Cover art for ‘11 – Sam Harris And Judea Pearl’

11 – Sam Harris And Judea Pearl

Artificial Creativity

Listen on Apple Podcasts Listen on Google Podcasts Listen on Spotify

References:

“Making Sense” podcast episode with Sam Harris and Judea Pearl: https://www.youtube.com/watch?v=NNDvhFbMD0s
Popper, “Objective Knowledge”
- Chapter 1 “Conjectural Knowledge: My Solution of the Problem of Induction”
- Chapter 7 “Evolution and the Tree of Knowledge”
Popper, “Conjectures and Refutations”, chapter 1 “Science: Conjectures and Refutations”
David Deutsch, “The Beginning of Infinity”, chapter 2 “Closer to Reality”

Subscribe to this podcast via RSS Back to all episodes

Transcript

This transcript was generated automatically and may contain mistakes.

Welcome back, folks. This is episode 11. Today we’re doing something a little bit different. I hadn’t planned this, but the other day Sam Harris released an episode of his podcast with Judea Pearl. Of course, it’s best if you listen to that episode first, but I hope to do this in a way that makes the topic accessible, even if you haven’t.

Most of you are probably familiar with Sam, but just in case you aren’t, he’s a widely read philosopher and neuroscientist. I used to be a fan of Sam’s podcast. That is until he had David on in one of his episodes and my interest in philosophy was completely taken over by him and by Popper. So now I don’t listen much to Sam’s podcast anymore, but I still get notifications when he releases new episodes. And this one is called “Cause and Effect”, and that piqued my interest. So I read the summary and it says that Sam, quote, speaks with Judea Pearl about his work on the mathematics of causality and artificial intelligence. They discuss how science has generally failed to understand causation, different levels of causal inference, counterfactuals, the foundations of knowledge, the nature of possibility, the illusion of free will, artificial intelligence, the nature of consciousness, and other topics, end quote.

Judea Pearl is a fairly famous computer scientist and philosopher. I’ve talked about some of these topics on this podcast, and I was curious what they had to say about this. And I’m not going to lie, I had a bit of a premonition that it wasn’t going to be very good. And that’s because I found Sam to have the wrong epistemology in some of his previous episodes, and I’m afraid this episode is evidence of that. And if you’ve listened to my podcast, parts of that description I just quoted may sound problematic to you.

I want to put some big grains of salt on the table here at the start by saying that I’m not very familiar with Judea’s work. The only piece of his I’ve read is the one in the Possible Minds book that recently came out. David also contributed to it and made by far the best contribution in my opinion. And I don’t have the book in front of me right now, and I don’t remember much of Judea’s piece. I also haven’t read his latest book, and they don’t really go into it all that much in this episode, unfortunately. I could wait and read his book first, but that would take me a little bit. And I want to respond to that episode in a timely manner. So what I’m left with is really only this episode of the two of them talking and trying to get to know Judea better that way.

I also can’t know if Sam’s agreeing with Judea on some points because he wants to facilitate the conversation or because he actually thinks the same way. But I’m going to assume that he actually thinks the same way, at least in terms of epistemology, because he’s said things like that in the past. So suffice it to say that Popper would turn in his grave if he listened to this episode. So what I want to do is go over some of the misconceptions that creep in here and there, not because I want to take the episode apart, but because Sam and Judea are in very good company. A lot of people think the way they do, and I think it illustrates wonderfully the power of critical rationalism, the Popperian approach, because I think we’ll see that it solves many of the problems that they try to approach.

Now, I don’t have the time or the knowledge to go into every topic they cover, so I’ll focus on their epistemology. And one thing I won’t cover, for example, is reductionism. Sam flirts a bit with reductionism here, but overall, he actually sounds a lot less reductionist than he has in the past. Nonetheless, Sam and Judea debate whether or not abstractions can have top-down causal power over physical systems, and I’ll just briefly say here that the answer is yes, absolutely, and Popperian epistemology solves this problem.

Something else I don’t have the time to cover in depth but want to mention is that they both agree that AI should be regulated. It’s not clear if they think AI research should be regulated or if AI programs should be regulated in their functionality, and if they mean AI or AGI. As I’ve shown in other episodes, the research is safe. But beyond that, calling for regulation and involving the government is a surefire way to bring the chances of making progress towards AGI close to zero. And regulating AGI is shackling it. It’s immoral.

