Episode 79: Perspiration vs Inspiration
- Links to this episode: Spotify / Apple Podcasts
- This transcript was generated with AI using PodcastTranscriptor.
- Unofficial AI-generated transcripts. These may contain mistakes. Please check against the actual podcast.
- Speakers are denoted as color names.
Transcript
[00:00:08] Blue: Welcome back to the theory of anything podcast. Hey, Peter. Hey, Bruce. How you doing?
[00:00:13] Red: Good.
[00:00:14] Blue: All right. Well, let’s talk about where we kind of left off with our discussion so far. So kind of just as a summary, we talked about the walking robot example from beginning of infinity, and how it happens to matches Deutch’s own theory or criteria or properties, whatever you want to call it of knowledge that he lays out in beginning of infinity. This realization was the thread I pulled on the tapestry of history of knowledge looking for a fully consistent understanding of knowledge. We discussed how the centerpiece of Deutch’s theory of knowledge seems to not be his criteria or properties or definition of knowledge, but the two sources hypothesis, ie that there are only two sources of knowledge biological evolution and human knowledge. What we have here is a definition or a set of properties or a criteria of what defines knowledge or what knowledge is. And then we have this theory that there’s only two sources of knowledge. And the two do not match is really what I’m trying to show is that there is some sort of gap between them. Now, the gap may not be completely unfilable. I’m going to argue that it is unfilable. I have argued that it is unfilable, but maybe I’m wrong. Okay, so let’s leave open the possibility that maybe the criteria are just missing some criteria today or something along those lines, but that ultimately will find that he’s right that there’s only two sources of knowledge. However, what I’ve really done is I’ve really emphasized not the two sources but the criteria itself and I’ve said okay if we just look at the criteria. What, what other things that are outside two sources would we have to include.
[00:01:58] Blue: And we’ve talked in particular a number of examples of that including the walking robot itself, but also animal learning and the immune system and the immune system I called out in a past episode is particularly interesting example. Because it is an example of knowledge stored in the genome, or, I guess, technically, George would say it’s not knowledge but something like knowledge that’s stored in the genome, or in DNA. But it’s not passed between generations because the immune system actually modifies DNA in its own cells and hyper mutates it to try to find the right antibody. And then it stores that final recipe in its cells, but it’s not in sex cells so it doesn’t get passed to future generations. So you actually have a case here that matches every single part of Deutsche’s criteria or explanation for what knowledge is, and yet still, and is yet is outside the two sources and that he claims isn’t knowledge. I’m trying to cause as many problems as possible with that set of criteria in an attempt to learn from it.
[00:03:09] Red: Stress, stress testing it.
[00:03:12] Blue: I’m stress testing it.
[00:03:13] Red: Yes.
[00:03:13] Blue: Okay. Okay. We’d also talked about a defender of the two sources hypothesis who I called Henry. And, but I also mentioned in passing that there was another one that I had talked to I’ve talked to a whole bunch and a lot of times when I use these people Henry or James, their conglomerations of several different people.
[00:03:34] Red: We’ve got kind of a strange hobby, I guess, going online talking to people.
[00:03:40] Blue: So Henry isn’t the only defender of the Deutsche’s theory of knowledge that has like when I’ve brought up to him, hey, look, this criteria that you claim represents physically what knowledge is it includes things that are outside the two sources hypothesis. What Henry did is he made up new made up new criteria on the fly. He said, Oh, no, no. And he was often stated as as simply making things more clear. Oh, no, the criteria are correct. But you’re a misunderstanding criteria. One, it actually means this and he would like clarify this is the same as either making the criteria more specific or it’s like implicitly adding additional criteria and the example I used with Henry. Is that he claimed that there was an implicit criteria that it had that knowledge only counted as knowledge if it was created through replicators and if it existed for generations. Now that’s not stated anywhere, clearly. But he believed that that was an implicit criteria that needed to be understood. But he also made the argument you weren’t allowed to write it down because if it was explicit, then that would be a problem. And my argument was, look, I don’t care if it’s implicit or explicit. It’s a problem either way. The fact that you didn’t write it down doesn’t magically make the problem go away. Right. And so in a sense, Henry and I were agreeing there was a problem. He just didn’t think it was a problem worth dealing with. And I thought it was. But this is something I’ve seen come up over and over again that the that the moment I point out to a defender of Deutsches theory of knowledge.
[00:05:16] Blue: Look, what he’s saying knowledge is and the basic definition that he always comes up with is that it’s adaptive information that that causes itself to remain so or that it it has causal power. There’s a few different things that he’s brought out and I don’t care which ones you use you got the same problem no matter which ones he’s come up with. Okay. And every time I pointed out, look, that includes the immune system that includes the walking robot because those easily fit that definition or that criteria or that set of properties. That the first thing that always happens is that the the properties are modified to try to get back to the two sources hypothesis. Therefore, I’ve come to the realization that the central part of Deutsches theory, at least as perceived today is the two sources hypothesis, not the criteria. Okay. Because the criteria are allowed to be varied and the two sources hypothesis is not allowed to be refuted.
[00:06:09] Red: So, I guess the definition I keep coming back to is just the knowledge is useful information. One of the one of Deutsches so that in that sense the immune system would easily qualify certainly qualify useful information. Yeah, that’s right.
[00:06:25] Blue: Okay. So, now, Henry’s not the only one though that makes up new criteria on on the fly to try to get back to the two sources I also mentioned another defender James. And I argued to him that even if he was right that the immune system didn’t create knowledge. It at least created adapted information in the form of a recipe to create antibodies. And so was worthy of its own theory. Okay, so I want to be clear, I was not arguing to James that the immune system created knowledge I was arguing to James that it was creating something useful. And that it was adapted information and that’s all I was arguing I was not in any way implying it was knowledge. Okay. And here was the response he gave me he said, but I’ll hold you responsible to say to say then what you find so special about adapting information created by the immune system. It does not seem so special to me, because the trial and error process in the immune system is 100 % deterministic. So the result is to and thus a mechanical consequence. I see therefore nothing special or relevant about this kind of adapted information. Okay, that’s not quite a quote but that’s it really comes close to exactly what he said to me I’ve modified it to make it clear what what the context was. Now recall that the immune system literally hyper mutates DNA in the in the genome to create antibodies. If one is successful those cells are allowed to replicate more while the others die off. So it uses DNA in the genome. It just doesn’t pass into the next generation of organism, though it does to the next generation of cells.
[00:08:03] Blue: So it uses replicators and it uses variation and selection that is how the immune system works. But because the process is what he says 100 % deterministic the adapted information created by those processes so uninteresting he’s claiming that it’s not even worthy of being discussed or having its own theory or anything it can be ignored. Okay, and that is the argument he was making to me. Now just to be clear, the immune system is stochastic random not deterministic so that’s literally he’s wrong in what he states here. And I could at this point dismiss him as just he didn’t know what he was talking about. But in truth, I think he’s making a stronger argument than the words he chose so I want to steal man what he was saying, even though I recognize that he was mistaken. I believe that it doesn’t actually matter one way or the other to his argument, whether we’re talking about 100 % deterministic or 100 % stochastic, right. Without a doubt the immune system uses stochastic and this was from the Nobles paper that I quoted they state that it is stochastic. Okay, so I’m backed by an actual paper by actual scientists that know what they’re talking about. But it wouldn’t make a difference either way if if the immune system was entirely deterministic or whether it was random. It wouldn’t change the fact that it’s still of a mechanical consequence which is what I think his real point to me was. Therefore I could imagine instead he had not made this mistake and he had instead argued to me. It doesn’t seem special to me because the trial and error process in the immune system is 100 % stochastic.
[00:09:46] Blue: So the result is to and is thus of a mechanical consequence. And I see therefore nothing special or relevant about this kind of adapted information and it wouldn’t make any difference to his argument. And this is something you’re going to find over and over again is that a lot of times when people try to argue. Oh, that’s deterministic that it’s a pointless argument because adding randomness doesn’t change the argument in the slightest like like. This happens a lot with the free will argument where people say well there’s no free will because it’s deterministic. And then someone will point out well actually we live in a quantum universe and at least from the point of view of a single universe there’s actual quantum events that are actual randomness. And it’s like what doesn’t make a slightest bit of difference to whether there’s free will or not like not even slightly right so it’s it’s strange that people concentrate on determinism. I think what they really care about is the mechanism the fact that I can so easily explain to you how the immune system works. And it’s clearly just a mechanistic thing and it’s clearly not doing something magically open ended like humans or biological evolution that it just doesn’t seem special. And so that’s I’m going to take his argument in this steel man form. I don’t know if you had the opportunity to listen to that new league Lee Cronin interview with Lex Fredman, which was cheese her so much in there that that I think you would be interested in.
