesn't count that as the type of thinking or the type of intelligence that he's looking for. Now, he says, now, I'd find these examples of abstractions more convincing if you couldn't explain them with classical conditioning. So he says, we've not we've not yet come up with any examples in the lab where we've shown animals able to do abstractions that we couldn't explain with classical conditioning. So he says, but in the wild, we see examples of it where it just seems really obvious you couldn't explain it with classical conditioning, and yet it happens. So he gives the example in the book, The Thinking Ape, gives the example of border patrol behavior of chimps. So chimps, they have this behavior where they go out on raiding parties and they attack other clans of chimps and kill the males and try to steal the females.

[00:21:20]  Bruce: They don't do it. So first of all, you can't explain that behavior as pre-programmed because they don't do it on a regular basis. Let's say they did it like every spring or something, then we might make a case for, oh, this is genetically pre-programmed behavior. Also, it's very complicated behavior. So it's difficult to explain it as genetically pre-programmed behavior just because of the complexity that's involved of getting a group together, going out and everything you have to do to make it successful, things like that. So you're generally eliminating pre-programmed behavior with a border patrol behavior. Now, could it be learned? Okay, border patrol behavior is learned. So, yes, and we would then consider that trial and error learning. So we wouldn't consider it kind of a deep insight. However, here's the thing that he points out. While the chimps are on a border patrol, if one of the chimps starts to become too noisy, the other chimps in the group will threaten it to get it to shut up.

[00:22:18]  Bruce: Now, he points out that this behavior can't have been learned by regular reinforcement learning because border patrols are really rare. They don't happen very often. And if they do it wrong, the feedback mechanism, you do it wrong and you get a negative reward, they'd be dead. Right. So here's these chimps doing this behavior while on a border patrol. And it's really hard to come up with a good explanation of how they understand the concept of, oh, that chimp is being too noisy. I better threaten him to shut him up. They're somehow making a connection between noise and potential danger that can't be explained through classical conditioning, can't be explained through genetic pre-programming. So he thinks this is probably an example of where the chimps are actually abstracting things in a way that isn't explainable just through classical conditioning.

[00:23:13]  Bruce: And we're going to use examples like this as we go along to kind of explain why he believes chimps do, in fact, are the kind of thinking animal he's looking for.

[00:23:22]  Sadia: OK, I guess noise in some cases might be beneficial as warning others, right?

[00:23:26]  Unidentified speaker: But in this particular case, they understand that. That's right.

[00:23:29]  Bruce: That's right. It's like the opposite of what it would normally be.

[00:23:32]  Sadia: It's just not like there's a specific program that always equates noise with that. It's more circumstantial.

[00:23:39]  Bruce: That's correct. That's correct. So the chimps are showing a kind of thinking abstraction that can't be explained through trial and error learning or classical condition or genetic pre-programming. So what Byrne is looking for is this deeper sort of understanding or thinking or intelligence, not any possible use of the term. So he, to differentiate, he coined the term insight. Now, I've been using that term in the last two and I need to explain it better. What I've been saying is, is the ability to use mental models. But of course, even this, I'm being a little abstract and I'm not explaining myself well. You really have to get into examples to truly understand what he's talking about, which is what we're going to do today and then in the future episodes. So Byrne describes this as insight remains, the term insight as used in everyday usage, is used as a down to earth lay term for deep [unclear] and discerning kind of understanding.

[00:24:35]  Bruce: So he feels like that term in lay usage comes really close to what he's looking for scientifically. So that's why he's co-opting the term insight to be what he's looking for. When did, so his book Evolving Insight, now you know what that means. He's saying when did insight evolve in animals? And which animals have it? Which ones don't have it? So and that's what he's trying to discover. This is like his whole career is to try to work this out. So Byrne does believe, this is a bit of a spoiler, he does believe that it's quite rare in animals. So he basically believes that only great apes have it. And then like in The Thinking Ape, which is one of his older books, he says, well, actually, probably cetaceans, so like dolphins and whales have it. He gives some examples of that. And then in his later book, Involving Insight [Evolving Insight], he goes, well, actually, maybe elephants and some birds have it.

[00:25:26]  Bruce: But it's not a lot of animals. OK, it's, by the way, some birds are the smartest animals on Earth. So that's more impressive. You know, it's not all birds, it would be specifically the ones that are particularly intelligent. So there are these groups of animals, a few animals that he thinks have insight. Now, he still weights that, though, like we haven't done anywhere near as many studies with dolphins as we have with chimps. So he's a lot more sure that chimps have insight than he is the dolphins. But he thinks dolphins do. And we haven't done a ton of stuff with elephants, say. I mean, that's a lot harder to maintain elephants. So we just don't have the data. And there could be other animals that he doesn't know about just because we've never studied them. Right. I mean, for all we know, maybe hippopotamuses have insight or something like that.

[00:26:17]  Bruce: OK. But that's not the way Byrne goes about this. He's basically saying, you know, I'm kind of assuming no animal has insight unless they actually demonstrate it in some way. OK. So he's saying we really, really only have strong evidence for the great apes, the ones that are the closest to the human family. And if you count dolphins, then we can see that there may have been some convergence, evolutionary convergence towards intelligence in that case. OK. Now, I need to do a little bit of an aside for this next because I'm going to now describe Byrne's methodology, why it's so awesomely Popperian. To understand this, though, I need to do a quick explanation of Popper's epistemology. Now, we've done that in the first four episodes and we've talked about that extensively. But there are numerous misunderstandings of Popper's epistemology. I've talked to a lot of Popperians out on the Internet, and I would say that in general, they don't usually understand it that well.

[00:27:14]  Bruce: There are certain common misunderstandings that come up. So I want to explain those quickly so that this will help understand why I'm so impressed with what Byrne is doing. So here's some common misunderstandings that kind of come up just from talking to people on the Internet, even amongst people who like Popper and have read Popper and consider themselves Popperians. So one of them is you label any theory that you like that you have a preference for as a best theory and then claim we need to embrace best theories. So basically, a best theory becomes means little more than this is the theory I prefer. That's not what a best theory is supposed to be. A best theory is supposed to be a theory that has strong empirical content that's been highly corroborated. We attempted to falsify and we couldn't. And it has no competitors. Things like paradigm theories, like quantum mechanics and general relativity.

[00:28:03]  Bruce: Theories like that are best theories. Your pet theory is not a best theory. That is a misunderstanding of the term. Another one is this idea of refutation. Everybody knows that Popper is about conjecture and refutation is kind of his own summary of his epistemology. One thing that commonly happens is that people will simply word their arguments in terms of a refutation. They'll use the language of a refutation. Typically, their arguments are actually just appeals to intuition. That's what most people do. I'd say probably 80%, 90% of arguments, someone you're having on the internet somewhere, you're probably not doing Popper's epistemology, you're probably just talking about intuitions. Oh, it feels like this is the truth to me or something along those lines. Or you're using an intuition pump, as we call them, something where you use an example and it pumps your intuition to seeing things a certain way.

