Episode Transcript
[00:00:00] Speaker A: Welcome to ID the Future. I'm Andrew McDermott. Today's episode comes to us from our sister podcast, Mind Matters News, a production of the Discovery Institute's Walter Bradley center for Natural and Artificial Intelligence.
You can learn more about the show and access other episodes at Mind Matters. AI.
Okay, everybody, welcome back to the podcast. This is your host, Pat Flynn, teaming up again with the awesome team at Mind Matters to discuss all things philosophy of mind. Today we are joined by Dr. Winston Ewart to discuss his very good contribution to the Minding the Brain volume called the Human Mind Sophisticated Algorithm and its implications. Lots of great stuff to discuss here. I'm sure we'll nerd out on some cool topics related to the nature of the human person and human cognition and all that good stuff. But Dr. Ewart, this is the first time that we have had the opportunity to speak. I'm looking forward to it.
If you wouldn't mind, I would love to just hear a little bit about your background. I'm sure the listeners would as well. So who are you? What do you do? How did you get into all this philosophy of mind business?
[00:01:10] Speaker B: Well, basically I am initially a computer nerd. So I grew up being very interested in computers, studied computer science as part of that. I eventually became interested in the subject of intelligent design by way of computerized simulations of evolution.
And this led me to getting a graduate degree at Baylor University where I worked with Robert Marx. And so that led me to be involved in intelligent design. And then I was asked if I would be interested in contributing a chapter to the Mind Matters book. And so that sort of led to me developing the argument I put forward in the chapter. And that's how I got into sort of the philosophy of mind and that sort of thing.
[00:01:54] Speaker A: Yeah, fascinating. Just not to, you know, veer too quickly off of topic, but what, what was your experience like when you're working with computer simulations of evolutionary theory? Tell us a little bit about that. The sort of work you were engaged in or research you were engaged in. And what surprised you about that work or research?
[00:02:14] Speaker B: What surprised me? Yeah. So I came at to it from a perspective of someone who was already on the intelligent design side of things, and therefore I was skeptical of the claims being made about these computer simulations.
And so I sort of was sort of taking them apart and seeing how they worked. And so as someone who was a computer nerd and did programming and that sort of thing, I was, well, sort of suited to sort of take it apart and look at the details of these things and find, well, they're never quite what they're presented as because they're not really simulations of a naturalistic process. They're simulations of a very intelligently guided process that has a lot of, you know, intelligent agency involved in the setup of the system to make sure it does what they want it to do. And then they take it and present it as, hey, look, evolution is working. And it's like, yeah, that's not really what's happening there.
[00:03:05] Speaker A: Yeah, that's, that's an interesting point and something that, you know, a lot of people might not fully appreciate. So when you say that these are, there's a lot of clearly intelligent input into a lot of these simulations and models. Can you just give us a few examples to help people understand what you mean by that?
[00:03:23] Speaker B: Yeah, one example, there's a very famous example from Richard Dawkins book A Blind Watchmaker where he evolves the phrase methinks it is like a weasel.
And that might be impressive except for the fact that what he did is just say, well, phrases that are more similar to me think it's like weasel no matter how dissimilar they are. But if they're a little bit more similar to that, you know, natural selection would favor that and then very quickly converges on me thinks it is like a weasel. But you know, that required that sort of idea, the target phrase to be pre programmed into that simulation and to define all sense of selection based on similarity to that, which very much, you know, not really what an evolution would do realistically. But you can look at other cases like there's a paper I wrote on a Steiner trees and someone had written this simulation was very loudly proclaiming that it demonstrated the power of evolutionary processes. But then when I would look through his code, I'd find things like they had overridden how many points there were in their Steiner tree to away from randomness to something that would better fit what they were trying to accomplish and they would restrict the locations of points and have these, all these extra rules in there which serve no purpose but to aid the evolutionary process towards the goal he was trying to reach with it.
[00:04:47] Speaker A: Yes. Okay, so that's, that's, that's helpful and that's certainly interesting, I'm sure to, well, everybody in general, but especially to listeners of this podcast. So is it, is it fair to say that a lot of these, what's going on with these simulations involves just specific fine tuning of certain things to generate particular outcomes that one is, one is aiming at?
[00:05:07] Speaker B: Yeah, so certainly that's a very common feature, you find these specific, I call them teleological fine tuning in these.
That's very, very common when I look through them.