[00:11:12] Red: But the one of the things he keeps saying, again and again, he comes back to is that the universe is not big enough to contain the future. And sort of this isn’t sort of an argument I think against the universe being deterministic. But then again, what do it and others would say that the most probably most physicists really is that the universe is deterministic, but that the future is fundamentally unknowable.
[00:11:45] Blue: Yes, it kind of seems like two ways. I don’t know. I keep becoming that just stuck in my brain a little bit. It kind of seems like two ways of sort of saying the same thing. Does that make sense? Yeah, it does. You know, I they are two ways the same the same thing. But let me just say that they’re not equally the same. Okay. The Deutsch version is by far the more accurate and that the the Cronin version is far easier to misunderstand and to get confused about. And it’s at some level. I mean, like we had Saudi on the podcast for like four episodes and she kind of is more in the Cronin camp. I don’t actually think that her version of this and Lee Cronin’s are all that similar, but they have some commonalities of
[00:12:28] Red: fairness.
[00:12:28] Blue: And I don’t think there’s any doubt that Saudi looks at those two as different, whereas you’re seeing them as kind of similar.
[00:12:37] Red: Well, it seems like sort of like just arguing about what the word determinism means. Yeah, this is what kind of what I come back back to. And
[00:12:44] Blue: as it turns out, the word determinism actually has multiple definitions that get used even formally like not surprising. One of the things that I did not know until Dennis Hackathol pointed it out to me was that when we refer to a deterministic function, we mean a function that has has no side effects so that it will give the same result with the same inputs every single time. Now, it’s not that hard to have a function on a deterministic computer that utilizes a variable that wasn’t part of its inputs and therefore is a quote non deterministic function. Now, it’s a weird way to use the word determinism, because we refer to the function as non deterministic. But it’s running on a deterministic computer, it is 100 % deterministic in the way we would normally use that term. But we call it non deterministic. And I don’t even know why it’s just historical reasons we call it that, right? People will want some point confused as to whether this was determinism or not. So they called it non deterministic in the term stuck. And
[00:13:49] Red: it’s an
[00:13:49] Blue: actual use of the term determinism today. That’s formal in computer science, but has nothing to do with how people normally use that term.
[00:13:58] Red: Yeah, I mean, it just kind of seems on a on a just common sense way. If you say, oh, well, that that over there, that’s that’s deterministic. He kind of has the connotation that, well, yeah, you could you could work out what what’s going to happen. But, you know, it did. So if you say it, it’s it’s you are you understand what I’m saying.
[00:14:22] Blue: Yes. So what you’re really trying to say, well, we need to do a separate podcast on this. Get to get to words and what words are, right? And particularly Douglas Hofstetter’s ideas about words. OK,
[00:14:37] Red: OK,
[00:14:37] Blue: and how they they work with analogies and things like that. The simple truth is, is that words have halos of meaning around them. Yeah. And the word determinism comes with philosophical baggage, whether you wanted to or not that includes in most people’s mind this idea that you can predict what the outcome would be. And in the vast majority of cases, that is just not true of determinism. There’s there’s rare exceptions. OK, but and yet the word has taken on that connotation so strongly over time that people cannot get out of their mind that that isn’t an inherent part of the essential meaning of determinism. I intentionally use the word essential meaning because words do not have essential meanings, but people think they do.
[00:15:26] Red: OK,
[00:15:26] Blue: and the confusion comes over this need to apply the essential meaning of determinism includes. I can predict the outcome when, in fact, real life determinism, that just isn’t true.
[00:15:37] Red: OK, that’s a compelling way to put it. OK.
[00:15:40] Blue: And so we’ll have to get back to that example. And then I’ll actually want to read some of what Douglas Hofstetter said Hofstetter says about this. And really, I think he’s onto something interesting here. And it ties in well with Popper’s own writings about words and meaning. I don’t fully agree with Popper, by the way. I think Popper was very much on the right track. But I kind of want to pull together Popper, what he said about essentialism, what Hofstetter said about it. And then I want to kind of tease out under what circumstances do I agree with them and where do I maybe feel there’s some wiggle room that we should maybe question what they’re saying a bit. But I feel like they both have really interesting things to say on the subject that certainly have informed my thinking on the subject. So OK, so let me get back to the steelman case. So I think the important aspect here isn’t the determinism, but the seeming mechanical nature of the process that when James is arguing to me. There’s nothing special about the adapted information that’s created by the immune system that what’s really catching in its craw is that it is clearly a mechanical process that’s not that hard to understand. And so it is the type of outcomes it’s going to come out with. It’s never going to invent an airplane. It’s never going to do anything except what it was specifically meant to do. And so it just doesn’t feel special to him because it’s mechanical. Right. And I think you’ll see people, not just Deutschians, not just critical rationalists, although often them that they’ll kind of say, oh, that’s mechanical, so it’s not creative.
[00:17:21] Blue: Or that’s mechanical, so it’s not creating knowledge or something along those lines. OK.
[00:17:27] Red: You know, I got to be honest, that argument kind of feels right to me. I don’t know if I’m going to argue it for it, but it feels. I mean, the idea that the immune system is never going to build an airplane. Yeah, that kind of rings. There’s something fundamentally different there. Yes, we can’t define even if we can’t define what it is. But
[00:17:49] Blue: in your gut, you can tell there’s something special about it,
[00:17:53] Red: right? It seems to me. But whereas something like evolution essentially could do, you know, I mean, it can make things fly and do all kinds of things that human knowledge can’t do, at least right now. And it can make immune systems or brains or who knows what else. It does feel very different.
[00:18:20] Blue: OK, so I’m glad you said that because that’s a good setup for what I’m about to explain. OK, so that is the steel manned argument. And I think it’s interesting for a variety of reasons and I’m going to now deconstruct it to show the problem with it. OK, but also why it’s it’s kind of getting it something kind of true, but just not in the right way. OK, so we’re assuming this argument, the one you just made, the one that James made, the one that you hear all the time on the Internet. OK, there’s kind of it’s kind of assuming that knowledge must be special in some way. And if I can’t
[00:19:01] Unknown: demonstrate
[00:19:01] Blue: to him the quality of specialness, he’s going to not only refuse to consider it knowledge, but he will refuse to even acknowledge it is worthy of a discussion or its own theory at all. OK, and also notice the clever way in which he he tries to put the burden of proof on me. Well, it’s up to you to prove to me it’s special. OK, what I want to point out is that this is exactly amounts to adding an additional criteria, just like Henry did. OK, essentially it’s saying the adaptive information does not count as knowledge if it comes from a purely mechanical process. Or possibly we could read it as the adaptive information does not count as knowledge if James or Peter or whoever you’re talking to about this doesn’t consider it special.
[00:19:53] Red: OK,
[00:19:54] Blue: these are not part of Deutsche’s existing theory, right? When he tries to say knowledge is physically, you know, these three properties and, you know, it’s adaptive information that causes itself to remain so. There is nothing in there about James finding it special or even about whether it was created by a mechanical process or not. OK, so this is additional criteria. I’m going to I’m going to argue. I’m not going to argue it is additional criteria that just clearly it is. OK, now that doesn’t make it wrong. Maybe those special criteria are true. OK, maybe knowledge requires that it doesn’t come through a mechanical process or requires that it is special in some way, but we can’t define what that specialness is well today. OK, I’m not ruling that possibility out. But I’m being explicit that we are adding to the properties of what counts as knowledge beyond what Deutsche’s constructor theory of knowledge calls for. OK, secondly, I want to point out that he’s just not entirely wrong. The adaptive information created by the walking robot or the immune system really is completely mechanical. And just it’s kind of obvious it’s never going to make an airplane, right? It’s something very different about it from what human beings do. So why is he so convinced that knowledge can’t be created by a mechanical process? OK, why is it that the moment we can see the mechanism and understand the mechanism, it makes us want to say, oh, that’s not creativity. OK, this is actually a variant of an argument made by Deutsche himself, namely his argument that knowledge is created via inspiration versus perspiration. The inspiration versus perspiration argument. OK,
[00:21:50] Blue: now let me say that the way Deutsche originally uses this argument in beginning infinity is really pretty innocuous and fairly benign. OK, and I’m going to actually quote it. I’m going to prove that’s the case. However, I do think that the way it has mutated over time, including sometimes the way Deutsche uses it himself, it’s something far less benign at this point. It’s a problematic argument. And what I really want to do is I want to call out that as it was originally used in beginning infinity, it said something meaningful. But as it gets used today, it is clearly a mistaken argument. And I want to kind of tease out the difference between the two, what it originally said and how it gets used today.