[00:28:57]  Bruce: It's well known that intuitive arguments like this, they're super common. We buy them. They're often very convincing, but they're not what Popper was looking for. And they are not examples of Popperian refutations. So Popper, to him, a refutation was something far more concrete. It was that you have a theory that has actual empirical content, so it can be tested, and you use deductive logic to work out the consequences of your theory. And then you show that there is a contradiction, a logical contradiction. Preferably, the logical contradiction is empirical. It's an experiment that doesn't do what the theory said it was supposed to do, although he'll also accept a contradiction against another theory. So let's say you had a theory that showed that you could violate the law of conservation of energy, but it only took place inside of black holes, so there's no way to actually empirically test it.

[00:29:53]  Bruce: The fact that Popper would probably still accept, well, actually, the very fact that it violates the laws of the law of conservation of energy, is in and of itself a contradiction that refutes the theory. So in a case like that, he would still consider that a refuting contradiction. But there's concreteness to what a Popperian refutation is. It is not an appeal to one's intuition. The other thing that comes up is the idea of an ad hoc save. People tend to do ad hoc saves to their pet theories, and then they call that a counter-refutation. So you try to refute their theory, and then they save their theory through an ad hoc explanation, and then say, no, I just countered your refutation. That's not what a counter-refutation is under Popperian epistemology. In fact, Popper's epistemology rules out all ad hoc explanations, period. It never allows them.

[00:30:49]  Bruce: Now, what is an ad hoc explanation? It is something concrete. It is any explanation that does not increase the empirical content of the theory. Okay, I'm going to say it again. It's any explanation that does not increase the empirical content of the theory. So if you are going to use an explanation to save your theory, the explanation you're using has to have itself new testable consequences that the original theory didn't have on its own that can then be tested. If it doesn't have that, Popper rules out the ad hoc save altogether. Now, I want to do a little bit of a defense of ad hoc saves. I do think they have a purpose. I think that they work very well as starting points for research programs where you say, okay, I think that this problem could be resolved by discovering a new kind of particle. Deutsch uses this as an example in one of his papers.

[00:31:47]  Bruce: When you just say, I'm going to say there's a new kind of particle, it's an ad hoc save. It does not remove the problem. Once you've actually come up with, here's what the particle is and here's the testable consequences of the particle, then it's no longer an ad hoc save. So an ad hoc theory or explanation can grow into a good Popperian theory and no longer be considered ad hoc. But so long as it's ad hoc, as long as it has no testable consequences, it can never be considered as having solved the problem. Okay, this is something that a lot of Popperians that I've talked to do not understand. They think that the moment they've given you some sort of response, no matter how ad hoc it was, that they have now saved their theory. This actually came up with an interesting example. We were talking about, I was, some of us online were talking about face blindness.

[00:32:36]  Bruce: Now, I don't know if you guys know about face blindness, but there's a module somewhere in the brain that recognized—set aside specifically to recognize faces and it can get damaged. And if it gets damaged, you won't be able to recognize people's faces anymore. You'll have to, I mean, like, there was a guy who had—some people were born with this, but it can also happen from specific brain damage. There was a guy who had face blindness. He had to tell his wife to wear a bow in her hair so that he could find her at the party. Now, it doesn't stop them from recognizing male and female. It doesn't stop them from recognizing a beautiful face from an unbeautiful face. They just can't seem to recognize if this is, you know, their wife's face or not. OK, and it seems so strange to us that there could be a module for this, right? So in the discussion, I pointed out that this is an example of where the brain seems to have very specific modules that seem to have nothing to do with the universal explainer modules where we use explanations to explain things.

[00:33:39]  Bruce: And because if it was, then if it just got damaged, it would just be a damaged explanation. You could just rebuild it as an explanation. And yet this never happens when a person gets their face module destroyed. They never rebuild it ever. So I pointed this out and I said, this is a refutation of your view. It's clear that there are some non-explanatory modules in the brain, which that's not really that big of a shock. This is something that Popperian should have known from the outset was probably the case. OK, how it actually works, I don't know. This is a mystery. I was suggesting maybe it's like machine learning. That's just a wild guess on my part. OK, that might work. And this person I was arguing with, he said, Oh, no, I'll bet you that just for some reason, children can learn explanations about faces that adults can't. I'm like, OK, wait, where is that a part of your universal explanation, universal explainer theory?

[00:34:30]  Bruce: It's not, right? I mean, you're you're completely doing an ad hoc save here and it has no testable consequences. He says, “Well, OK.” Well, maybe maybe what happens is, is that a lot of times it does get rebuilt, but it just happens so fast that they're not noticing it. OK, well, now you're making an ad hoc save where you're saying it either gets rebuilt really fast or never. How's that a part of your theory? How does that have testable consequences? So this would be an example of where I'm sure he very sincerely thought he was countering my my Popperian refutation. But in fact, he was just doing an ad hoc save. OK, now the last one is that often I've seen people label non-testable theories as best theories. This one surprises me a little because I would have thought most of the others I could understand the misunderstanding. This one surprises me.

[00:35:17]  Bruce: We've kind of talked about the buggy animal behavior. So Dennis Hackethal, who's been on this podcast in the past, he keeps kind of a list of buggy animal behavior. And we're going to actually talk about that today. And then he actually put out a post about why he feels animals aren't sentient. They feel nothing. He thinks they're just meat robots based a lot on his list of buggy animal behavior. Now, I'm not getting into animal sentience. Animal intelligence may exist without animal sentience. And so, you know, who knows? Maybe he's right. OK, here's the thing that's surprising. It's not whether he's right or he's wrong that matters. His theory is a 100% non-testable theory. OK, it is a zombie theory. He's—just to explain zombie theories… I know I feel things. I know I feel pain. I know that I feel excitement or pleasure or curiosity. And Cameo acts like she does, but maybe she doesn't.

[00:36:11]  Bruce: Maybe she's behaving like she's seeing those things, but she's not really feeling any of them. And she is therefore what we call a philosophical zombie. OK, and have you guys heard of philosophical zombies before? Are you familiar with that or is this something new to you guys?

[00:36:28]  Cameo: Yeah, I'm familiar with it.

[00:36:30]  Bruce: OK, so zombie theories are made up by philosophers. Nobody worries about them except philosophers. And they are specifically used by philosophers to show skepticism. OK, they're trying to say, see, you can't ever disprove this theory. They're meant to be untestable theories. OK, now, Deutschians would agree that in the case of humans, they're silly, that they get eliminated up front from the discussion because they're untestable. And yet they will use them for animals for some reason. OK, without realizing that actually Popper's epistemology eliminates all non-testable theories like that. Up front, they do not get to be part of the critical discussion. Now, Dennis, for some reason, doesn't seem to realize that. And he thinks that he has a best theory about animal sentience. When really, there's two, you know, let's for the sake of argument say two competing theories.