Anyway, the main problem they discuss that I want to focus on here is the problem of induction. And what I found is that they address the right questions and sort of circle around a solution, but then never quite get to it. They start out pretty good. Sam says early on that the goal of science is to come up with good explanations, and that’s true. And they rightly stress the fact that many scientists are either not aware of this or simply aren’t in the business of creating good explanations and instead try to reduce everything to formulae or statistics. And that is indeed a dangerous trend, and it’s not what science is about at all.

But this realization writes them a check they refuse to cash because as they start talking about knowledge and knowledge creation, they begin to make some mistakes. For example, even though Sam just said that the goal of science is to create good explanations, they then talk about the so-called foundations of knowledge, the burden of evidence, and so forth. And they do this even though they mention Hume’s problem. This is the problem of induction. Induction is a process by which people allegedly create new knowledge.

A popular example is that you see one white swan, and then you see another, and you generalize from these observations that all swans are white. And then the idea is that the more white swans you see, the surer you can be that your theory that all swans are white is correct. So the origin of the theory is supposedly observation, which is empiricism. We already refuted that in episode four. Empiricism can’t be true because once you apply it to itself, it breaks apart. Empiricism itself is not derived from the senses, so it doesn’t work.

And the idea that further observations support your theory in any way is something we refuted in episode five, where I introduced the functional approach to explanations. A revelation whose content you already know surely won’t count as a revelation. And no matter how many times you test your explanation or function, you have always only tested it a finite amount of times, which you need to compare with the infinity of times you could test it at every spacetime coordinate.

In any case, the problem Hume pointed out is that there is no valid process by which you can claim to know anything about future observations from past observations. Hume’s problem was a problem at the time, but it’s been solved. Popper solved it some 80 years ago. We know the solution. And the solution is that there is no inductive process whatsoever. That simply isn’t how people operate and create new knowledge.

New knowledge is created through guesswork, through trial and error correction, and it begins not with observation but with problems. Popper explains the solution in detail in chapter one of his book Objective Knowledge. And in chapter seven of that same book, he also explains why the method of trial and error correction can sometimes look and feel as though it was inductive, because guessing solutions is second nature to us and happens rather quickly. Those times it feels like induction. And because the method of trial and error correction contains induction as an approximation. So Popper was miles ahead of Sam and Judea here almost a century ago.

Now they continue to talk about Hume’s problem. And I believe Sam at some point says that we have no direct knowledge of causation or of explanations or something along those lines. I think what he means here, and he says this later on, is that any explanation reaches beyond what you have so far observed. This is the problem of induction again. Now, when you’re an empiricist, this sounds like a problem. But when you’re a Popperian, this is a wonderful thing. It means that your explanation has reach, and that’s the whole point. We want to be able to explain things and make predictions, because otherwise we could all just sit around and wait to make another observation.

I think it really depends on what he means by direct knowledge here. I mean, does he mean that we do not have direct access to explanations? Then my answer is that’s false. Explanations are the one thing we do have access to in our minds. Everything else we can only access through those explanations. And so in that sense, our knowledge of the world is always at least one step removed, and sometimes many steps depending on how many explanations are needed. And these explanations can involve things like scientific instruments, which physically remove us further from reality, but mentally bring us closer to it. I recommend you read chapter 2 of The Beginning of Infinity for more on that. So if alternatively, Sam means that we don’t observe anything directly, that’s correct. Every observation is theory-laden.

Judea tries to address the problem of induction with a three-layered model of how we acquire knowledge. He says there is seeing, which is when we observe the world and change our beliefs accordingly. The second he calls doing, which is when you run an experiment, and he believes this is where we get most of the knowledge about the cause and effect in the world. And the third he calls imagining, where you can manipulate theories about the world in your mind.

My response to this three-tiered model is that at the most he’s onto something with a third tier, but splitting them apart doesn’t work. There aren’t three different sources of knowledge. There’s only one, and that is always your mind when it conjectures explanations and solutions to problems. Popper has done the work. And I doubt Judea has something better to offer. They don’t go into all that much more detail about this three-tiered model. I would imagine, again, if I read Judea’s latest book, I would learn more about it.