[00:22:32] Unknown: Does that make sense?
[00:22:34] Red: Yes, it does.
[00:22:36] Blue: OK, so let’s go back and let’s look at what he actually said. So on page 36 of beginning infinity, he says the inventor Thomas Edison once said, none of my inventions came by accident. What it boils down to is 1 % inspiration and 99 % perspiration. OK, this is the basis for the inspiration versus perspiration divide that that then becomes a part of this argument.
[00:22:57] Red: OK.
[00:22:59] Blue: So this idea has been imbibed by online Deutschians in particular as something like if the algorithm is just is just an automated and mechanical process, then it is not true creativity and thus does not create knowledge. Now, nowhere does Deutsch actually say this in beginning infinity. In fact, I’m going to demonstrate that he not only didn’t say this, but that he really kind of indicates that that just isn’t a correct reading. The problem with me saying that, though, is that I’ve actually seen Deutsch make some variant of this argument himself, where he’s basically implying that if it’s a mechanical process, then it can’t be true creativity and thus it can’t be creating knowledge. So I don’t know. I mean, like there’s the difference between the way Deutsch used it in beginning infinity, which was a far more limited case. And even the way Deutsch uses it today. And that’s why I want to do a podcast episode on this because I really feel the need to tease out the difference between those two uses. OK. Notice, though, that James’s argument is typical. He identifies the process is mechanical. Thus it can be automated or one might argue it is automated that the means system just is an automation. And thus it doesn’t create knowledge QED. This is the essence of James’s argument to me. And this is actually Peter, the essence of your argument to me that you just made. OK. And basically everyone I’ve ever talked to that has brought up, oh, that’s mechanical. Right. Why do they understand twice this way? And is this what Deutsch actually said? OK. Well, on page 36, Deutsch says the inventor Thomas Edison once said again, none of my inventions came by accident.
[00:24:38] Blue: What it boils down to is 1 % inspiration and 99 % perspiration. OK. Here’s the thing, though, despite Edison saying this about human creativity. I want to emphasize this again. Thomas Edison was saying this about human creativity. OK. That it is 1 % inspiration and 99 % perspiration. OK. Deutsch feels that this is misleading. And here is what he says. Some people say the same about theoretical research where the perspiration phase is supposedly uncreative intellectual work. Deutsch argues that, quote, but the fact that a computer or a robot can perform a task mindlessly does not imply that it is mindless when a scientist do it. This is all still on page 36. And then he gives an example on page 36 of computers playing chess using an exhaustive search, whereas humans use creative and enjoyable thought. That last part is a quote. And then on page 58, Deutsch argues the perspiration phase can be automated. This is where the whole automation starts to come in and why it is that James is arguing with me over whether it’s automated or not. This seems to be what James had in mind. The immune system or the walking robot isn’t creating knowledge because it is automated and thus uses perspiration, i.e. it is mechanical instead of creative, i.e. inspiration. Given this view, James has no reason to look into how the immune system works because whatever it is doing can’t be that impressive or even worthy of further discussion because it’s all part of the perspiration phase. Okay. So James’s core argument is that because the immune system is just mechanical or automated, he sees this is a quote, nothing special or relevant about this kind of adaptive information.
[00:26:27] Blue: This is not, this really is not an atypical reading of Deutsch. And I’ve had many, many, many, many, many other people make similar arguments dismissing the need to even consider adaptive information created outside the two sources using this means. But here is, but here we now have a pretty big problem. And I really want to make sure people understand that what we’re trying to do here isn’t to dogmatically defend a dogma, but to intentionally seek problems in the theory and try to work them out. Okay. And here’s the problem. Everything in physics can be simulated on a mechanical machine called a Turing machine. So everything is mechanical. Literally, everything is mechanical. Every algorithm, including the AGI algorithm, once we know what it is, will turn out to be mechanical. Okay. And in fact, the same argument goes for deterministic. Okay, whatever it is the AGI algorithm is doing, it will be deterministic because algorithms are always deterministic. Okay. So to be fair, unless you believe in something like assembly theory, right, that would be going against what I actually don’t think it would be. Oh, now I haven’t studied assembly theory close enough to know that for sure. And I’m planning to study it and then do a podcast on it. Okay. But let me let me just explain why we believe all algorithms are deterministic. Okay. There’s really only two things deterministic and stochastic. Right. I mean, there is there’s nothing in between. There’s nothing that like if I were to say, give me an example, something that’s neither deterministic nor stochastic, like literally go try to give me an example of that. You can’t. Right. There’s an excluded middle here. The whole world is broken into determinism or stochastic. And that is it.
[00:28:23] Blue: Okay.
[00:28:24] Red: Yeah.
[00:28:25] Blue: Now there’s a completely interesting question here. Okay. Is the Turing machine a universal computer? That is the question. And we’ve, Deutsche has answered that. And the answer is at least so long as we’re assuming quantum physics holds. The answer is that yes, it is the quantum Turing machine. I guess I should probably say is the universal computer and really technically so is the classic Turing machine. It just has a different computational class. Okay. But they can both they can both compute any algorithm. Okay. Now there’s a really interesting thing that people bring up who if they don’t understand computer science and they don’t understand computational theory. It’s very natural, very, very natural to say, well, wait a minute, nature is full of randomness. It’s not deterministic. I mean, quantum physics, at least from the point of view of not the multiverse, but the point of view of the unit of a single universe. It is without a doubt, random. Therefore, how can you, how can you say that, that all algorithms are deterministic? It may be their algorithms that require random processes and that they can only work with random processes. Okay.
[00:29:35] Red: That’s what I thought until I was like 45. Yes. Yeah.
[00:29:39] Blue: Okay. Now, there’s a the thing that this question is good as a question as this is, it’s absolutely misunderstanding like fundamentally misunderstanding computational theory.
[00:29:51] Red: Okay.
[00:29:52] Blue: They have already thought of this, of course. And one of the proposals that has been made over the years is what if we made a Turing machine and we attached a random number generator to it? Can it compute algorithms that the Turing machine can’t? Okay. And this is a really interesting question that we want an answer to. And this is not something that somebody just barely thought of. This was thought of a long time ago. And the answer is we’re not completely sure, but we don’t know of any. Every single algorithm we can conceive of that we do actually know of some algorithms that perform better if you allow that random number generator. Okay. So I’ll even go that far, but we don’t know of any algorithm in existence that requires randomness. And the reason why is because there’s this thing called pseudo randomness. And every single algorithm that you would think requires randomness, you can plug in pseudo randomness and it will still produce good results. Okay. Not necessarily identically good results. So randomness does maybe have a place, but there is no algorithm that requires randomness that we know of. Therefore we are conjecturing and we have conjectured and that there simply aren’t any algorithms. And that’s part of the church Turing thesis as we understand it today. Okay. And no, it could be wrong. Right. I mean, it may be that Deutsches proof using quantum physics is going to turn out to be false because once we find out what quantum gravity is. And it may be that there is some algorithm out there that requires use of randomness. Okay. The key point here is, is that that is just a wild conjecture, a wild speculation.
[00:31:41] Blue: And the current theory says no, there are no algorithms that require random processes.
[00:31:47] Red: That seems to really get to the heart of it. So would you say that in order to falsify the church Turing Deutsch thesis, you’d need to find an algorithm that requires randomness.
[00:32:00] Blue: That’s one of the ways you could falsify. There are many ways that you could falsify the church Turing Deutsch thesis. And it’s well understood how you would go about falsifying it. Okay. Just nobody knows how to. In fact, they can even theoretically falsify it. Right. Like if you don’t hold yourself to the laws of physics and you say, OK, I’m going to build a computer that doesn’t follow the laws of physics. It’s very easy to falsify it. Okay. So typically we would define falsifying it as. Well, it has to actually be physically possible to implement the computer. Okay. And we don’t know of any way to do that. Okay. That’s why it’s our best theory today. Hmm. Okay. So. Having said all that. That’s why I can say with complete confidence, at least so long as the church Turing Deutsch thesis holds. That the AGI algorithm is going to be workable as a hundred percent deterministic algorithm. That’s entirely mechanical. Now, once you realize that’s the case, you realize why this argument, no matter how good it feels in terms of intuition, has clearly got something wrong with it. And the question is what is it like? Like I want to know, like you don’t get to just tell me, well, that’s mechanical and therefore it’s not creativity. I want you to explain to me how an algorithm could be non -mechanical. And if you can’t do that and you can’t, then I consider that a bad argument until you can. Right. If you can, if you can actually show me, okay, here’s what I mean by a non -mechanical algorithm. And if you can explain that to me and give me examples of it, then hey, we’re in business.