[00:37:22]  Bruce: I'll actually show later that the idea that animals have sentience is actually the better theory. But for the sake of argument, that's not a great theory. I admit we don't really know for sure. And I don't want to make statements that are too strong. But the idea that we know that that is a best theory that animals don't feel things, that is a non-testable theory. It is not even allowed as a Popperian theory into the discussion at this point. Now, what would Dennis need to do to make it such? He would need to come up with consequences of that theory. He would need to say, OK, if an animal is a zombie and isn't actually feeling pain or feeling things, what predictions would that theory make that are different than if the animal was just feeling things? OK, well, of course, that's not the point. He's trying to come up with a theory that eliminates any way to test it and can't be refuted.

[00:38:15]  Bruce: But you never want non-refutable theories under Popper's epistemology. OK, does this make sense? Do you see where I'm coming with this, that these are common misunderstandings that come out of Popper that we want to correct for this next discussion here?

[00:38:28]  Unidentified speaker: Absolutely. Yeah.

[00:38:31]  Bruce: All right. So now why did I say all this? OK, let's go back to the buggy animal behavior. So Deutsch uses the example of a squirrel that tries to bury nuts on concrete to show what? Well, he's trying to show that animals don't understand things. And he's trying to show that all the animals, maybe he's trying to show, some people will say, that all the animals' knowledge is in its genes. OK, and Dennis has collected this very large list of buggy animal behavior that's similar to that, where you can see the animal doesn't understand. Now, what do these examples really demonstrate from a Popperian standpoint? What they're trying to do is they're trying to take a specific example. Squirrel tries to bury its nuts and doesn't understand it's on concrete. They're trying to generalize. OK, they're trying to say, OK, that means the squirrel is entirely automatic behavior, just like this automatic behavior.

[00:39:20]  Bruce: OK. Well, can you do that? Can you take a specific example and generalize from it? What is that called when you try to take a specific example and generalize from it?

[00:39:28]  Sadia: I mean, isn't that kind of like in what induction is supposed to be?

[00:39:32]  Bruce: That is induction. Yes, that is the word induction means to generalize from specifics. When you're trying to do an argument like that, you're doing induction. OK. Now, I think there is something to this a little bit. It does sort of demonstrate something to our intuitions. We say, oh, wow, a squirrel behaving in that way, a human wouldn't make that mistake, at least not that many times in a row or something along those lines. There may be something valid there that's getting lost, right? But they're attempting to generalize from a specific and that is induction. It doesn't matter how many buggy animal behaviors you collect, right? For the same reason, it doesn't matter how many white swans you collect. If you have five billion buggy animal behaviors on your list, then that's the same as having five billion white swans. You still can't generalize to all swans are white, OK?

[00:40:26]  Bruce: It simply does not mean that. This is why the buggy animal behaviors list don't really prove anything. You might argue here, OK, now wait a minute. What if someone were claiming that animals have human level intelligence? Well, no, no one's claiming that. At least no scientist is claiming that. Everyone knows animals don't have explanatory knowledge like humans. OK, but let's say somebody was. Now, this is why I say it seems like there's something a little bit valid about it. The fact that I see a squirrel doing automatic behavior for trying to bury a nut, I pretty much I feel like I can intuitively say, oh, that shows that the animal, the squirrel doesn't have explanatory knowledge. And I think this is really what they're trying to do. OK, they're they're trying to find this intuitive example like this. Now, it's tempting to say that this is OK, but really it's not, not under Popper's epistemology.

[00:41:19]  Bruce: Let's let's be honest, Popper's epistemology is a harsh mistress. OK, it does not allow even examples like this into the critical discussion. Now, why would that be? Consider the idea that Popper always requires that things are done in terms of constraints or universal laws that can therefore be contradicted. So let's take the universal law. All animals have human-like intelligence. OK, that would be what they're trying to refute. All right. And so the refutation is to show a lack of understanding. Again, I can see why they're doing this. This makes sense to me too. I find it somewhat convincing. All right. But it is not valid under Popper's epistemology. And the reason why is because an intelligent person may show a lack of understanding. Whereas—and that would be a verificationism. OK. But an unintelligent, here's the thing you really have to realize.

[00:42:13]  Bruce: An unintelligent animal can't show intelligence. But an intelligent animal might show a lack of intelligence in some cases. That's why you want to get away from trying to use inductive arguments. And there is a way to still go about it. It still accomplishes the purpose that they want to. And Byrne does it well. And that's what I'm trying to build up towards. How would you really go about doing this not with buggy animal verificationist type arguments, but with Popperian style arguments? This is what Byrne does really well. So first of all, he starts with best theories. So he's not trying to do vague things like, when do animals think and are they intelligent? And, oh, I call that behavior intelligence. He's not doing any of that. He starts with the idea of trial and error learning. He says, look, I know animals can do trial and error learning, and I am upfront counting that as unintelligent for our purposes.

[00:43:04]  Bruce: No matter how impressive it is, I'm counting it as unintelligent because we know they can do that. So any behavior that I see an animal do that can either be explained as pre-programmed genetic behavior or can be explained via trial and error learning, I'm going to assume that that is the explanation, even if it's not. I'm going to assume that it is. So if I see an animal doing something that looks intelligent, and then I say, OK, well, we could explain that through trial and error learning. I don't know if we don't know its actual history, so we don't know if it actually was trial and error learning, but reasonably it could have come from trial and error learning, but I'm going to assume it was trial and error learning. And he's very harsh this way with his own experiments. He enforces this idea, if it can be explained by trial and error learning, then it will be, I am going to count it as trial and error learning.

[00:43:58]  Bruce: OK, or if I can explain it through genetic preprogramming, then I'm going to count it as genetic preprogramming. And this is the reason why this is so Popperian is because he's starting with, he's effectively starting with universal law. Animals have no insight. OK, and then he's looking for experiments that refute that universal law. He's trying to find examples of where animals behavior can't be explained by genetic preprogramming or by trial and error learning. I already gave one example, the example of the border patrols and the chimps quieting each other. OK, this seems like an example where you can't explain it by one of those two means. Therefore, we're left with the idea that animals have insight. OK, even then, he's super guarded. That's why he says, well, that's only in nature. We really want to come up with an experiment that shows it so that it's replicable.

[00:44:51]  Bruce: OK, he can't replicate something that's in nature. He never discounts observations in nature, but he counts them as not as strong evidence. OK, now this is why I say this is perfect. You're basically saying, look, I'm saying all animals are unintelligent and then I'm looking for counter examples. That is the correct Popperian way to go about it. Because an animal that is unintelligent can't ever behave as if it's intelligent. OK, but an animal that is intelligent can behave as if it's unintelligent. So he's figured out the correct Popperian universal law that is actually refutable. That's what he's done. And he's got to he's going about this the right way. Now, there is one more issue, which is you can actually explain all animal behavior without referencing insight. If you call it a coincidence, you could say, oh, the animal just happened to try something that he's never tried it before.