But again, they keep circling the problem of induction without ever really getting to the gist of it. And Judea says at some point that, quote, if you want to get to a causal conclusion, you must have a causal assumption someplace, end quote. And that’s getting a little bit closer to the solution.

There’s also justificationism finding its way into the conversation because Sam at some point says that he sees the microphone in front of him and believes there’s a microphone because if there weren’t a microphone on his desk, he wouldn’t think there was one. Now, this isn’t exactly wrong, and it’s an example to illustrate counterfactuals, which is something they’re talking about. But I think Sam invokes this example also because it serves as something that he can be certain of. The microphone is right there in front of him. So if he can be certain of anything, or at least more certain than of other things, it’s that the microphone exists as part of his immediate surroundings.

But the real reason he knows with the microphone in front of him is that in his mind, he has conjectured theories about optics and microphones and other things which combined allow him to deduce the existence of the microphone in front of him. So he doesn’t observe the microphone directly, but only through theories. And these theories have no foundation. They’re not secure. All we can say is they have improved over time through error correction. But looking for certainty or foundations leads to an infinite regress.

I think the biggest thing they miss is this. An explanation is the thing that tells you the cause of something. It’s a mental representation of the cause. To explain something is to describe the thing that caused the problem you’re trying to solve. And to understand something is to understand the cause of what happened. Popper’s explanatory approach solves this. Explanations are everything. What I don’t understand is how Sam doesn’t bring this up. And he’s had David on the podcast twice, as I recall.

Okay, well, big deal. So they have the wrong epistemology. Who cares? Well, they should care. They’re wondering how one might build AGI. And you can’t do that without the right epistemology. It’s not that without the right epistemology, it’s going to be harder or take you longer. It’s going to be impossible. AGI is knowledge creation instantiated in a computer program. So if you don’t know how knowledge is created, that means you don’t understand it, and if you don’t understand it, you can’t program it.

And Sam and Judea talk about the problem of induction as if it was this mysterious thing that has eluded us all this time. Popper shows step by step how he solved the problem of induction in chapter one of his book Conjectures and Refutations. And then he goes, quote, thus the problem of induction is solved, but nothing seems less wanted than a simple solution to an age-old philosophical problem. Skipping some. My contemporaries do believe that there are philosophical problems and respect them, but they respect them too much. They seem to believe that they’re insoluble, if not taboo, and they’re shocked and horrified by the claim that there is a simple, neat, and lucid solution to any of them. If there’s a solution, it must be deep, they feel, or at least complicated. However this may be, I’m still waiting for a simple, neat, and lucid criticism of the solution which I published, end quote.

Judea says that he wants to teach computers about causality, and so he’s looking for a way to store explanations on computers. That is a novel problem that goes beyond the problem of induction. And the functional approach, storing explanations as functions, solves this.

Again, I run the risk of embarrassing myself for not having read Judea’s book. And I’d actually love it if you know what I’m missing here, and you could point it out to me by commenting on this episode, but it seems to me that Sam and Judea believe philosophical problems exist, but they respect them too much, and they ignore the best existing solution to induction, and that’s Popper’s solution. Problems are soluble. Induction has been solved.

What I want to see is someone like Sam having a conversation about how the method of trial and error correction creates good explanations. And I mean in detail, in the mind, because that’s the cutting edge of where we currently stand. So let’s move on from there. I’d love for him to have David on his podcast again to just talk about creativity in detail.

Okay, well, in the next episode, we’ll get back to solving some problems ourselves. It turns out there are problems with the Turing test, and we’ll look at a simple, neat, and lucid solution, as Popper would call it. I’ll see you then.


What people are saying

What are your thoughts?

You are responding to comment #. Clear
Markdown supported. cmd + enter to submit. You have free speech here. You’re responsible for what you write. Terms, privacy policy
Your real name is preferred.
This small puzzle helps protect the blog against automated spam.

Sign your comment with GPG to create or add to a public profile with all your comments.

Paste a detached signature of your comment.

              
Paste your public-key block if you haven’t before. You consent to your key’s contents, including your name, being displayed to the public.

              

Preview

Playing ~