[00:05:18] Speaker A: Yeah. Okay, interesting. Well, that's obviously something that could be an entirely separate conversation, but thank you for, for entertaining me there for a moment. Now, turning back towards philosophy of mine, of course, these all things, all these things are generally related.
You've got this, this article, this very called the Human Mind Sophisticated Algorithm and its Implications. So you gave us a little bit of background on how you got involved in this project. But yeah, tell us a little bit more about the general overview of this article.
What are you setting out to accomplish? What inspired it? What sort of things do you think we need to know by way of background before we start actually diving into some of the details here?
[00:05:58] Speaker B: Yeah, so the basic, the basic idea that I took here was to say, well, let's consider the human mind as a computer.
Now, I think there's more to it than that, I think particularly explaining consciousness. But I think when you look at human intellectual ability, our ability to do math or reason through things, we could model that as a computer running a really sophisticated algorithm.
And then we look at, well, what would that algorithm, if we think about that algorithmically, we can apply what we know as theoretical computer scientists to what that algorithm and what that would actually tell us about the implications of humans running on a very sophisticated algorithm.
[00:06:44] Speaker A: Okay, good. So let's, let's, let's start at the sort of the kitty end of the pool and help people just become familiar with some more of the background and a few of the terms and we can dive into some of the technical weeds of your argument, which is very fascinating, your overall paper.
So I guess the first thing to consider is that there is a certain model out there, computational theory of mind. Right.
That is often proposed to explain human cognitional ability. And that's, it's certainly an interesting model. It's a very, very popular model, at least among certain types of thinkers. And it seems to have become of increasing interest lately, especially with, with AI and all that sort of stuff, which we can talk more, more about as we move along here.
But early on in your article, you make the very useful division when talking about the human person, the human mind, between different aspects of the mind, between phenomenal consciousness and cognitional ability. And I think that's probably a good place to start. Right. Because these are different features of the world and the human person that may or may not require different explanations. So I think it's important to at least break these apart. Conceptually, but before we start to see if they succumb to the same sort of analysis. So, so help us there. How should we, how should we start to sort of divide the, the problem up, if you will, and then how it relates to computational theory of mind. Yeah, right.
[00:08:11] Speaker B: So the two aspects in particular I look at, you've got the phenomenal consciousness you mentioned, and then sort of the problem solving cognition.
So when you're looking at phenomenal consciousness, that's sort of like, you know, what your, your first, your first person experience, like, you know, what it is to feel cold or delight at solving a problem or happiness or, you know, the experience of seeing things with your eyes. That's sort of your, you, you actually experience these things and you have that experience.
[00:08:43] Speaker A: Yeah, the, the what it is likeness to taste chocolate or, or hero.
[00:08:47] Speaker B: Yes, that sort of thing.
[00:08:48] Speaker A: Symphony or something like that. Right?
[00:08:49] Speaker B: Yeah, right. And that's, you know, obviously something we have firsthand experience that we do that.
And the other thing you have, the problem solving cognition. And that's where you sort of, sort of anything you can reduce to answering a question of some sort. So you like, what is two plus two?
Or is this a valid logical argument? Or is it likely, given, you know, the fact that the sun has risen every day of my life, that it's going to rise again tomorrow? Those sorts of things, they're intellectual things. You see, that's quite different than experiencing something because you're sort of coming to a conclusion based on your data.
And that's what I think. There's sort of the problem solving cognition side of it.
[00:09:32] Speaker A: Okay, good, good. Let's, let's take a look at the first one, which you don't dedicate all that much space to because you think the, the question sometimes isn't particularly well defined in certain respects. So how should we think about human phenomenal consciousness? Is this something that can succumb to a sort of materialistic or computational theory of mind? What are your thoughts on this?
[00:09:53] Speaker B: Yeah, well, I think that if you think about consciousness and if you're trying to reduce it to computation, to me that's like trying to reduce it to math equations.
So I'm pretty sure that the quadratic formula isn't conscious.
[00:10:11] Speaker A: I'm with you.
[00:10:12] Speaker B: I actually have run across some people who actually dispute that and I don't really understand their mindset, but I'm sort of starting. I think for most people, common sense says math formula isn't conscious and computation at the end of the day is just basically math formula. And So I think it doesn't make any sense to say that the abstract notion of computation could be conscious.