[00:45:44]  Bruce: That was the first time it was trying it and it happened to work. OK, if you're going to explain it like that, that's an ad hoc save. So he discounts up front. He says, I'm—any—if you're just going to use coincidence to try to explain it, you know, that does not count. You're not allowed to save the theory through referencing to coincidence. Now, apparently, a lot of his peers actually do use coincidences like that to try to show the animals aren't intelligent. So he's discounting that again. That's a proper Popperian epistemology. You do not count ad hoc saves. They do not enter into the equation. So Byrne's methodology is the correct Popperian epistemology. You start with the idea animals have no insight, they're unintelligent, and then you try to you try to refute that. You try to come up with examples where the animal behavior can't be explained using that universal law.

[00:46:36]  Bruce: OK, this is what I love about Byrne. OK, is that he really is going about this in the right way and coming up with some fairly impressive results, I think. So one other thing, he avoids terms like understanding and uses insight instead, and he gives an example of why he does that. So he gives the example of a cat. So there was a cat that had learned that if its owner was sitting in its favorite chair, that it could pretend like it needed to go outside to relieve itself. And the owner would get up out of the chair to open the door and it could then take the chair. So we would say, oh, that cat was being deceptive. Now, Byrne's not necessarily denying that. The word deception might include this cat's behavior, but he would point out that the word deception, usually in the human mind, includes the concept of some sort of theory of mind, where if I'm being deceptive to Sadia, then I'm thinking about what she's thinking about.

[00:47:36]  Bruce: I'm thinking of her as a mind separate from mine. I'm thinking about, okay, I need to act this way and that will cause Sadia to think this way. The cat's probably not doing that. From Byrne's point of view, the cat's not doing it. It has no theory of mind for its owner. What it actually did is it learned it through trial and error learning. So you can imagine a cat that one day it wanted to sit in its chair, but it couldn't because the owner was there. And then it thought, oh, I should relieve myself, but that wasn't what it really wanted to do. So it goes to the door and meows. The owner gets up. It sees the chair is available. That's actually its true intention is to take the chair. That's the top intention. So it runs over and it takes the chair. At this point, it has learned, if owner is in chair and I meow at door, then owner will let me have chair.

[00:48:23]  Bruce: It doesn't, it's not thinking of itself as intentionally deceptive. And yet the act is a learned deception. Do you understand the difference here of what I'm trying to get at?

[00:48:33]  Sadia: Yeah.

[00:48:34]  Bruce: So he talks about this is zero order intentionality. So like a hawk moth with a pre-programmed flick of its wings to show eyes when a predator is near. So a hawk moth looks like it has eyes. It looks like it might be an owl or something like that. So if a predator is near, it flicks its wings, the predator sees the eyes and it deceives the predator. That'd be zero order intentionality. Really, the hawk moth is just a program doing something. OK, then there's got first order intentionality. That would be the cat. The cat doesn't know that its owner has a… it's deceiving its owner, doesn't have any concept of deception. But it does have an intention to accomplish a goal, which is taking the chair. So that'd be first order intentionality. Then there's second order intentionality, which would be mentally representing the mental states of others.

[00:49:22]  Bruce: I want him to think X. Second order intentionality requires actual insight, the other two don't. So from Byrne's point of view, the first two are unintelligent or don't show insight, and the last one shows insight. Okay, now I talked about how the cat could learn to deceive. So let's give some examples from experiments. So let's say a monkey is given a tube with food in it, and it has to learn to use a tool to get the food out of the tube. Okay, now monkeys can use tools, chimps can use tools, apes can use tools, a lot of birds can use tools, tool use is actually far more common amongst animals than you might think. So the monkey grabs this tool and it just shoves the tool in and it tries all sorts of stuff and it gets the food out. So that would be, we would assume that that is trial and error learning because that conceivably could be done through trial and error learning.

[00:50:17]  Bruce: Therefore, we assume it is. Now, they want to test this further though. So they try putting a trap on the tube. So if the monkey pushes the food one way, the food falls into the trap and then it can't get it at all. Where if it pushes it the other way, it can get the food. So the monkey then tries all sorts of different things very quickly. The monkeys are fast and it figures out, oh, I have to push from the left side to get the food out. So that it doesn't fall into the trap. Now, we want to test to see if it understands the concept of the trap. And it's because the food is falling into it. So they switch the tube to be upside down so that the trap is upside down. So now it can't trap the food anymore. And then they wait to see what the monkey does. Well, the monkey continues to push from the left side. So it has learned the rule, I'm going to push from the left side, and that's how I get the food out.

[00:51:14]  Bruce: And that's the way that works. It doesn't have any real concept of because there's a trap there or the food might fall in. Okay, so by experiment, we've now shown that this monkey is actually just doing trial and error learning that there's no insight. Now, great apes can be given the exact same experiment and they understand what the trap is. And once you twist the trap around, it will now know it can come from either side and it can get the food out. So Byrne would say that this experiment is evidence that apes do have insight, whereas the monkeys don't. So he would make a cutoff now. He would say great apes have insight and monkeys don't have insight. So that would be an example of how we do with an experiment. He gives other examples, a lot of them come from nature. And in fact, I'm going to give you several examples here just to kind of build an intuition for what it is Byrne's looking for.

[00:52:06]  Bruce: So he gives an example of pedagogy. So there are animals that will teach their young. This is actually, I don't think it's super common, but there are many animals that do it. So cheetah mothers, they will bring prey back to their cubs and they'll disable the prey so that it can't escape easily. And then they'll let the, initially, it might be dead and they let the cub play with the dead prey when it's particularly young. Later, it'll be still alive, but it'll be injured so that it can't escape easily. So the cub has to use its meager skills to try to kill the prey. And then it will slowly make the prey less disabled over time so that the cub gains skills in a linear fashion. Pretty impressive, really, right? So what they wanted to know is, is the animal, is this just pre-programmed behavior from genetics? Or is this actual insight? Is the cheetah mother actually understand the concept of, oh, I need to disable this prey so that I can get my baby to learn slowly?

[00:53:14]  Bruce: Well, in this case, they said, OK, reasonably, if the mother is doing it based on age, then it could be biological programming. However, if it's doing it based on competence, biological programming is not a good explanation anymore. So all we really have to do is we have to assess if the cheetah mother is doing it based on age or doing it based on competence. OK, well, the answer is, it's by age. So Byrne, based on his harsh Popperian rules, says that means we're going to say it's biological programming. Whether it is or not, we're now going to assume it is. OK, we rule it out because it can be ruled out. Now, he gives another example with apes. He says there was an ape mother, this has actually happened twice, that saw that its child was not learning how to crack a nut with a stone. So apes will take a nut and they'll put like one stone down as a hammer—or as an anvil and one as a hammer and then they'll crack the nut.

[00:54:11]  Bruce: OK, this is actually, Deutsch talks about this in Beginning of Infinity as one of his examples. The child maybe at first isn't very good at it. It hits it and the nut goes away. It might be hard to find. There's some risks involved there. So the mother took the stone, showed it to the child and then rotated it very, very slowly over several minutes so that it was in the right position that it could crack the nut and then showed the child how to crack the nut. And from that, the child learned to crack the nut by positioning that stone in the right way. Now, obviously, this would be stone specific. So this is different than the cheetah mother example. So that's one point in favor of insight right off the bat. Furthermore, he's only ever seen chimps do—mothers do this like twice. So it can't be pre-programmed behavior because then all chimp mothers would do it.