[00:33:46] Blue: But until that moment comes, this is not a good argument. It is a bad explanation. Okay. And I don’t know any way around that. That is just how critical rationalism works. Right. This is a hard, fast thing that I can’t break out of. It doesn’t matter to me how intuitively obvious the argument seems. I completely acknowledge it seems like an intuitively obvious argument that we should be able to dismiss something as not creative and not creating knowledge if it’s just a mechanical process. I totally at an intuition level get why people use that argument. And I know it to be wrong under current theory. So I don’t know what else to say. I just can’t accept the argument unless you can first show me what’s wrong with the church during Deutch thesis. Okay. And that’s kind of where I’m stuck at this point with this argument. So it wouldn’t make sense at least to take Deutch’s argument as meaning all mechanical processes don’t create knowledge as clearly most people seem to be taking it. Sometimes even Deutch himself seems to be taking it that way because that would be at this point a supernatural claim. Now, is this what Deutch actually meant by his perspiration argument? Well, at least in the beginning of infinity, I’m going to argue that he never says this and it is not what he meant in context. Okay. That as much as people interpret it that way today, maybe even Deutch himself interprets it that way today, that that is not the argument he made in beginning of infinity. So Deutch now, why do I say sometimes Deutch seems to be using this using the argument this way? Well, this is an episode 44.
[00:35:20] Blue: I actually had a conversation with David Deutch. And when I tried to explain to him some of these things, he almost immediately went to, well, that’s a perspiration algorithm. And I at that point explained to him, well, I don’t know if I can buy that argument and he stopped thought about it for a while and then came back and gave me a much better answer, which was, well, it doesn’t define its own problems, which I thought was incredibly brilliant, right? He took my concern seriously. He thought about it. He didn’t necessarily say, oh, you’re right. It was just more like, okay, I can see that argument is not going to work with you. Let me think of a better one, right? That fits what you’re thinking better. And this is all contained in episode 44. However, let me just say it is a bit difficult to tease out Deutch’s original meaning. And so I can see why people find it confusing and end up accidentally falling into the, if it’s mechanical, it’s not creative point of view. But let’s do a closer reading of what he actually said.
[00:36:17] Red: I think how I kind of interpreted it was, yes, the human consciousness is ultimately mechanical or algorithmic, as would, is consistent with the church terrain, Deutch thesis. But, you know, there’s still something that’s clearly missing, very mysterious going on there. I mean, if we understood it, then we could program it and we would understand what human consciousness is. So, you know, we don’t know what the mysterious missing element is, but, you know, clearly there’s something there. I guess that’s how I kind of interpreted it. Well,
[00:37:01] Blue: okay, I guess I don’t need to finish the rest of this podcast. I think you basically summarized my whole point.
[00:37:06] Red: Okay, well, that sounds like a compliment, I guess then. All right, let me go ahead and do the podcast anyhow, and I will try to tease that out.
[00:37:16] Unknown: Okay,
[00:37:16] Blue: so what about Deutch’s argument that the chess algorithm uses perspiration instead of creativity? All right. Deutch isn’t wrong that chess algorithms back when beginning infinity was written were just mechanically tried every possible move out as far as it could. Plus, usually a clever human created board evaluation algorithm. If you want to learn more about this, I actually cover this in detail in our AI episodes. I forget which episode numbers those are. But it should be obvious that everything, all these episodes tied together in various ways because they represent how I’ve tried to improve my thinking and my attempts to figure out how to move forward with thinking about AGI and how hard that is. Okay, but chess algorithms, big blue, which was the chess algorithm that beat the world champion back in the day, which would be the one that Deutch would be writing about in beginning of infinity. It contained a very clever human created board evaluation algorithm. And then it was just really good at searching out every possible move as far out as it could. And it got to the point where no human could beat it because it could search out forward forward a number of moves that no human could possibly compete with. Okay. In other words, it was 100 % perspiration. It’s just trying every single possible move and it’s just searching every single possible board position it can as far as it can before it runs out of time. And it’s able to get far enough that it can beat any human. Okay. No human could possibly compete with that. Yet humans back then often beat such algorithms with prior to deep blue with some small portion of the same computing power. Okay.
[00:39:04] Blue: And this is what Deutch, this is the world context that he’s discussing this in, that when humans play chess at the time, they’re somehow usually beating these other than deep blue. Like today I can make a, we can make chess playing programs that can beat any human including the world champion and it can run on your desktop. Right. But back then that wasn’t physically possible. And so your average chess playing algorithm humans would, would often if they were good enough beat them. And they would do it with some small portion of physical processing compared to what the computer was doing. Okay. So we knew that humans weren’t playing chess in the same way that big blue plays chess. And in fact, there was a number of interesting studies about this at the time. Okay. So they studied this and what they found is that humans were doing pattern recognition. So they could recognize at a glance patterns on the board no matter where the pieces were located on the board. And now how do we know that? Okay. Well, they came up with a really clever experiment. And, you know, when, whenever people say, oh, they can’t know that, like, just don’t ever say that to a scientist. Like just don’t because scientists are incredible at coming up with ways to test things. And like if you think, if you buy into the idea, I know we’ve talked about this a few times that, that you cannot possibly know anything at all about if animals feel things or not. Do not say that. It just isn’t true. Scientists are incredibly clever at coming up with test cases.
[00:40:41] Blue: And while it’s never certain, which it would still be fair to say, well, they don’t know for sure, but that’s always true. They can basically take all the competing theories and just make them problematic. Right. And come up with super clever test case. And we gave the example of animal grief, like that’s an incredible test case that I would never have thought of. And I would have thought it was impossible to even try to test if animals feel things or not. And yet they came up with a way that once I saw it, I went, OK, well, I can see it doesn’t prove anything for sure. But man, I can’t even come up with a good alternate explanation anymore. Right. Not at least not one that’s super that isn’t super vague. And just never tell a scientist you can’t test that. OK, so how did they actually test this? What they did is they would put together board conditions. And then they would try to get people to recreate that. You know how you do like the memory experiment where you see like a plate of stuff and then you try to write down as much as you can remember or whatever. It’s the same sort of thing where you get to look at a board and then you have to try to recreate the board as closely as you can. OK, if you are a regular person, your ability to recreate the board you just saw would be not that good. OK, let’s say level X, OK, on average, you can recreate the board with X level of fidelity and it’s kind of low. OK, I don’t know exactly what the numbers would be, but it’s you can imagine a set normal level.
[00:42:13] Blue: OK, if you give that to a master, a chess master, then one of two things happens. If the board set up was random, the chess master’s ability to recreate the board is exactly the same as a normal person’s. But if the chess board was not random and in fact represented a completely valid. Board position that would happen in gameplay. Then the chess master would be able to recreate the board with a very high level of fidelity. OK, so clearly what was going on is the chess master was super good at picking out patterns. And once those patterns existed, he had choked them and they weren’t just random pieces on a board anymore. And he could therefore recreate the relationships that he saw at a glance and he could come up with what the board position was. OK, do you follow this experiment and how they did it and why they did it?
[00:43:11] Red: Well, as someone interested in chess, I find this quite an interesting assertion. So what you’re saying is that when these these chess masters who can just like basically they just can play chess in their head. Right, where there’s there’s like that scene in the Queen’s Gambit where she’s in the in the car. Great show,
[00:43:28] Blue: by the way.
[00:43:29] Red: And yeah, yeah. And they’re just playing playing chess in their minds. And, you know, I think a lot of really good chess players can can do that. What you’re saying is that if you take someone like that with that kind of ability and just give them a chess board to like just a completely random configuration of pieces to memorize. They wouldn’t be any better at it than anyone else.
[00:43:54] Blue: That’s right.
[00:43:55] Red: OK, well, that’s that’s interesting. I never thought about it like that. So really what they’re doing is like memorizing the the history of the board or
[00:44:05] Blue: the relationships. Like what we’re not entirely sure like doing it right. They are somehow chunking it in some interesting way. OK, finding patterns and chunking it up. We know that’s how humans work, right? When we can chunk something, it allows us to like think about like if I were to give you a random set of numbers, you would have a hard time remembering it.
[00:44:29] Red: But if
[00:44:29] Blue: you chunk it up, suddenly you can remember a much larger set of numbers. That’s why they make
[00:44:34] Red: it into a song or something.
[00:44:36] Blue: There’s various tricks that we do great relationships. OK,
[00:44:39] Red: OK.