Now I think you can then say, well, where does consciousness come from? And I think you could say, well, whenever you build a computing machine in the real world, there's consciousness generated as a side effect or something like that. That's somewhat different than what I'm saying here. Just that's not like the abstract computation itself generates consciousness, but somehow the rules of the physical world cause consciousness as
[00:11:03] Speaker A: a side effect that somehow emerges once a certain degree of complexity is attained. Or something like that. Something like that, yeah. Obviously, I think we're both probably skeptical of that, but it's important, as you say, that these are. These are different sorts of proposals. Right.
And yeah, I mean, that seems right. That seems definitely right. That. That would be almost a serious, you know, profound confusion of categories to think that phenomenal consciousness is just somehow equational.
I'm with you there.
So could it be physical? That's a different question as well, isn't it, Winston? Right, right.
[00:11:39] Speaker B: So that could be a different question. If there could be physical, maybe there are consciousness particles that exist that are responsible for generating it and we just haven't found them yet. Or maybe there's sort of a non physical world that's interacting with ours. Those are, you know, possible ways of doing it.
[00:11:56] Speaker A: Good, good, good. Okay, so is there anything else that, for our purposes, for, you know, your article that needs to be said about phenomenal consciousness?
[00:12:03] Speaker B: I don't think there's. There's not really anything further. I basically just, you know, distinguish it in the paper. Just say, I'm not talking about this. Focusing on the cognition and problem solving
[00:12:15] Speaker A: and the problem solving ability. Great. Okay. Good.
[00:12:18] Speaker B: All right.
[00:12:18] Speaker A: Hey, here's a. Here's a question that maybe we should have, you know, gotten some clarity on right at the beginning. What's an algorithm? What do we. What do we mean by an algorithm?
[00:12:26] Speaker B: So an algorithm is essentially a step by step procedure that can be followed. Followed. It doesn't require any creativity or choice or sort of intelligent choices.
[00:12:37] Speaker A: You can just.
[00:12:37] Speaker B: You just follow the procedure. So when you were taught math as a child, you were probably taught to follow. Here's the certain procedure you follow in order to do long division or multiplication or whatever it is. Right? You don't get to make any choices during that decision. You just have to follow those particular rules. Rules. And that's what an algorithm is. It's just the set of rules that you're following that requires no creativity. You just Follow the defined procedure.
[00:13:06] Speaker A: You know, you made this very interesting and important historical note in your paper that computers were originally humans. That's true, yeah. So give that little detail for people. I think you might find that fascinating. Right.
[00:13:20] Speaker B: Computers historically were originally people who were following these procedures.
So because we hadn't figured out how to build electronic computers yet, they would actually have large groups of people and they would give them these calculations to follow, and they would sort of follow those procedures. And that's somewhat where they were developing the theory of algorithms was they were, they would have these people and they, you know, didn't want a bunch of math geniuses that they had to hire to do this work, but they would fire people who just had the ability to do your basic math calculation calculations and give them, here's the procedures you follow and you go and you do that. And so that was even used, I think, in some of the early space Apollo missions. They actually even were still using banks of human computers to do some of their calculations.
[00:14:07] Speaker A: Yeah, that's cool. It's fascinating to think about. Right. And certainly we've advanced a little bit since then, but I mean, fundamentally. Right. The mechanics are still the same, right?
[00:14:15] Speaker B: Yeah, Fundamentally it's still the same. All we just figured out is, oh, for these really straightforward mechanical things, we can build a machine to do that.
[00:14:22] Speaker A: Yeah, very cool. Super fascinating. Okay, all right, so let's move forward. I mean, so your, your general thesis, right, seems to be that you're, you're quite sympathetic to the idea that the, that the human mind, with respect to our, the cognitional side, the problem solving side, is a very sophisticated sort of algorithm. But this isn't particularly good news for like say, reductive materialists and atheists and neo Darwinians. Right?
[00:14:50] Speaker B: Yes. I think once you actually think through the implications which we'll get into, it's actually very problematic for them.
[00:14:56] Speaker A: Yeah, And I wanted to. To pick that out because I think a number of people, people who think about this are resistant to the idea of saying that, that this aspect of us is algorithmic. So you sort of part ways with.
[00:15:09] Speaker B: Yes, I part ways probably with many of the other people in that volume.
[00:15:13] Speaker A: Yes, I just wanted to hang a hang of. Hang a lantern and find. But that's good. It's good to get this diversity of thought. But ultimately you guys are sort of winding up in the, in the same position. Right. You know, obviously if people think that this is non algorithmic, then this kind of throws different models that are very much favored by, you know, reductivist materialist type of thinkers. Right.