[00:55:06]  Bruce: So it really looks like in this case, the mother actually had a concept of pedagogy in mind where it was trying to teach. Oh, I need to teach my child how to crack this nut so it had to rotate it the right way, very specific to a specific stone. And then the child got the point and learned to do it. Now, here's the weird thing about that. That would require insight. So that would be another example of apes showing insight that can't be explained by one of the other two methods. But he's never seen a mother then suddenly have it click. Oh, you know what? This concept of pedagogy and teaching, it's generally valuable. I ought to use it in lots of different cases. Apes don't do that. So on the one hand, they actually have enough insight to understand pedagogy, but they don't understand the concept of pedagogy and how to then try to reuse the abstraction of pedagogy in other circumstances.

[00:56:00]  Bruce: Does that make sense? Now, based on experiments like this, you wouldn't ever make a conclusion based on a single experiment. That's the other part of Byrne's Methodology that's important. So if you, if any individual case, there's always room for doubt because it might be a coincidence. Okay. But imagine you, you repeat this approach across time. So you go and you look for deceptive circumstances, you know, where cats and dogs use deception. You look for where apes use deception. And let's say that, by the way, let's say that, let's say that the cat didn't use deception only in one case when you're in your, in the chair and it meows at the door. But let's say that it's constantly coming up with novel sorts of deceptions. OK, at that point, we would have to say, oh, that has to be insight because it's not trial and error learning and it can't be genetically pre-programmed behavior.

[00:56:53]  Bruce: OK, cats, anytime they show deception, cats and dogs, anytime they show deception, it is always explainable through trial and error learning. They went through and they tried to find counter examples. They could not find any counter examples. That is not true for apes. Apes, they'll take a bunch of examples, they'll explain away as many as they can as trial and error learning, and they'll always have some left. Where they'll go, oh, yeah, we can't explain those with trial and error learning. OK, so he would then tentatively conclude that apes have insight and cats and dogs do not. And this is how Byrne goes about experimenting for when animals have, which animals have insight, which ones don't. OK, does this make sense? Do you have any questions or criticisms of Byrne's approach?

[00:57:37]  Unidentified speaker: I don't.

[00:57:37]  Cameo: I agree. He seems very Popperian in his reasoning.

[00:57:42]  Sadia: Yeah, same here.

[00:57:43]  Bruce: OK, I've got several of these, but let's just end this with a bunch of examples. I'm just going to give you a whole bunch of examples that kind of show different animals showing insight, most of which are the great apes. But I'm going to give you a couple other non-examples that are outside of the great apes just to kind of mix things up a little. OK, so chimps will grab a tool and modify it. Now, they don't do anything super complicated, what I mean by that. They will remove branches from a stick so they can use it as a poker. They'll use a stick as a poker to stick it into a termite mound so they can pull the termites out and eat the termites. Now, a lot of animals use tools. Not very many modify the tools. So right off the bat, this is suggestive of chimps having insight. What's more impressive, though, is that the chimp will grab a tool in one location, modify it for the use of what it needs it to as a poker, then it'll walk to some location that isn't in its view and then use it on the termite mound.

[00:58:43]  Bruce: This, how would an ape do that? Right? I mean, first of all, fishing for termites can't be genetically pre-programmed behavior. We know basically that apes learn that through trial and error learning, or through learning mechanisms, because it's specific to whatever their environment is, okay? It literally must have some sort of mental representation of A, what the tool must look like to be useful, and then it has to modify it to match. And then it also must have some sort of mental representation of I'm going to now use this tool in a different location I'm not currently in, and I'm going to accomplish the following. Okay, how else would you explain this example, except by attributing to chimps' insight? Another example, an ape was shown a picture of humans, there was apes that were shown pictures of humans with problems, and they were rewarded if they could select the item that the human needed.

[00:59:34]  Bruce: So an example would be they'd show a picture of a man shivering from cold, and he has a heater, but he has no matches. So then the ape would have to select out from a list of pictures that the matches was what the human needs, and then he'd get a reward for selecting that. The ape can do this. The apes can actually learn things like this. They can put together, oh, that man needs the matches and pick it correctly. It can't be genetic or trial or error learning because of the novelty of the circumstances. It seems like it's showing a theory of mind, oh, that guy is shivering. However, you might explain it still as, well, there's association between matches, fire, heat, being cold, something along those lines. So it does show a kind of insight, but it doesn't necessarily show a theory of mind, but it might. We'll get to that. Does insight include a theory of mind?

[01:00:22]  Bruce: A theory of mind meaning that you have a theory that other beings have minds separate from your own. Now monkeys, this is one from monkeys. Now remember, Byrne does not attribute insight to monkeys, so it's kind of interesting that he does have some examples even for monkeys. He gives an example of where the monkeys were, they would like, a human researcher was in the wild with these monkeys, and the monkeys would disappear, and then suddenly reappear later, and the researcher wouldn't know where they had come from. Well, what the monkeys, they realized later what the monkey was doing is it was hiding behind a tree, and then it was keeping itself in the cone of invisibility the tree afforded it, until the human was in a different position, and then it would leap out and surprise the human. Now, is this show theory of mind? Maybe. But it may show something a little bit more basic.

[01:01:13]  Bruce: It may just show that the monkey has enough insight to understand the concept, if I can't see the human, the human can't see me. Most animals can't do this, by the way. Most animals have no ability to know something like that. So monkeys, which Byrne would not attribute insight to, do, at least in this case, show something like a very limited form of insight. Now this is one of the things that comes up from reading Byrne's books that's a little bit frustrating, is that he will give examples of, he'll kind of say, the only animals that I think have insight are, you know, great apes, maybe dolphins, and you know, maybe elephants and some birds. And then he'll give examples in other animals that don't fall into that. And he doesn't necessarily have a better explanation. Animals are weird like this, right? It may be that they do, that some of them do have a kind of limited narrow insight that is genetically channeled or something along those lines.

[01:02:07]  Bruce: So when he's looking for insight, he'll admit animals do sometimes show it that don't fall into his group of animals that have insight, but he's looking for ones that have a more general ability to do it. Okay, so he still discounts the monkeys, even though in some circumstances they show some insight. Chimps, the use of rocks as hammer and anvil, they'll actually pick out an appropriate rock to be at the anvil and an appropriate rock to be the hammer, which is very suggestive that they have some sort of insight into I need my rock, my tool to look like this to be useful. Then another one that's really interesting is one chimp was trying to do that, trying to do the blow on the anvil stone, and it wasn't working because it was wobbling. So he went down and he picked a sliver wedge to put underneath the anvil to level it and then was able to break the nuts with the stone.

