Max Bennett: belief

Transcript context

…Yeah. This is where I don't see myself as a philosopher, but this is where I do think scientists need to include philosophers. Because when questions become nonscientific, I think the scientific instinct is to argue that we don't draw distinctions between things that scientists that the scientific method can't draw distinction between. But the problem is there might be moral differences between them. So for example, it might be scientifically impossible for us to have we have no methodology for differentiating whether 2 systems that look indistinguishable in their outputs and inputs, which 1 is sentient. And so scientifically, might say, well, for for the because we can't differentiate the 2, we're gonna say they're the same. But that doesn't mean they're the same. That just means that because we have no methodology for for drawing a distinction between them, we're from a scientific perspective, we're not gonna draw a distinction because we're entering philosophical territory. But if you if you take that and then you start talking about policy implications and and the actual values we attribute to them and how we introduce these things to society, I think we need to include a sort of a philosophy lens here because that might not actually be the case that they're the same just because we can't distinguish them. So that's just 1 thought. Mhmm. Other thought on world models. So so 1 distinction I wanna draw because I've seen a lot of confusion on the Internet about about the world model dilemma. So there's a difference between a world model and a model. It is undeniable that language models have a model. And all that means is clearly, in order for g p t 4 to correctly predict the next token in these really complicated language questions, it clearly has some model of something. And because we can ask common sense questions about the world to it and it answers many of them correct, you can say this is a model of aspects of our world without question. I think that would very hard to argue that's not the case if you look at the performance of of g p t 4 and many of these these questions. But what most people mean when they say a world model is they mean a specific process of stimulating an ordered stimulating an ordered states of consequences of different actions and and identifying the the end results of these actions in your head. Another way to think about what we mean by world model is the ability to reason about interventions and causality. Mhmm. And so so this is the Judea Pearl sort of argument here, which is with our world model, we can hypothesis test. I can say, I imagine that if I do this thing in the world, I think this will be the consequence of it because that's what I see in my head. Now I have a hypothesis. Now I'm gonna actually do that thing in the world and see if my hypothesis is correct. And so that's very different than what's happening in a language model where its understanding of the world derives solely from its input data. thing in the world and see if my hypothesis is correct. And so that's very different than what's happening in a language model where its understanding of the world derives solely from its input data. Versus in a world model, my understanding of the world comes from the delta, the difference between what I hypothesize is gonna happen in the world and then my actual experience of it. And this distinction really matters the more we are gonna start offloading our cognition to these systems. Because, for example, everything that ChatGBT knows is on the basis of its input data. Mhmm. And so that means if false information or wrong information is in the input data, ChatGBT is gonna know that information. There's there's absolutely no hypothesis testing embedded into ChatGPT versus our true AGI agent that 1 will will be invented. What it would do is it's going to hypothesize aspects of the world, and it's going to test its own hypotheses. And so if you give it false information, if it reads articles about how the Earth is flat, it's not gonna just start talking about how the Earth is flat. It's going to say, okay. Well, that is incongruent with my model of the world. I'm gonna now render some tests where I could differentiate them, and I'm gonna perform those tests, and then I'm gonna conclude that the world is not is not flat. So so, when 1 says that Chachapiti does not have a world model, I think some people misinterpret that as suggesting that it's just dumbly looking at the statistics. And I think that's not at all what we're saying. In order to correctly look at the statistics of language, clearly, it's built up a very rich and complex model of the text that it's seeing, and that's how it's able to predict the next word so well. But it's not what most people mean when we say world model. Yeah. I mean, couple of things on that. Yeah. I think I think people conflate the machinations of of language models with how we represent them statistically or or abstractly. Because if you look at lot of papers, they they actually represent it like a probability, you know, like a joint probability distribution. And, of course, you know, the way that language models work is completely different to that. But you are bringing in some very interesting things. So first of all, we are agents in the world. So the the agential lens is quite interesting. We interact with the world, so we're not just learning from observational data. It's quite interesting, actually. I was talking with Nick Chater. You know, we said, oh, why is it that in our everyday experience, we see, we experience the world in in 4 d color? And he said it's because it's interactive. So in your experience plane, you can actually seek new information. Right? You can saccade your eyes. You can get new information, and you can you can touch things. And when you're doing future or past simulations, you don't have that interactivity. So there's something about interactivity, which is really important. But it even then, right, you know, we could you know, how far how far could you go? So a complete 1 to 1 simulacrum of the world wouldn't be a particularly good model. And in in physics, there is no causality. Right? So it's just dynamics. So causality is actually something which emerges very, very far up. So we're talking about a model which is an approximation of the real world, which may or may not include causality. It probably would because we're it's an interactive model, and it has this kind of agential map. But I guess we're just kind of drawing the line somewhere, and we're saying, well, that is a world model. And yeah. Well, I…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.