Kind of out the window. Right. But you're. You're taking a different route, a different strategy here. You're saying, no, we can. We can accept. And there are some reasons to accept that, that. That it is algorithmic, but the implications of that are still rather negative and dire for those particular worldviews. Is that a fair summary?
[00:15:52] Speaker B: I think that's a fair summary.
[00:15:54] Speaker A: Okay, good. All right, let's start making your case. Winston, where do we begin?
[00:15:58] Speaker B: So I think that what we begin with is discussing what I called the uber task.
So, computer scientists, we study something called the halting problem. And the basic idea is, if I give you an algorithm and I ask you, if you were to follow this procedure, would you ever actually finish, or would you keep following the procedure forever and ever?
So I think I have an example in the chapter where it's like, you know, step one, dance a jig, and step two, go back to step one. And if you were to follow that procedure, you would never stop dancing your jig because you'd keep going back to step one and just repeat itself forever and ever. And so that's, of course, a very trivial example. And you could look at that very quickly and say, well, this doesn't halt. It never finishes.
But if you had a, you know, step one, dance a jig, and step two is repeat the first step a thousand times, you can look at it and say, well, it would take me a long time, but I would eventually finish.
[00:16:56] Speaker A: But it would halt, right?
[00:16:58] Speaker B: It would eventually halt. And so we think of there sort of being as two possible kinds of algorithms. There's those algorithms which halt and those which don't.
[00:17:06] Speaker A: Got it.
[00:17:07] Speaker B: And so we call differentiating those the halting problem. And it may seem like that's a kind of esoteric and boring question, but the key point is that by selecting different algorithms, we can actually use that to define pretty much any kind of cognitive. Cognitive problem can be described as an instance of the halting problem problem. So we can take it and say, here's an algorithm for which the halting problem is equivalent to that original question.
[00:17:37] Speaker A: Okay, good.
Do you have an example or way that you like to illustrate this or get this across to people?
[00:17:44] Speaker B: Yeah. So one of the examples in the chapter, I think I talk about the Goldbach conjecture. And so the Goldbach Conjecture says that every even number can be expressed as the sum of two primes.
If you have 88 is 17 plus 71, that sort of thing. And it turns out for as many numbers as We've checked for all the even numbers. You can always find two problems, add them together, and get to that number. We've never actually developed a mathematical proof for this. It just empirically, it looks to be true for everything we've tried.
We don't know, maybe there's a counterexample that's really large.
And so you could ask the question, is the Goldbach Conjecture true? Which is that this is true for all numbers.
Now, you could write an algorithm out that says, okay, well, let's look. That says, look for a counterexample. So start with 2, check if the Goldbach conjecture is true, go to 4 as the next even number, check if the Goldbach conjecture is true. And then you could just repeat that over and over again, going through all the numbers.
And so if the Goldbach conjecture is true, that will never halt, it will never finish, because it's going to keep looking for a counterexample.
[00:18:51] Speaker A: Sure.
[00:18:51] Speaker B: But if the Gobacks prediction is false, then it's eventually going to halt. And so if you could quickly and easily solve the halting problem for any algorithm, you could determine whether that's true or not just by sort of writing out that algorithm and doing your halting check on it.
[00:19:07] Speaker A: Okay.
[00:19:08] Speaker B: And so that's sort of an example. And it turns out there's number. There's lots of different issues in number theory that you could check this way by looking for a counterexample. You could also take, for example, if you want to say, like, does this, does this conclusion logically follow from these premises?
And you could write an algorithm that says, well, let's systematically try every possible proof to see if we can find a proof that makes this work. And then if you can't find a proof, well, that's never going to halt because it's never going to find a valid proof. But if there is a proof, it will halt. And so again, you can actually, if you could solve the halting problem, you could quickly, you could check, does this halt? Will I find a solution? And then that would tell you whether or not there's a proof for whatever it is that you're looking at.
[00:19:57] Speaker A: Okay, good. All right, so that is, that's a clear illustration of the problem. So what's the connection here for human cognition and computational theories of mind and trying to answer the question of whether our it is algorithmic or not.
[00:20:11] Speaker B: Right, right. So the case that I'm making there is that for any sort of the cognitive tasks that we're interested in, you could express it as a version of the halting problem.
And so we can think of it.
Humans are able to solve some of those problems, not all of them. Right. We are not able to prove whether or not the Goldbach conjecture is true.
[00:20:31] Speaker A: Right.