[00:44:40] Blue: So maybe this isn’t too surprising, but like that is what they found out from the experiments. OK.
[00:44:45] Red: Yeah, no, that’s interesting. So
[00:44:47] Blue: based on that, we know that human beings did not play chess the same way that a computer plays chess computers playing chess. It’s not like human beings don’t think out forward, like try to work out if I make this move, then he’ll make this move, etc. But human beings just don’t have the computational power to just broadly do that across every single possible move. So instead, what they do is they significantly pare down the search by using these pattern matches. OK, they kind of just know these are the most likely to be moves I should be exploring. They kind of just sometimes even just as a gut feel that may not even have an explanation when we get to go. We’ll see that it’s often entirely a gut feel that this would be a good move and this would be a bad move. And then they try to think out how many moves they can. And however much time they’ve got, they’ll try to work out some future moves. But clearly they’re not doing an exhaustive search like a computer does, right? And instead, they’re doing this pattern match. OK, now here’s the problem with this argument. While it was completely correct at the time, Deutsche made it. We then invented AlphaGo. Now, AlphaGo, if you want to know more about this, see Episode 34. It’s far more like how it plays like far more like a human does. It does not concentrate on brute force moves. Why? Because go you cannot possibly play well just doing brute force exhaustive search. The search is just too large. So what they needed to do is they needed to teach the computer to do the same sort of intuitive pattern matching that humans do.
[00:46:31] Blue: And that’s what machine learning is. OK, so and that’s what they ended up doing. So AlphaGo replaced human intuition about what patterns are good patterns by using reinforcement learning. And so it’s much closer to what a human master does today when playing go or chess than what Big Blue did when it was playing chess. OK.
[00:46:55] Red: Does that lead into what these LLMs do with language? They can look for these patterns in kind of a similar way. Yes. OK,
[00:47:07] Blue: so the advent of machine learning has been around forever. But I don’t think we realized how capable it was until a couple of decades ago. Right. It took everybody off guard that when once you get enough data and enough computing power that you can run the data through that machine learning can start to look a lot like human intuition. And in fact, let’s in fact, this is in my episode 34. But AlphaGo, it’s intuition for what how good the board is and is so good that it can make it can look forward only one move. And it can determine which move to make is the best move at a level that puts it at professional levels. OK, so it literally doesn’t have to even do a board search anymore. It just simply has to try out every single move one move forward and it can already play at professional levels, not to the level where it can beat a go master. That requires more of a board search, but it can play at professional levels by looking forward one move because it’s that good at recognizing board patterns. This is far closer to what humans do. So this example that Deutsche is using, I’m going to say it has not aged well that the advent of AlphaGo has made it so that he’s not wrong. But the way the example he’s using needs some improvement now. OK, we need to tease it out better. So AlphaGo does still use it doesn’t have to use perspiration anymore to play well. Unless you look at the pregame and here AlphaGo does use perspiration. OK, so essentially it plays billions of games to learn which patterns are the best patterns.
[00:49:02] Blue: OK, now I want to make this distinction because this is going to be important to my explanation. Deutsche was talking about perspiration in gameplay itself, which is the way we played computer chess back then. Today we don’t need that kind of perspiration to play chess well. OK, but you do still need a different kind of perspiration specifically prior to the game where they were basically AlphaGo or AlphaChess needs to play itself billions of times to learn the good patterns. OK, so AlphaGo masters so AlphaGo does still use perspiration and it’s tempting at this point to just shift and say, well, well, that’s what Deutsche really means. He just means there’s perspiration somewhere in the process. OK, even though his example was perspiration in gameplay, Bruce, you’re you’re you’re taking him to literally he just meant perspiration somewhere in the process. The problem with saying that is that AlphaGo masters human AlphaGo masters also use perspiration and chess masters to write. They learn those patterns over thousands of games that they play. That’s how they get good at it. Right. And so they’re also using perspiration. Now, at this point, you could arbitrarily say, well, Bruce, thousands of games don’t count as perspiration, but billions of games does. Thus, QED perspiration is what separates AlphaGo from a human go master. You could say that I will admit. But boy, that is a completely arbitrary distinction at this point. Right. What you’ve done. And this is why we did an episode on degrees of freedom. Freedom is you’ve conveniently introduced a degree of freedom into your theory so that you can defend it from from refutation, which is precisely what you should not be doing as a critical rationalist.
[00:50:57] Blue: OK, you need to take seriously the fact, even though, yes, there’s a difference here that I have now shrunk that difference substantially. And I’ve forced you into a position where you the only way you can try to declare victory now is to arbitrarily make a distinction between billions of games as perspiration versus thousands of games as perspiration. That should catch your interest at this point that I’m on to something important and that there’s something interesting that we can learn here. OK, so let me suggest that human AlphaGo masters do put a lot of perspiration into getting good at a game. And because of that, even if it’s not as much perspiration, I’m uncomfortable with trying to make a simple divide here between humans and machines. What I really want to know is what is the divide. And of course, no one knows the answer to that, but we’re trying to figure out what it is. And it’s really important that we push as far as we can and get into territory where we know we don’t have answers because that’s where the search needs to take place. OK, rather than become comfortable with, oh, it’s just perspiration. And then arbitrarily try to make a distinction like, oh, well, humans don’t use as much perspiration as computers. Therefore, you know, perspiration and try to end the argument there. So I do see Deutsche’s point, though, that even with AlphaGo, there is a perspiration phase and it is still different than the human perspiration phase in some way, namely that one takes billions of games and one takes thousands of games. OK. But I want to say that it is closer. AlphaGo is clearly closer to how a human master plays AlphaGo. And
[00:52:29] Blue: clearly it is not, but it is still clearly to Deutsche’s point not identical to how a human plays. OK, I suspect the answer here is that humans have explanatory knowledge that AlphaGo does not. And I think Deutsche would probably approve of me saying that. OK, that is to say that humans that play chess and go use three kinds of knowledge creation. One, brute force trying out future moves as far as they can. Two, intuitions from experience or implicit knowledge of how to prune the brute force tree as well as intuitions as to what a good board position looks like. And three, explanatory knowledge that helps them decide how to prune the tree and what a good board position looks like. Big Blue, when it beat Gary Kasparov, used only source of knowledge number one. AlphaGo use source of knowledge number one and two. And a human uses source of knowledge one, two and three. So humans are still different than AlphaGo, but AlphaGo has reduced that difference and shown it can’t be just perspiration. That third process, this explanatory knowledge that they use to help them play, that is what Deutsche calls and explicitly defines creativity to be. He defines creativity as the ability to make new explanations in beginning of infinity. It is something that humans can do in machines at least so far can’t and we don’t even understand it well enough to even talk about it meaningfully or make an algorithm for it. What I really want to know is what do you specifically mean by creativity? And we can tell creativity and can we tell creativity from non -creativity merely by seeing if the process uses perspiration or not. That’s really what I’m trying to ask.
[00:54:25] Blue: Okay, now Deutsche defines creativity as the capacity to create new explanations from page 30, page 30. Surely AlphaGo creates no new explanations. So when defined this way, sure, it is not creative. I have no doubt that under that definition of creativity, AlphaGo is not creative, but neither is biological evolution as biological evolution creates no new explanations either. And so it is not creative in that sense either. I really want to stop and dwell on this for a bit. If you’re taking creativity to mean the capacity to create new explanations, biological evolution is not creative. And in the episode 44 where I talked with Deutsche, he admitted that to me and said, of course I use the word knowledge and creativity in different ways in different contexts. Okay. And to him that made no big deal that evolution is not creative in this sense because clearly it’s creative in some other sense. All right. And that was really how Deutsche approached it with me when I talked with him about it is he just didn’t care that biological evolution was not creative in the sense of creates no new explanations. Because he knew plain well that the word creative didn’t always mean the capacity to create new explanations. And in the case of biological evolution, it was creative in some different sense. Okay.
[00:55:50] Red: To be fair, I mean, he’s, you know, I think he just sees philosophy as the search for truth rather than the search to define words. Right. I mean, there’s, I don’t think there’s anything wrong with that.
[00:56:03] Blue: There really isn’t.
[00:56:04] Red: Yeah.
[00:56:04] Blue: Here’s the problem though. When people want to say the immune system is not creative. It’s not too hard to see that since there is no single definition of creative that really the word creativity could easily be stretched to include the immune system. Okay. That it’s just a different kind of creativity. Okay. So getting caught up on the word creative really makes no sense at all. It’s got nothing to do with anything. Okay. The immune system is clearly not as creative as a human. I’ll give you that easily. Like I’ll just give you that. Right. But it is creative in its own way. Okay. And what do I even mean by that? Well, I don’t know. Right. I mean, like the word creative is kind of one of those vague terms that kind of means different things in different contexts. Right. And so just this whole idea, it’s mechanical and it is therefore not uses perspiration and is therefore not creative. It is. It is a flawed way of looking at trying to even understand what it is that humans are doing.