[01:02:59]  Bruce: Again, this is very suggestive that this chimp has some sort of insight. It understood enough of the physical concepts to go, oh, I need to put this wedge. And again, how would you explain that as genetically preprogrammed? This isn't normal chimp behavior. This is very specific to a specific circumstance and problem. Another example of where Byrne would say, yeah, that is evidence, a corroboration of the idea that the chimps have insight. Another one that's interesting, that suggests at least a limited theory of mind, is there was a chimp, again, this is a human researcher in the wild, and this baby chimp wanted to groom the human. And the humans weren't supposed to touch the chimps. So first of all, it might be dangerous for the human, but it's actually even more dangerous for the chimps. They might get a disease from the humans. So the humans were supposed to stay away from the chimps.

[01:03:47]  Bruce: And here's this chimp, baby chimp, infant chimp, I should say, coming up and wanting to groom him. So to get out of that, he used deception. This is a human using deception. So, you know, of course, humans have theories of mind. So he pretended to see something in the distance. And so the infant got up and went to go look and stopped grooming the human and moved away from the human, which is what the human wanted. OK, so he used deception to get what he wanted out of this infant chimp. Once the chimp could see that there was nothing there, he came back over to the researcher, hit him over the head with his hand, then ignored him for the rest of the day because he was angry at him. So it's hard to understand this behavior if you're not going to attribute apes with enough insight to understand both the fact that it had been deceived and had some sort of theory of mind about what the human was thinking compared to it.

[01:04:42]  Bruce: Now, later we're going to see apes don't have a great theory of mind. It seems to be fairly limited. But this is an example of where it does seem to actually have one. Another one that's interesting was that there was a chimp that wanted to... So chimps like to look at babies because they think they're cute, just like we do. Chimps, if there's a baby chimp, the chimps would want to come over and see it. And the mother doesn't want the other chimps to be near the baby because it's dangerous. And the mother in this case was like an alpha mother. And so the chimp knew he'd be in trouble if he tried to go look at the baby. So he pretends he built a nest. He pretended to build a nest, which is normal chimp behavior. And then when he was done, he decided he didn't want that or acted like he didn't want it. And he built one that was closer to the baby.

[01:05:24]  Bruce: Then he [unclear] like he didn't like that one. He built one that was even closer to the baby. The final nest he built was so close to the baby, he could sit in the nest and stare at the baby, which is what he wanted to do. And because it had happened very slowly and the mother had thought he was just trying to build a nest and wasn't actually trying to interfere with her baby, she didn't stop him like she normally would do. So this is another example where how do you explain this behavior if you're not going to attribute to the chimp some sort of insight into, hey, I can deceive this mother by building nests and getting slowly closer, and at least having some sort of concept of limited theory of mind, that this will deceive the other chimp.

[01:06:08]  Cameo: And can that be, can it be argued that coincidence is causing that?

[01:06:14]  Bruce: You know, you can argue anything as coincidence. And that's why Byrne eliminates coincidences upfront, but that's also why he wants to see multiple examples before he really concludes that this species of animal has insights. Because any one example can be explained away via coincidence.

[01:06:35]  Cameo: Yeah, I agree. Interesting.

[01:06:37]  Bruce: Okay, here's another one about deception. He points out that if we were to ever see an arms race of deception, which humans do, that would suggest a theory of mind with deception. Okay, the second level of intentionality with deception, which requires insight. So we have a female chimp named Belle and alpha male named Rock. And they're in captivity. And Belle is shown where food is inside of a maze. And then put back with the other apes. And then the apes go look for the food together. And Belle knows where it is already. Now, Belle knows from experience, from trial and error learning, we're going to assume, that Rock will steal the food if he gets the food before her or if he gets to it right after her. OK, so Belle learned that Rock would go to where she was sitting and then search for the food. So she started to sit further away from the food so he would have to search further.

[01:07:37]  Bruce: And over time, he got to where he was used to searching further and further away. Well, Belle then used this and she waited for Rock to be really far away in the wrong direction, then she would run to the food. So up to this point, to me, this is very suggestive of theory of mind deception, but Byrne would say, look, you could explain this through trial and error learning up to this point. So we're going to explain it as trial and error learning. OK, now here's what happened. Rock then started to pretend to look away and then would look back suddenly and look to see where Belle was running to and then try to run there first. OK, well, suddenly turning around is not native chimp behavior, so it can't be genetically pre-programmed. And for him to suddenly start doing it like this and not have learned it by a trial, where they're watching him do it, he didn't learn it by trial and error, this is now suggestive of insight, that Rock actually had some sort of concept of deception and how to overcome it and had come up with a strategy to overcome it.

[01:08:39]  Bruce: So she started to intentionally lead him to the wrong spot so that he would search in the wrong spot and then try to run over to the right spot. She also learned to take Rock to where one piece, so maybe there would be multiple pieces of food, there's one place where there's one piece of food, another place where there's two pieces of food. She would take Rock to where the one piece of food was, let him steal it, and then she'd run off to where the two pieces of food were. So now we're talking about an arms race of deception. We're talking about something that really would require insight to explain because you can't explain it via trial and error learning anymore.

[01:09:16]  Sadia: I guess one of the things, just to interject here, that I'm seeing through all these examples is that in all these studies or these observations, there seems to be an element of surprise. Because the thing is that if there was something that was genetically programmed, then obviously you would see more of that thing, you know, a repetitive thing. And then, you know, how many of these elements of surprise do we need to say that there is some novel behavior here that can't really just be explained, you know?

[01:09:51]  Bruce: And you know what? The answer is you're guessing. Okay, that's the correct Popperian answer.

[01:09:55]  Sadia: Yeah. Yeah.

[01:09:57]  Bruce: So the fact that you can always explain it as trial and error learning with cats and dogs, but you can't with chimps, that's really what he's going on. Right. And the fact that it gets reconfirmed over multiple experiments or multiple observations. Okay. Another one that's interesting. Viki is a chimp that would pretend to pull a toy on a string. So it doesn't really have a toy on a string, but was pretending. And then she would pretend that the string got entangled on a real object and even pretended to be upset that she couldn't get it untangled. So this is, chimps do stuff like this. Does that suggest, you know, what does that suggest? I don't know. You know, I mean, like, it's hard to imagine this being genetically pre-programmed. Like, it's impossible to imagine that behavior being genetically pre-programmed. And it's hard to see how it could be trial and error learning either, right?

[01:10:44]  Bruce: Because there's no value in it. She's just having fun, basically, right? So in a case like this, it's another example where we would probably attribute it to insight, some sort of insight into play. Another one that's interesting that's kind of related to that is there was a mother. So we talked about chimps being taught sign language. And I told you how I wasn't really that impressed with Koko's language abilities. But there was a chimp that had been taught sign language, and they wanted to see if she would teach her baby chimp sign language. So they intentionally, whereas she'd been taught by humans, they intentionally wouldn't do signs in front of the baby. And then they wanted to see which signs it would pick up from the mother teaching it to it. And the mother would actually mold the hands of the baby, just like the humans used to do for her, to try to teach the baby signs.