[00:20:31] Speaker B: But there are many other ones that we've pretty well convinced ourselves are true because we've constructed proofs for it or whatever. And so the one way of thinking about humans, then, is that we're partial halting detectors.
We can determine whether some subset of programs halt or algorithms halt, but we can't determine another subset.
[00:20:50] Speaker A: Yeah, I see.
[00:20:52] Speaker B: Now, are. The set we can determine is pretty large and sophisticated, which is why I say we must have a very sophisticated algorithm, but it is still limited.
And that's sort of the intuition that, to me, makes me sympathetic to the human mind being algorithmic, because we know the same thing is true if we were to build an algorithm to solve the halting problem.
So it's one of the sort of foundational proofs of theoretical computer science is there's no algorithm that can solve all halting problems.
[00:21:22] Speaker A: Okay, all right, I see. So that connection is becoming clear to me and hopefully to the audience is there, obviously, as you said, a number of people, even in this. In this volume, resist the idea that all human cognitional ability is algorithmic. So in your consideration of these dear thinkers who take a different route than you, Winston, what do you think are some of the most perhaps problematic considerations for your particular theory?
And how do you. How do you think about those?
[00:21:57] Speaker B: Yeah, so I think the what, in my view, when a lot of people are sort of looking at this, what they do is they're implicitly in their mind, considering human abilities versus a relatively simple algorithm.
So they're like, well, humans are way, way better than our best algorithm at finding mathematical proofs.
[00:22:21] Speaker A: Sure. So it's a difference in degree. Not kind necessarily.
[00:22:25] Speaker B: Right. But to me, it's like, well, yes, but the question isn't really, are humans smarter than our current algorithms? And the answer is absolutely yes.
But are humans smarter than any algorithm that could exist?
[00:22:38] Speaker A: And you're saying not necessarily, right?
[00:22:40] Speaker B: Not necessarily.
And in particular, we can know from. If we think of cognition as solving some subset of the halting problem, we know you can always take any algorithm capable of solving some halting problems and expand it to detect more halting problems.
[00:22:57] Speaker A: Yeah.
[00:22:58] Speaker B: By simply adding on to it, we say, well, here's a pattern of things that we know. Halts detect that pattern as well. Or here's a pattern of things that doesn't halt, detect that Pattern.
[00:23:08] Speaker A: Yeah. Okay, got it. So I'm thinking of our, of our dear friend Bob Marx. Right. And he does want to say that there is obviously non algorithmic aspect about us, but from what I remember, and Sorry, apologies to Dr. Marks for not remembering everything he said on this topic. I know he focuses a lot on phenomenal consciousness. Right.
[00:23:28] Speaker B: I think that's definitely where the best case is.
[00:23:30] Speaker A: And so you, you two are in, in agreement there for to be sure. Right. And again, it's just really, just thinking about this cognitional ability where you're saying, yeah, no, not, not. Don't be so quick. Right. To dismiss the idea that, that, you know, this, that this really could be algorithmic through and through, however impressive, however sophisticated. Right. All right, all right. So anything else you want to say about that or the halting problem or just in general support of your sympathy towards human cognitional ability being a sophisticated algorithm before we move on to some of the implications?
[00:24:08] Speaker B: Yeah, I guess the only thing I, I guess what I would say is for me, if you look at humans limitations on certain things, it kind of makes sense in my mind that they seem algorithmic. So like, one of the ones I talk about in the paper is someone else looks at busy beaver numbers, which is this sort of game programmers play to say, you know, given a very short program, how big of a number can I describe?
And so you look at it, and while humans, they've managed to figure out this largest number that you can describe in some very short programs, but then we've actually kind of fizzled out in terms of once the program size gets large enough, we lose the ability to figure out what the smallest one is because there's so many programs to look through.
And to me, well, that looks very much like what you'd expect if it's. We're reaching the limits of our algorithm in terms of what we're able to see there.
[00:25:09] Speaker A: In part one, we have discussed his position on human cognitional ability and it essentially being just a very, very sophisticated algorithm. Now at first you might think, okay, well, isn't this what a lot of, you know, atheists, materialists, and people who hold to computational theory of mind say?
What does that mean? Is, does that mean that that worldview is true and that there's nothing of any, like, huge, you know, grand significance to us? And we're about to say, not so fast, not so fast. Because in part one, we differentiated between different aspects of the human mind, the phenomenal consciousness aspect, the what it is like aspect of what it is like to Taste chocolate or to hear a symphony. And Winston, in agreement with many of the other contributors to Minding the Brain, said, yeah, that's, that's definitely not just a formula or an algorithm, but when it comes to our cognitional ability, our problem solving ability, well, there's more of a case to be made there.