[00:57:04] Red: Okay.
[00:57:05] Blue: So you can’t really say creativity is the ability to create new explanations per se. Like even that can’t be quite right because there’s this thing in machine learning called explanation based learning. An explanation based learning give as a computer the ability to create explanations. Now, I don’t explanation based learning is not some giant creative breakthrough. In fact, it’s inferior to regular neural nets. Right. It is. It has not lit the world of machine learning on fire. Now you could make a couple of things about, okay, well, maybe they’re not real explanations, but like nobody knows what an explanation is. Right. I mean, there’s no way to correct. And Deutsch actually claims that it’s impossible to define what an explanation is because you could always invent new modes of explanation. Okay. Surely what explanation based learning is precisely what popper had in mind where it’s it’s specifying universal laws that can’t be violated and a single violation refutes it like that’s exactly how explanation based learning works. Okay. So any serious attempt to say, well, what popper had in mind when he spoke of explanations machine machine learning explanation based learning would qualify as explanations. It would probably qualify as well as anything you can imagine as explanations because we just simply can’t imagine anything more than what they’re doing or they would have tried it by now. So I want to suggest that even just saying the ability to create new explanations that that that is an deficient understanding of human creativity. Yes, human creativity includes and in somehow it’s really important that we have this capacity to create new explanations, but even that by itself can’t be enough to understand what makes human creativity special.
[00:58:54] Blue: Because if that was then explanation based learning should count as equivalent to human creativity and clearly it is not equivalent to human creativity.
[00:59:04] Red: Okay. So
[00:59:05] Blue: I’m going to say something else is missing. We don’t know what, but it’s related to the problem of open -endedness in some way which is why we did an episode on open -endedness before I started this whole thing. Now it seems to me that when Deutsch and his defenders of his theory use the word creativity, they’ll try to make it sound like they’re saying something explicit, it’s the capacity to create new explanations. But what they really mean is whatever it is that humans are doing that makes them so much smarter than any animal or machine today, that is what we mean by creativity. Oh, and it’s something unspecified somehow unspecified related to explanations. I think that’s what they really mean by creativity, right? Is anytime I make them get explicit, I’m going to be able to find counter examples. It’s clearly something that we don’t understand and they’re referencing whatever that is that we don’t understand and they want it to sound more explicit than that. But really that’s what they mean. They really mean that thing that humans do that we don’t know what it is that’s creativity.
[01:00:07] Red: Well, I think that’s what I played right into with my earlier comment. I would agree that that’s a fair interpretation then.
[01:00:16] Blue: Okay. Now let me point out though that this isn’t unfair. Humans are doing something special compared to animals and machines.
[01:00:24] Red: Sure.
[01:00:25] Blue: Now, I’ve argued throughout this series, I’ve argued that the difference is that the search is open -ended for humans and for biological evolution and that it’s not for all the other algorithms we know about. Okay. Now, I don’t know that for sure. Okay. And even me saying that isn’t really an answer because of course I have no clue how to even explicitly define open -endedness, right? I can only give you vague examples of it. Okay.
[01:00:49] Red: And it seems to me Deutsch would probably agree with that too. I mean, isn’t that sort of the whole premise maybe of the, or at least one premise of the book is that we’re at the beginning of infinity? I mean, that’s kind of another way of expressing this concept of open -endedness, right?
[01:01:06] Blue: Well, I think so. Yeah. If Deutsch really had fully imbibed it in that way, then I don’t think he ever would have used the walking robot example. And in the next podcast I’ll explain why I think that is. Okay. So I actually think that there is a hidden flaw somewhere in Deutsch’s thinking, but that he’s basically on the right track, if that makes any sense.
[01:01:27] Red: Mm -hmm.
[01:01:28] Blue: Once you, the idea that the problem is that it’s an open -ended search defines the problem differently than trying to figure out what’s physically different about what it is that humans produce. Deutsch, his, his constructor theory of knowledge is fundamentally about how do we physically define what knowledge is? When what I’m suggesting is it’s got nothing to do with what knowledge physically is. It’s got to do with the nature of the open -endedness of the search. And therefore I’m going to argue that while Deutsch may in principle agree with me humans are open -ended and things like that, that this has implications for his constructor theory of knowledge that he hasn’t fully accepted yet, including the fact that he needs to drop the two sources hypothesis. But, and this is where I’m really going with this, right? We could define knowledge physically. I’ve got no problem with doing that. And I like how Deutsch went about trying to do that. But trying to do that is ultimately a red herring. What you really need to be looking for is what makes the open, what makes an open -ended search special. And that’s the real place you need to be putting your research as an AGI person. So let me get back to the why I think and that it’s still not unfair to say that even if what Deutsch and defenders mean is creativity is that thing that humans do that we don’t know what it is. That’s not really unfair. Okay. Like, I need a word for whatever that thing it is that humans do that I don’t know what it is. And naming it creativity is not necessarily a bad idea.
[01:03:05] Red: Yeah.
[01:03:05] Blue: So it is as long as we understand it’s something undefined, right? Like we don’t want to we don’t want to mistakenly think that human creativity means the capacity to create new explanations. If that’s only one little part of what we mean, right? It’s there’s something else more going on. Okay. And that’s really what I’m trying to argue here. Yes, that’s that’s important, but it’s insufficient. Now a fair tweak to be more accurate would would be that human creativity, not every kind of creativity is rooted in the capacity to create new explanations in an open -ended way. And I think that’s really what Deutsch means to say. Okay. I think he really means to say that creativity is the capacity to create new explanations in an open -ended way. And that he didn’t maybe at the time know about the problem of open -endedness. I’m sure he does now. And he didn’t realize that it was an important part of what we mean by creativity. Okay. It is also unclear why perspiration must be the opposite of inspiration. Okay. And let me let me make a case here. Edison was referring to human creativity as 99 % perspiration. And yes, Deutsch thinks Edison did not understand his own experience. I’m not so sure. Deutsch is right about that. Let me use some examples here. Okay. So first of all, let’s look at Edison himself. Why did Edison say that it was 99 % perspiration? Well, he had this idea, the inspiration that we could somehow make a light bulb by finding some piece of metal that would just. Light up. And then we could keep it from breaking. Even though it’s, it’s highly, highly enough heated that it creates light. Okay.
[01:04:53] Blue: But he didn’t have any idea how to work out which piece of metal to use or under what circumstances to use it. So he went through thousands of different variations, trying different things informed by heuristics of what he thought might work until he found the right combination. And then he, that was the light bulb. Okay. Whatever else you might say about this, that’s, this does sound like a lot of perspiration to me. And I honestly think I would find this very, very, very boring. Now Edison, maybe he did find it boring. Maybe he didn’t. Maybe he was interested enough that he decided to push through the boringness. Like I don’t know that I’m prepared to say, oh, you know, Edison didn’t find, didn’t find it boring. Maybe he did. Maybe he didn’t. He never says, I know that he did choose to refer to this process as 99 % perspiration. And that’s, that’s really all I know. And from there I kind of get the sense that part of the human creativity of coming up with a light bulb involved a lot of kind of boring perspiration. And I think that’s actually what Edison did me. Okay. Maybe I’m wrong, but I’m trying to read a mind of a man who’s long dead, but so is Deutsch. Right. And I actually don’t think that’s a bad reading of what Edison actually meant. Let’s use a different example, though. Let’s use Einstein. So famously, Einstein had a flash of inspiration. He had called it the happiest thought he ever had. Right. I can’t remember the exact quote. Okay. And it was this, it was this idea that you can’t tell the difference between actual weight.
[01:06:27] Blue: Like I weigh myself and I, you know, come out to be some number of pounds that I can’t tell the difference between that the force of gravity as a force. Or if I’m in an elevator and the elevators accelerating and it just shows me as weighing, you know, 170 pounds or whatever. Right. And he had this thought that those two were equivalent. And that led to that, that inspiration, that moment of realization that those two could be thought of as the same thing or that they would be indistinguishable in any case is what led to him eventually working out general relativity. Okay. Now, how long did it take him from that moment of inspiration to get to the actual theory? Well, it was eight years of very hard work, very, very hard work. In fact, one of the things that that, you know, some sociologists have argued, I don’t want to go too far with this, but I just thought it was interesting. Is that some people have argued that if I can’t remember the exact stuff on this, but like Einstein was like married at one point and then he got divorced and it’s kind of believed now he might have been autistic and he didn’t have very good interpersonal skills. One of the things that some people have argued, maybe wrongly was that part of what made Einstein Einstein was that at the time a good deal of the time he had nothing to do no social contact.