[01:11:34]  Bruce: And the baby picked up like 17 different signs, with no human having ever taught it to it and got the meanings correct, just from the mother teaching sign language to the baby. Now, this is actually nowhere near as fast as if, this is very impressive, in terms of an experiment, this really does show some sort of insight on the part of chimps, to be able to figure this out, and to understand the concept of language and things like that. Even if it's really basic, just basic reference, that this symbol represents something else, is very impressive. However, this is the, that speed, 17 signs or whatever it was in the first two years, whatever it was, is much slower than if a human is teaching it sign. So they started to realize, oh, we better teach this baby sign. So they, they started teaching the baby sign language. And from that point forward, they, she had human teachers and she picked up sign much quicker when the humans were teaching it.

[01:12:27]  Bruce: Another one that's interesting is they had an experiment with the box. And so the idea is, is that the box is locked. But at some point, random point, the box will unlock and you'll hear a click so that you know it's unlocked. And at that point, the apes have been trained that they can open the box and there's some sort of prize that they want inside. It may be food or something. One example of this experiment, that a beta chimp heard the click so he knew the box was open, but the alpha chimp was nearby and would take the prize from him. So he pretended to be disinterested and the alpha chimp then walked away. And then the alpha chimp hid and spied on the beta chimp to see what he was going to do. And the moment he saw that he was going to open the box, he runs out and opens the box and steals the prize. This is page 134 of Thinking Ape, by the way.

[01:13:18]  Bruce: So again, adult chimps do not have a hide-and-spy behavior, so it can't be genetically pre-programmed. And this is completely novel. This is completely novel behavior that solves a problem in a specific circumstance and required no trial and error. Okay, it wasn't learned through trial and error. So again, it seems like it's suggestive of insight. Another one that's fun, this one comes from a dolphin. The only dolphin example I have is that, I do have other dolphin examples. The only one I'm going to share is there was, how dolphins will be in tanks and they can kind of swim up and interact with you in the aquarium or whatever. So this dolphin comes swimming over and there's a man smoking and he's doing puffs of smoke. And so the baby dolphin swims off, gets some milk from its mother, comes back over to the man and puffs out the milk as if it's a puff of smoke so it could join in.

[01:14:10]  Cameo: Oh, that's crazy.

[01:14:15]  Bruce: So, you know, first of all, there's no reason for this behavior in terms of rewards, right? So it can't be, it certainly can't be genetically pre-programmed and it can't even be trial and error because trial and error is based on rewards through valences that you get food and there's some sort of reward that you genetically need, right? It's been programmed to want. This is a dolphin having fun, right? It's joining in. It wants to do what the man is doing. So it came up, abstracted the concept of a puff of smoke into a puff of milk.

[01:14:46]  Sadia: It's purely based on the fun criteria there, right?

[01:14:49]  Bruce: Yes, based on the fun criteria. Okay, so this is a pretty impressive example of how, by the way, the other example that I wasn't going to share, but dolphins can be taught language too. The very fact, and when I say language, none of their language is super impressive. They usually have a grammar of like three symbols, both for chimps and for dolphins. They're super limited grammar. But the very fact that they can learn symbols, that they can learn this symbol means this, and that they clearly understand that, and that they can then put these into… and communicate with them. You can give commands to the dolphins that have learned language through symbols. You can tell them to do novel behaviors that they've never done before, where they understand the parts, but they don't understand all the parts together. You can link them together, and it will do this novel behavior because it understands the concepts in the language.

[01:15:43]  Bruce: So this is very insightful behavior. And now we've talked about American Sign Language for apes, but really the best language teaching to apes hasn't been American Sign Language. It's actually quite hard for apes. They can't form their hands well. So there's these Yerkish symbols that they use, and the best experiments in terms of teaching apes language has been with these Yerkish symbols, which they, similar to the dolphins, they'll have like three of them in a row, and there's kind of this really super basic grammar. They can't seem to get beyond like three symbols in a row, and they have to be in certain orders for the grammar. There's certain types of grammars that they can pick up and certain types they can't. They'll play a game where a human will hold food items on its lap, and then the ape needs to point to the correct symbol of the item it wants, and then both of them will eat the item.

[01:16:33]  Bruce: Okay, so maybe it points to an apple, and then the human and the ape get to eat the apple. And so they will teach this to these apes. Then they put the apes alone to see if they would play the game on their own without a human. And the apes would do it for fun. They would have these items of food, which no human is in the room. Okay, so they can do whatever they want, right? And the apes will actually choose to play the game. And one of the apes will point to the symbol that it needs. And maybe it will miss it. Maybe it will, like, it points to the right one, but the other ape isn't paying attention. And, like, neither of them will eat, right? They will refuse to break the rules of the game because they are having fun playing the game, right? Even though this is not being enforced by the humans. And they will actually continue to play the game.

[01:17:22]  Bruce: And there are other games they give examples of. They have taught these apes that they will then play on their own, okay? Which, it's hard to explain this in terms of, I mean, if it's just a matter of getting rewards, they should just eat the apples, right? They shouldn't stop and play the game. So there is clearly this concept of the game that they have learned. It's not just a learned behavior. They are actually playing the game. And showing self-restraint so that they can enjoy playing the game together. Another thing is interesting is they would intentionally withhold the symbol. So like maybe they don't give it the symbol for an apple. And so the ape wants the apple and it doesn't have the symbol for an apple. It might grab a can or something that has a picture of an apple or something similar to it. And it will actually improvise a symbol that then it can communicate with the other ape.

[01:18:08]  Bruce: Oh, it wants the apple and it can figure that out. So this really does show that apes have the ability to show reference. Where one thing represents another thing. OK, so Washoe the Chimp. This one's really interesting. And this is a one-off. So you could always explain it through coincidence. But this one's super interesting to me. So Washoe the Chimp lost her baby. And she became depressed. OK, because her baby had died and she wasn't taking care of herself. We'll talk about this in a future episode maybe about animal depression. Having mental anguish. So this chimp had been taught the Yerkish symbol so they could communicate with the chimp. So they went to her and they told her through the symbols that they were going to get her a new baby. And her depression immediately lifted. And she became excited. And she ran around super excited that her baby was coming back.

[01:19:01]  Bruce: Then they gave her the baby. And of course, it's a different baby because they can't bring babies back from the dead. And she gets super upset. And she won't take the baby. And she goes back to being depressed again. And then later, though, she thought the baby was cute. She ended up accepting the baby. And then the depression lifted again. It's really hard to explain this behavior if we're not going to accept that the chimp actually had a understanding but a misunderstanding. When it was told, we're getting you a new baby, it had thought they meant her baby. And that was why the depression lifted. And then once she had realized, oh, they actually just meant a baby, then she became upset again and went back into depression. So this is really showing some level of real language understanding, okay, even if it's very limited. And then this last one is Alex the Parrot.

[01:19:51]  Bruce: So let's do a parrot example. So for birds, they discovered they could train parrots by, they originally thought it was really hard to train parrots. And then they discovered that parrots will learn things if you train a human in front of it. So like you have one human train another human, like you have one human say, you know, if they perform some behavior, then they get the reward. And if the parrot sees the humans doing that, it will learn the concept. And then it will start to train to do the, to get the reward also. And they started to realize they can substantially train birds to some fairly complicated concepts by using this technique. OK, so Alex the parrot could respond and parrots can speak. So Alex the parrot could respond correctly in English. This is a quote from from Byrne from Thinking Ape on page 173. Alex can respond correctly in English to questions about objects and their properties.