And indeed, he made that case. So please listen to part one, if you haven't already.
But the implications are not so great for many people who hold to a sort of reductive, materialist worldview. And we're going to explore that right here and right now. So, Winston, thank you so much for taking the time to be here.
[00:26:25] Speaker B: Yeah, it's good to be here.
[00:26:26] Speaker A: Okay, so where should we begin here? Because there's a few things I want to explore, and obviously they all concern the implications of a lot of what we discussed in part one. I want to explore AI, but I also want to explore why you think this understanding of the human mind is actually pretty bad news, you know, for people who, who think that you can sort of reduce the human person to atoms and the laws that used to combine them and Darwinian forces and all that sort of stuff. Right, right. So, yeah, help us, help us find a place to start, Winston.
[00:26:58] Speaker B: So I think the, the key thing is, sort of linchpin of my argument is the idea that a, A, an algorithm could only construct an algorithm less sophisticated than itself.
[00:27:11] Speaker A: Ah, yes. Okay. Yeah. Help us understand what, what is meant by that.
[00:27:14] Speaker B: So we've got this, the idea of the halting problem. So my argument, you know, all kinds of cognition can be reduced to examples of the halting problem, including the problem of constructing a halting problem, a halting detector. And so basically it means if you wanted to build an artificial intelligence, that's equivalent to building a program that's able to determine whether certain algorithms halt.
And that itself, you could, you could build a program that verifies the correctness of a halting detector by sort of simulating, checking through all of its programs, trying to find one that it gets wrong.
And if you can't find a program it gets wrong, then, you know, just like when we were searching for a counter example to Goldbach's conjecture, we check to see whether there's a counterexample to Goldbach's conjecture. We could also check for a counter example to sort of a halting detectors claims.
Now, the key thing is that the fundamental proof we have in computer science that you cannot build an algorithm which solves all halting problems basically shows actually that no halting detection algorithm can correctly detect whether itself will halt.
So basically what I argue from that is, well, that basically a halting detector can only detect whether a less sophisticated halting detector is correct, because if it could detect itself or a more powerful halting detector, then that runs into a contradiction in the standard proof for the impossibility of halting detection.
[00:28:56] Speaker A: Okay, really fascinating there. So my first immediate question is, you know, assuming this is well known among the relevant experts, why do so many people seem so optimistic about singularity? If you want to explain what that is for us and all, you know, this, you know, essentially Terminator type scenarios and all that fun stuff. Right, yeah.
[00:29:15] Speaker B: So the singularity is the idea that there, that eventually you were going to have an artificial intelligence smart enough to create an even smarter artificial intelligence, and then that will create an even smarter artificial intelligence and there'll be this sort of breakaway explosion of artificial intelligences that far surpass human ability and leave us far in the dust and change the fabric of the universe forever. Yeah, that's the basic idea of the singularity. And an implication of my argument is that doesn't happen.
[00:29:43] Speaker A: It can't happen.
[00:29:44] Speaker B: Says you're can't happen because an algorithm can only drive.
Now you ask if this is well known. So what's well known is the proof that you cannot develop an algorithm which correctly classifies the halting status of all algorithms.
That's well known. The idea here that the implication is that you can't.
That an algorithm can only construct a less sophisticated algorithm that is basically original to me.
So that's not a well known conclusion you can get in the chapter I talk about drawing a blank on the guy's name.
[00:30:21] Speaker A: Was it Penrose?
[00:30:22] Speaker B: Yes, Penrose.
[00:30:23] Speaker A: There we go. Always happy to lend a hand.
[00:30:27] Speaker B: Penrose has actually a somewhat similar ish argument that he makes, but he draws a different conclusion than me from it. So he realizes the problem and says, well, his solution is that humans, the human mind, must not be computational.
[00:30:44] Speaker A: Sure, yeah.
[00:30:44] Speaker B: And so, you know, I take a different fork there on his argument and say, well, it is computational. The problem actually is. Another part of Penris's argument was specifically that, well, the only way this could work is if humans had a really sophisticated algorithm. And where would that come from?
[00:31:00] Speaker A: Yes, right. And he's not willing to hold, of course, to divine a divine origin or divine intervention.
[00:31:06] Speaker B: So a divine origin, of course, solves a different problem, solves the problem a different way. And so it's similar there, but in general there's not sort of a Widespread awareness, I think, of this particular issue.