[01:07:51] Blue: And except to work on this problem that he found interesting and that if he had been like more successful socially, like a normal person would be, that he wouldn’t have had a sufficient time to work out his theory and that we may actually owe general relativity impart to the fact that Einstein was not the most socially capable guy. Now, I don’t know if any of that’s true or not, right? I just throw it out there as an interesting theory I came across once. But it does give a feel for the fact that Einstein wasn’t out there socializing and then occasionally having flashes of inspiration. General relativity required a ton of perspiration. Okay, like lots and lots, eight years of intense perspiration. Okay, if is so here would be my question if we’re going to try to make this perspiration inspiration divide, right and make it a hard criteria like this is is only the first flash. Is that the inspiration and that counts as knowledge creation and then the rest of the eight years doesn’t count as knowledge creation. Like obviously that’d be silly, but like you almost could read it that way if inspiration is creativity and perspiration isn’t creativity, then and creativity is knowledge creation and perspiration isn’t knowledge creation. Then like I wouldn’t I would really feel the need to ask that question and seriously say so only that initial thought that initial inspiration counts as knowledge creation and the eight years doesn’t count as knowledge creation. I mean that that would be bizarre right. Does this mean that everything else that came after time including the final theory is just not knowledge creation because it was all perspiration.
[01:09:36] Blue: Now obviously, I’m cheating when I say this and I know someone’s going to say this they’re going to say well I mean I’m sure he had other inspirations along the way. Okay, my point here is is that whatever inspiration is, it seems to include a lot of perspiration and therefore I’m just not sure that we can easily say perspiration is the opposite of inspiration. Whatever inspiration is it includes a bunch of perspiration. Okay, and then I have to ask another question. Okay, that initial flash that Einstein came up with that that that came unbidden to his mind. No work on his part right it just suddenly has this this flash where he goes. Oh, force of gravity could be thought of as a force or it could be thought of as a person in an elevator and the elevators accelerating right that initial flash. Was there no perspiration before that led to that inspiration. Well, I would say no right I mean just like with Alpha go. We can get rid of the perspiration as part of the gameplay, but there still had to be perspiration before, but that’s just as true for Einstein. So, Einstein spent considerable perspiration going to lectures and educating himself. Not knowing that by chance premium geometry would be key to turning his simple inspiration into something real. Okay, so but for all that perspiration prior to the initial flash. General relativity wouldn’t have happened. Okay, so you can’t make the easy divide between perspiration inspiration that that people often want to make. So human chess and go masters spend considerable perspiration reading books, watching famous games formulating theories about gameplay. All right.
[01:11:20] Blue: So I’m going to argue that there is no divide between inspiration perspiration that maybe you can have perspiration without inspiration but you can’t have inspiration without perspiration, and that the two are actually deeply deeply tied in a deeper way than people tend to use it.
[01:11:39] Red: Okay,
[01:11:40] Blue: now I’m going to I’m going to go on to show that actually do it seems to have understood this, and that as much as this going to be a little bit of a shock. He actually says some of these things in the beginning of infinity, and that it’s really just it’s sort of got lost in the process. Okay. So and then for that matter, how do we even know how much perspiration the general intelligence algorithm will use if I had the general intelligence algorithm here today. And I knew what it was, would it not use perspiration like like, what does that mean, right what would what would a non perspiration intelligence algorithm even look like right isn’t knowledge creation epistemologically supposed to be from trial and error isn’t trial and error a kind of perspiration. I hope you’re seeing where I’m going with this here. Okay. So what little we know about how the intelligent the general intelligence algorithm works boils down to spending lots of time perspiration, trying to solve a problem and failing until you really understand the problem well in fact this is what Popper said from objective knowledge page 181. So we learn to understand a problem by trying to solve it and by failing. And when we have failed 100 times, or maybe even become we may even become experts with respect to this particular problem. And this is how humans get their inspirations is through a whole lot of perspiration by trying to solve a problem and failing. Okay. Then what seems to happen is is the unconscious mind, the conscious mind may stop thinking about it, while the unconscious mind continues in the background.
[01:13:19] Blue: Often for months or weeks or months or maybe even years before that flash of inspiration comes. Now doesn’t it seem plausible that the unconscious mind is doing a great deal of perspiration but you just can’t sense it right you don’t have access to it. I’m going to argue that that’s at least a possibility and in fact it even fits well with our current best theories on the subject. What we kind of know is going on. Is that when you stop when you’re trying to solve a problem creatively solve a problem you don’t know how is that you keep trying things and you get you learn the problem better. And you do that very consciously. And then you kind of just put yourself into this, you know go off and you’re doing something else you go on a walk you watch TV show you sleep on it sleeping on it seems to be a big one because it gives the brain a chance to do quite a bit of perspiration while you’re asleep, right and it rewires itself. And you wake up the next day and suddenly this solution that your unconscious mind has come up with. Appears in your conscious mind. And it feels like this inspiration from God, right. And you just go. Oh my gosh that’s the answer. And sometimes it isn’t sometimes it isn’t like a lot of times, we often think of it as like it’s the answer but a lot of times you then go and test it and it doesn’t work out. Okay, but your unconscious mind has been doing something and it’s been doing it for possibly months, right.
[01:14:48] Blue: Well what’s it was what was it doing what we kind of know that it was trying out different neural circuits in some sort of search, right, isn’t that perspiration. So even in the case where you’re unconsciously aware of it. I suspect that once we understand the general intelligence algorithm we’re going to find that it is a 100 % perspiration based algorithm. Now maybe I’m wrong since we don’t know what it is someone could argue this with me, maybe legitimately, and they could say no it’s going to do something else that isn’t perspiration, to which I think my question to that person would be, then what is it. And of course they can’t answer the question, because we don’t know what the general intelligence algorithm is. And the only thing we know of today is perspiration. That’s how we do all trial and error today. Okay, so let me make my argument like this while I don’t know if in the the general intelligence algorithm when it when we understand what inspiration is I don’t know for sure it will be a kind of perspiration. I see no reason at all to rule it out as you being a perspiration based algorithm. Now it may be that it’s not perspiration conscious perspiration. In fact, not maybe that’s that’s what we’re going to find, right, is that there’s this conscious part to it where you try to work out the problem consciously, you fail, and then the subconscious works on it in the background, once it’s been kind of set up to do that. And I think that’s exactly what we’re going to find I think we’re going to find that it is a perspiration based algorithm. Okay, I could be wrong.
[01:16:44] Blue: So at least it would make sense to keep an open mind on this question and not consider it a hard. If it uses perspiration, it’s not creative.
[01:16:53] Red: Okay.
[01:16:55] Blue: I need to actually now defend what Dwight said in beginning of infinity. So that’s what we’re going to now go on to do. Okay, because I actually think what he says in the beginning of infinity is far more meaningful than simply a way to chop off algorithms as well that’s perspiration so it’s not creativity. Like that may be the main thing people use it for today, but like that is not the way he uses it in the book. Okay. So let me just say this though. What would a non perspiration but evolutionary method look like, you know, honestly that seems like a contradiction of terms to me wouldn’t all evolutionary algorithms be search algorithms and thus be perspiration algorithms doesn’t Darwinian evolution consist of a considerable itself consists of a considerable amount of perspiration billions of years of perspiration. Okay, so that’s why I just don’t buy the chop it off. With using the fact that it’s perspiration. Now, as it turns out, a close read of Dwight suggests that Dwight’s never had this problematic reading in mind, at least originally. He seems to be talking about more the subjective experience of inspiration versus perspiration and that for a human perspiration can be fun. And this seems to be his main concern. Now, we all know that Dwight spies into this thing called the fun criteria. So I actually think that this plays well to the fun criteria. Okay. And really wasn’t intended originally to be a way to slice off algorithms and say oh that’s perspiration so it’s not created. All right. Okay, so if we look at it more in this way, as Dwight is talking about human creativity does not need to be a lot of boring perspiration.