[01:20:50]  Bruce: What color? What shape? How are they the same? How are they different? Focusing on shared or odd one out properties, his use of numbers up to six is particularly striking. He was not fazed by transfer tests with novel arrangements of objects, so there is no possibility that he is simply memorizing old arrays of objects or the distinctive patterns that certain objects make. He was not even put out by arrays of completely novel objects for which he knew no verbal label. When given mixed array objects, two hair clips and two bobby pins, for example, he would be asked the number in each subset and the total, and he could get the right answer for those, even though this is a novel set of objects he's never seen before. So it seems that Alex has really acquired the concept of numbers as well as the linguistic concept of taking turns and simple syntax.

[01:21:40]  Bruce: OK, so this is an example where, you know, how do you explain, it can't be explained by trial and error learning because it requires levels of abstraction that goes beyond simple trial and error learning. And it can't possibly be genetically preprogrammed because it's completely novel behaviors for [parrots]. OK, I've kind of thrown a whole bunch of examples here at you. And this is actually the end of what I had planned for today. Any thoughts or concepts on Byrne's methodology or on any of the examples that I showed you to try to explain how he has come up with, which animals have insight?

[01:22:19]  Cameo: What I can't make sense of is how after looking over all of this research, how did David Deutsch come away with his interpretation of Byrne's research?

[01:22:31]  Bruce: That's an interesting question. I'm not sure. I asked David Deutsch on Twitter about how much he had researched. I wanted to make sure I had read any of the papers that he had read. He had read two of Byrne's papers. He had not read his books. So I don't think he was even aware of all of Byrne's research. I think that he simply read the papers on behavior parsing, which we'll discuss, I think, next week. Behavior parsing is actually very interesting for a number of ways, both for Deutsch's point of view and for Byrne's point of view, at different levels, strangely enough.

[01:23:07]  Sadia: Although, hasn't David Deutsch recently, I can't remember where he commented, or maybe I'm wrong, that it seemed like he was contradicting his original view.

[01:23:19]  Bruce: Oh yeah, it was in our podcast, Sadia. You remember where we did the people asking questions of David Deutsch?

[01:23:26]  Sadia: That's right, that's right.

[01:23:28]  Bruce: He admitted that he thought probably dogs couldn't be faking their shows of feelings, that that wouldn't make sense from an evolutionary standpoint, that probably to show excitement, they have to be excited. Then he admitted though he was still skeptical of that view. He said, yeah, but in other ways you can show that they just don't understand anything. And he's right about that. So I do think that he started off with a harder viewpoint on this, but he's a true Popperian fallibilist. He has definitely suggested that maybe he's wrong and even given examples of how he might be wrong.

[01:24:07]  Sadia: I actually grew up having a lot of pets, and I was actually surprised to see like chickens who usually you don't see much of an expression. I remember one of my roosters literally got kidnapped by one of my neighbors. I was in Pakistan and I looked for it and that rooster was pretty close to me. Like I used to pet him. He would put his head on my shoulder and go to sleep. And, you know, we were pretty chummy chummy, but I never really thought he was that intelligent. But when he got kidnapped, I started calling his name. He was, by the way, his name was Heckel and I had a Jekyll too, Heckel and Jekyll. But anyway, so I called him out and I'm like, where's Heckel, Heckel, Heckel? This was the first time I saw this type of behavior. He actually responded back to me and I traced him where he was and I discovered where he was because he gave me the signal that he was in distress.

[01:24:58]  Sadia: Even though the people weren't treating him badly, all they did was took him and locked him up. If he was just such a dumb animal who was now just in the cage, but not really, you know, I mean, he was taken away from his surroundings. And, you know, usually chicken will make this noise which says danger, right? And he was making that sound and I literally discovered where he was. He was able to communicate where he was.

[01:25:21]  Bruce: He had done an abstraction, right? It's a very limited abstraction, but he had done an abstraction.

[01:25:25]  Sadia: I mean, when else did he came across kidnapping or, you know, most of the time the dangers are usually there's a hawk and they usually alert everybody and, you know, and to be honest with you, sometimes you see these animal videos and you're like, you know, you're kind of suspicious, almost like those UFO footage type videos that, oh, the animal is creating art and, you know, and then they never tell you what was, what went behind the scenes of how they got up to that.

[01:25:52]  Bruce: That's right.

[01:25:53]  Sadia: So that's one of the reasons I've always been really suspicious. I mean, the only thing I ever fall back to, you know, is that, I mean, you know, maybe some of the direct experiences I've had or in the case, you know, maybe, yeah, I don't know. I mean, even the people who have written books on it, sometimes you're suspicious that how much, you know, the tendency for them to anthropomorphize and then presenting what they saw, like how much of it has been affected by how they feel about animals.

[01:26:23]  Bruce: Yeah. Do you know, before reading Byrne, I mean, I've known people who like, there's lots of dog lovers out there. Right. I mean, like tons of dog owners. And if you were to ask your average dog owner, they definitely attribute intelligence to their dog that, according to Byrne, likely the dog doesn't have. Yeah. Well, they have emotions from Byrne's point of view. So that's not what I'm really talking about. But they would attribute theories of mind to dog. Right. That this dog understands what the other dog is thinking. Things like that, which according to Byrne is not possible. Now, keep in mind that everything's a guess. Everything's a conjecture. Maybe we'll eventually find that dogs have insight after all or something like that. But the fact that he can do so many experiments and all of them can be explained by trial and error learning, I do find that a fairly convincing argument.

[01:27:14]  Bruce: Right. I suspect that on the one hand, dogs are quite intelligent compared to, say, AIs. On the other hand, dogs are nowhere near as intelligent as sometimes we try to anthropomorphize them as, as owners. Right.

[01:27:30]  Sadia: Yeah.

[01:27:31]  Bruce: It's going to end up being something in between. They probably do feel things. They probably do have some really interesting forms of intelligence, in which case here I probably mean abstractions that they can abstract things. On the other hand, they probably just as a preview for future episodes, they probably have no sense of self. According to Byrne's experiments, they probably have no theory of mind, no concept of deception. Other than if I'm out the door [wording unclear], that means I can take the chair, no concept of why that works. They probably have no real physical understandings of things, of laws of physics, basic things that humans might have, kind of the way things work. I think we're going to be surprised that they're way more intelligent than we thought in some ways and way less intelligent than we thought in other ways. All right.

[01:28:24]  Unidentified speaker: All right.

[01:28:25]  Sadia: Well, nice talking to you there, Bruce.

[01:28:28]  Bruce: Yes. Thank you, Sadia. Thank you, Tracy.

[01:28:32]  Unidentified speaker: Bye-bye.

[01:28:32]  Bruce: Bye-bye. 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.

[01:29:23]  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.