[00:31:16] Speaker A: Okay, so how would you draw it out for people who are familiar with the relevant proof and you want to try to convince them of taking, I guess, either your or Penrose's option. Right.
What's the simplest way to make that connection?
[00:31:30] Speaker B: So the basic proof for the impossibility is they say, well, imagine if you take your halting detector and you sort of add an addendum to it that says, okay, if I would halt, if the program itself would halt, then loop forever.
And if I would loop forever, then halt.
So it's a contrarian program that says, whatever I think I should do, then I'll do the opposite.
And so that, of course, leads to contradiction because either it got it wrong about what it. What it would do, in which case it's not correctly detecting the halting status, or it can't do the opposite, which shouldn't. It's a very trivial algorithmic step to sort of do the opposite of what your prediction claimed it would.
And so that only. And that conclusion is easily avoided if you say, well, I can't run on myself. And that's really the only way to avoid it. And so you can construct an algorithm which in principle could detect, you know, everything else except for itself. And that would be consistent with the proof.
[00:32:40] Speaker A: Ah, yes. Okay. I see. All right, so is this strong enough to say that you really can, at the end of the day, have.
And I know you talk about this in your paper, gradually increasing intelligence.
[00:32:52] Speaker B: Right. So I think it does rule out this sort of gradually increasing intelligence, because if in order to build a more sophisticated intelligence, you somehow have to know that what you're adding onto it is correct.
[00:33:05] Speaker A: Yeah. Right.
[00:33:06] Speaker B: And how do you know that if you can't evaluate the correctness of that already, and if you could evaluate the correctness of it already, well, then you're already that intelligent.
[00:33:15] Speaker A: I see. So you're. I mean, ultimately your. Your argument, sort of a master argument against the possibility of the singularity, but also a very reductionist evolution, Neo, you know, Darwinian evolutionary explanation of human cognitional ability. Is that right?
[00:33:35] Speaker B: Yes.
[00:33:35] Speaker A: Yeah. So the idea there is, while God could, of course, create beings of increasingly gradual intelligence, you don't have, like, ultimately gradually increasing intelligence because you have the ultimate intelligence at the bottom of things. Is that right?
[00:33:48] Speaker B: Yeah, that's the basic idea.
[00:33:50] Speaker A: Yeah. Okay, that's a really fascinating argument. What I would like to maybe just explore towards the end here are just as, you know, as you usually do with stuff like that is provocative. Right. This is interesting. Certain objections. What objections have you found so far to your, to your work along these lines and apparently you still hold the same position. So why do you think they fail?
[00:34:14] Speaker B: So I haven't had a lot of people take notice of the argument I laid out here yet.
[00:34:19] Speaker A: Oh, they will now. So get ready.
[00:34:21] Speaker B: Yeah, well, I gotta be ready for that.
[00:34:23] Speaker A: Well, I like one thing, I just want to point out that I like. Well, maybe we could even phrase it this way if, you know, and certainly happy to hear any informal injections you may have heard too. But what I like is like it make your, your argument makes a prediction. Right. If tomorrow, like the terminator systems come online and we find ourselves in a singularity, I guess you'd be changing your mind, right?
[00:34:42] Speaker B: Yes, I mean, that would be the least of my concerns at that situation.
[00:34:45] Speaker A: Yeah, I've been refuted now, right? There's bigger concerns. Right.
Are there any other predictions you think that your particular understanding might, whether with respect to AI or anything in general?
[00:34:57] Speaker B: Yeah, I mean, I think it predicts that we're going to see increasing sophistication of AI because basically it says the, the prediction is that humans can't create something more intelligent than themselves, but they certainly should be able to create something increasingly close to themselves as they work on it over time.
[00:35:20] Speaker A: Sure, yeah.
[00:35:21] Speaker B: And so I think even since I wrote this chapter and ChatGPT came out and we've had a bit of an AI revolution, I think that, you know, that prediction has actually turned out fairly well.
[00:35:32] Speaker A: Oh, wow. So you wrote this even before Chat
[00:35:34] Speaker B: really had CHAT GPT?
Um, it was, it was a little bit of a thing that, like just how long it takes things to progress through the publishing process is everything, the books obsolete now that CHAT GPT has come out?
But you know, I think that it's. The stuff we have in the book stands up well to.