[01:18:55] Blue: I feel like his argument becomes way more cogent. Okay, now I kind of argue that that may not be the case for Edison, right. And I don’t know that for sure. And so I put some hesitancy here because my guess is and like you talk to any scientist, and they’ll tell you that a lot of their jobs not that fun. So maybe we could read Dwight here as saying it could be fun or not necessarily that it is fun, but that it could be fun. And this might be the best way to read Dwight here. Because I’m fairly certain that the way we do science today that many scientists do not enjoy every aspect of their job and that a lot of it’s kind of meaning boring, mindless work. However, having said that, let’s look at what Dwight actually says because it really is kind of fascinating what he actually says. Okay, so Dwight admits that human knowledge creation does involve perspiration. This is the main thing that this is why I did such a build up to this point, right is because this whole argument from his book has become. Oh, that algorithm is perspiration so it can’t be creative and perspiration has come to mean the opposite of inspiration. I feel like that’s a just false reading of what Dwight actually says. So let’s take page 342. And here’s an actual quote from beginning infinity in complex decisions the creative phase is often followed by a mechanical perspiration phase in which which one ties down details of the explanation that they that are not yet hard to vary.
[01:20:36] Blue: Yet can be made so by non creative means case and remember creativity in Dwight’s mind explicitly means the inspiration right so here he’s he’s basically saying and I want to make sure this is clear. He’s saying that a good fear a good explanation that is to say a hard to vary explanation is not merely a function of the creative phase, but requires a perspiration phase, like you have to have that as part of the kind of human ideas, you do not have just inspiration. Okay, what he’s really trying to say here is he’s he’s he’s clearly tying perspiration and mechanical, let me read that part again. The creative phase is often followed by a mechanical perspiration phase. This is why people equate perspiration and mechanical together because he did in fact, equate them together in this quote. Okay. But he’s admitting that human knowledge is a combination of inspiration and perspiration in fact he’s insisting that’s the case in this quote. Okay, because a good explanation is hard to vary and inspiration doesn’t produce hard to vary explanations. Let me say that again if you don’t believe it, I’ll read the part again where he says that in complex decisions the creative phase is often followed by a mechanical perspiration phase in which one ties down details of the explanation that are not yet hard to vary, but can be made so by non creative means or automated means. Okay, so do it is in fact saying human ideas human creativity is this combination of inspiration and perspiration human good explanations are a combination of inspiration and perspiration.
[01:22:19] Blue: His main concern seems to be not to claim perspiration doesn’t create knowledge as James clearly believed, but that human should do the fun inspiration work and leave the boring stuff to the machines. Here’s a quote from page 36 or 37 of beginning affinity, but more profoundly I expect that Edison was misrepresenting his own experience a trial that fails is still fun. A repetitive experiment is not repetitive. If one is thinking about the ideas that it is testing and the reality that it is investigating. If Edison or those graduate students he gives examples of graduate students or any scientific researcher engaged in the perspiration phase of discovery has really been doing it mindlessly. They would have been missing most of the fun, which is also what largely powers the 1 % inspiration. Notice that doge here is explicitly saying that scientific research does have a perspiration phase. And that he’s really not arguing you can’t you can just get out of it he’s advocating for getting out of it. If you can, but he’s arguing that even if you don’t get out of it even if you need to do it. That if you do it right it can still be fun. And then he goes on and he says which is also what largely powers the 1 % inspiration it’s the perspiration phase and not doing it mindlessly that largely powers the 1 % inspiration notice how he ties those two together they are not the distinct things that people now make them out to be. Okay, and again I do want to emphasize I do have my doubts that I think Edison when he said it was 90 % inspiration that he wasn’t always having fun.
[01:24:05] Blue: But I do think that doge is right that it can be made more fun, and I also think that we can learn to automate it like imagine that Edison had lived in a modern era where he could say you know what we’re going to try every we’re going to try every combination of of environment how much vacuum there is and every single kind of filament, and we’re going to automate that so I don’t have to do it, and then we’re going to see what the results are. I mean like why not right we absolutely can use our creativity or inspiration to take that necessary perspiration phase, and we can make it more fun for ourselves or we can automate it entirely and let some other process take care of it. Okay, and this seems to be how he actually use the perspiration inspiration divide in the beginning of infinity. He points out that even the perspiration will be fun to a human he gives examples of graduate students working on a galaxy catalog galaxy catalog which to me sounds rather boring, and how we eventually automate the process via an algorithm. And here’s the quote page 36, perhaps those galaxy cataloging computer programs were written by those same graduate students distilling what they had learned into reproducible algorithms, which means that they must have learned something while performing a task that a computer performs without learning anything. Page 36. Okay. Now let me dwell on this a bit though because this does show a crack in dutch’s armor that I think is very important. dutch’s example has not aged well.
[01:25:35] Blue: Why, because today we probably wouldn’t have the graduate students take what they’ve learned and have distilled through the process of trying to catalog galaxies and write an algorithm, we would probably let machine learning write the algorithm for us. Okay, and that would almost surely work better than having a graduate student write it because it would come up with properties that we would never think of. And we just know just from experience that machine learning will easily outproduce a human in writing the sort of algorithm something like a galaxy catalog the algorithm. Now that wouldn’t have been so true back when beginning affinity was written but today it’s certainly true. Okay because machine learning is kind of long way. There would be no need for the distilled learning of the graduate students to write the algorithm anymore because machine learning would do that work. Now dutch insists that they must have learned or created knowledge to be able to do this, but he wants to rule out calling machine learning algorithm that does the same thing knowledge. Well that’s a problem, right because if the fact that the machine learned that the graduate students doing catalog galaxy cataloging somehow distilled something and learn something. And that’s the difference between a human doing the computer doing it now that computers are also learning from doing it and actually distilling out knowledge to suddenly decide what it only counts as knowledge of a human learns not if a computer learns well that’s a problem. Okay. All right so in summary, you can’t rule out the immune system or other examples as not creating knowledge based on the fact that they are perspiration algorithms, you cannot do it it is impossible.
[01:27:23] Blue: We don’t even know what inspiration or creativity means other than it’s the thing humans do that machines and animals don’t somehow know how to do related to explanation. And then I think it likely involves considerable perspiration. But that doesn’t mean dutch isn’t making a fair point. I think a correct read of him is that human creativity has something special about it that isn’t merely conscious perspiration and can even be outright fun for humans with the right mindset. Okay, now, however, there are some interesting things going on here that deserve further attention. Recall that James insisted that I needed to prove to him that there was anything at all worth theorizing about here. It goes a lot, a lot further than merely saying he’s going a lot further than merely saying the immune system doesn’t create knowledge which I was agreeing with him I was saying the immune system does not create knowledge the way you understand knowledge. Right. That under the Deutsch definition of knowledge I was fine with that. Okay, I was trying to say but it is adapted information and it’s interesting. And he’s trying to argue that not merely that it’s not knowledge because I’m agreeing with that. He’s trying to say the immune system is doing something thoroughly uninteresting. It’s not even worth having its own theory about. Okay. So the immune system at a minimum, as we’ve seen is an evolutionary epistemological process of variation selection or natural selection. I personally find that quite interesting. But James is a quick is basically equivalent to denying the importance of evolutionary epistemology poppers and Campbell’s theory of how epistemology is connected across everything via variation and selection.
[01:29:06] Blue: So there is something important going on here that when James tries to say that the whole perspiration inspiration divide has actually caused him to go not merely to claim immune system isn’t knowledge but to claim that essentially evolutionary epistemology is not that interesting. And I don’t think that’s right. Like I think that’s absolutely the opposite of right. In fact, I think it’s the opposite of how you should go about even trying to figure out the general general intelligence algorithm is the fact that defenders of the two sources theory don’t really seem to mind making up new criteria on the fly to defend the theory is something else I want to kind of pay a little bit more attention to also. So I’m okay. So I’m now done. That was that was the end of this part of the discussion. So thank you, Peter.
[01:29:56] Red: Okay. Well, thank you, Bruce. I think that was an above average episode and I you kept me interested for for almost two hours there. So thank you.
[01:30:07] Blue: You’re welcome. All right. Bye 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 Deutch’s philosophy as well as other interesting subjects. If you’re enjoying this podcast, please give us a five star rating on Apple podcasts. This can usually be done right inside your podcast player, or you can Google the theory of anything podcast Apple or something like that. Some players have their own rating system and giving us a five star rating on any rating system would be helpful. If you enjoy a particular episode, please consider tweeting about us or linking to us on Facebook or other social media to help get the word out. If you are interested in financially supporting the podcast, we have two ways to do that. The first is via our podcast host site anchor. Just go to anchor.fm 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 four strands.org. There is a donation button there that uses PayPal. Thank you.
Links to this episode: Spotify / Apple Podcasts
Generated with AI using PodcastTranscriptor. Unofficial AI-generated transcripts. These may contain mistakes; please verify against the actual podcast.