[00:35:53] Speaker A: Yeah, it's, it's neat to be able to kind of see that because I know your, your contribution is the only one that was kind of on the other side like before, you know, chatgpt really like kind of exploded onto the scene. So it is cool to see that now that the book is out and, and ChatGPT is, you know, pretty much being used by everyone to see how well a lot of these predictions and anticipations have stood up.
To be sure. Is there anything about, you know, AI now that has, has like really surprised you at all or that, you know, maybe you think is, is a tension for your model or is it all pretty much, you know, Accommodated by how you're thinking about human cognitional ability.
[00:36:32] Speaker B: Yeah, I mean, I think that the abilities have. I mean, I'm impressed by, you know, what chat GPT2 can do. And I was somewhat surprised, although I had seen, I think, a GPT2 model and already come to the realization of, okay, this technology is, is better than I thought it would be at this point in time, and that was before I wrote the chapter that I would have seen sort of some of the precursors to Chat GPT that maybe didn't get as much press.
But I think that in general, I feel it does fit very nicely into kind of the framework that I was thinking of in terms of we expect the progression, but we think there are limits to it, that there are limits
[00:37:14] Speaker A: and limits where we would presumably be able to know if they were surpassed. Right.
And I think that's a neat feature that yours has those specific sorts of, of predictions as well.
All right, very, very cool. Is there anything else about your, your article or your argument that you. That's important that you think we should mention? Obviously we're going to encourage people to pick up Mining the Brain and read it if they want to dive into the more technical details of which there are many. We're just mostly trying to get people interested. But, yeah, anything important you think we missed or glossed over?
[00:37:47] Speaker B: I think maybe one thing to think about, it's worth emphasizing is the argument is, of course, that I make, is that an intelligence can only create a lesser intelligence than itself.
And so the implication then would be something more intelligent than us must have created us.
But there's an interesting hitch there because, well, we can't just postulate an infinite regress of more and more intelligent algorithms responsible for us.
So at some level you have to break free of that and say, well, whatever designed us ultimately must not have been an algorithm. It has to be something that is somehow transcendent and beyond algorithmic status, which I think is very, very suggestive.
[00:38:30] Speaker A: I think that is very suggestive indeed, especially of the traditional sort of understanding of God in line with classical theism and this or that. So that's always great, but, you know, you're not engaged in a full project of natural theology here, so.
[00:38:46] Speaker B: No, but I think it's definitely very interesting that it sort of points very clear, very firmly in that direction as the explanation.
[00:38:53] Speaker A: Yeah, because what you wouldn't want to get to. Right. Presumably is just some like, brute, immense algorithm. Right, right.
[00:39:00] Speaker B: I mean, because then you have somewhat the same problem. Where did that come from exactly.
[00:39:04] Speaker A: So what you want is something that can explain the algorithms without itself being an algorithm, right?
[00:39:09] Speaker B: Yes.
[00:39:10] Speaker A: Yeah, I'm totally on board with that.
That sounds great.
Dr. Ewart, thank you so much for your time here today. Before we say goodbye, two questions. What are you working on next? And where can people keep up with you and your work?
[00:39:25] Speaker B: Well, people who are familiar with me will know that I recently published the second edition of the Design Inference with William Dempski.
I'm currently working on a sort of follow up to that called get your own dirt.
[00:39:37] Speaker A: That's a. I like that title. That's fun.
[00:39:38] Speaker B: Yeah, it's a reference to a joke that you've probably heard about. You know, a bunch of scientists go to God and says, we don't need you anymore. We can create life.
And God says, show me. And so the scientists go and grab some dirt and God says, wait a minute. Get your own dirt, right?
[00:39:53] Speaker A: Yeah.
[00:39:54] Speaker B: And so it's sort of playing on that, making a sort of a design argument that says, you know, even if you're going to claim the world is such that evolutionary processes or things like this actually work, that just pushes the problem back and you're getting your dirt from somewhere.
So that's one of the major projects I'm working on next. I have a website, winstonyork.com you can go there, see links to my various projects and papers.
[00:40:18] Speaker A: Excellent. Dr. Ewer, it has been a pleasure. Thank you so much for your time today.
This has been Mind Matters News.
Explore more at Mind Matters News.
That's mindmatters.
[00:40:42] Speaker B: AI.
[00:40:43] Speaker A: Mind Matters News is directed and edited by Austin Egbert. The opinions expressed on this program are solely those of the speakers. Mind Matters News is produced and copyrighted by the Walter Bradley center for Natural and Artificial Intelligence at Discovery Institute.
[00:41:03] Speaker B: Sam.
