AutoGrad Changed Everything (Not Transformers) [Dr. Jeff Beck]

[00:00:00] Speaker 1: So my PhD is in mathematics from Northwestern University. I studied pattern formation in complex systems, in particular combustion synthesis, which is all about burning things that don't ever enter the gaseous phase. Bayesian inference provides us with a normative approach to empirical inquiry and encapsulates the scientific method writ large. I just believe it's the right way to think about the empirical world. I remember I was at a talk many years ago by Zubin Garamani, and he was explaining the Dirichlet process prior. This is when the Chinese restaurant process, all that stuff was like relatively new. And his explanation of it, it so resonated with me in terms of like, oh my gosh, this is the algorithm that summarizes how the scientific method actually works, right? You get some data, right? You get, then you get some new data and you sort of say, oh, how is it like the old data? And if it's similar enough, then you sort of lump them together. And then you sort of, and you build theories and you properly test hypothesis in the fashion. That's, that's, that's the essence of the Bayesian approach is it's about explicit hypothesis testing and explicit models in particular generative models of, of the world conditioned on those hypotheses. It, I believe it is, it is the only right way to think about how the world works. And it's the, and, and, and it encapsulates the, the structure of the scientific method. I mean, if I'm being perfectly honest, what actually convinced me the brain was, the brain was Bayesian had a lot more to do with behavioral experiments done by other people. My principal focus was on, well, how does the brain actually do this? So I'm referring to experiments, you know, showing that like humans and animals do optimal Q combination. We're surprisingly efficient in, in terms of like the information that comes, using the information that comes into our brains with regards to, again, these low level sensory motor tasks. [00:01:53] Speaker 2: Oh, interesting. So it's almost like we, we're so efficient that the only explanation that makes sense is that we must be doing Bayesian analysis. [00:01:59] Speaker 1: Yeah. More or less. I mean, it's a bit more precise than that. It's, it's not just efficiency. It's, you know, like the Q combination experiments I think are really compelling. And so the, the idea behind a Q combination experiment is that I give you two pieces of information about the same thing. Um, and one, one piece of information is more reliable than the other. And the degree of reliability changes on a trial by trial basis. So you never know a priori that like, say the visual Q as opposed to the auditory Q is going to be the more reliable thing. And yet, nonetheless, when people combine those two pieces of information, they take into account the, the relative reliability on a trial by trial basis. And that means that they're optimal in a sense. Now we have to like, be super careful with our words. They're relatively optimal because they're not actually using a hundred percent of the information that the computer, like your, the visual information that you use, you don't use a hundred percent of the information that the computer provided you. Right. But you know, there is some loss between the computer screen and your brain mediated in principle by, but it, the system behaves as if, right. It has optimally combined those two cues. It is taken into account uncertainty. This also is cause it would like how we really do think about the world. Like we take into account uncertainty all the time in our decisions, right? You know, this, if you're driven in the fog, you're aware of this. The 90, 90% of what the brain does is decide what to ignore. Yeah. Right. And because, because if, if we didn't, right, we'd be screwed, right? We get, we receive an insane amount of information. Most of which does not even, we don't even bother to process. Right. So yeah. Yeah. [00:03:34] Speaker 2: Yeah. Is, is that definitely the case though? Do you think that we could actually be processing more information than we know? [00:03:41] Speaker 1: We are definitely processing more information than comes out in behavior. Yeah. So a lot of that is, is because, you know, we are continually learning and like learning, you know, you have to, you know, you, you close your eyes for five years and your visual system decays, right? You lose fidelity, right? It, it forgets. It requires constant inputs simply to maintain this understanding of like the low level statistics of the visual world, right? And then with like, without input, like your, you know, your WMO. So the question is, is, is that, is that using all the information or is it just using like the low level information? Yeah. And it's information that we don't like directly perceive, but it's still, but it's still definitely being used in a sense when it comes to, you know, you know, but it's, but what, what is it being used for? It's being used to track these sort of low level statistics that, that we sometimes need, but don't always need. And so this is why I say that, that like, you know, when we say context matters, you know, you can think of that in terms of like, we were able to flexibly switch between tasks, which means having a lot of resources and having a lot, you know, maintained and having them still be in good working order just in case we need them. Right. And this is why like the self-supervised or unsupervised learning approaches that are ubiquitous for like starting, you know, getting your LLMs to give you, you know, sort of your reasonable prior over, over a language is the sort of stuff that your brain is definitely doing. So in a sense it is using everything, but it's not really using all of the information that's present. Right. And that's sort of, I think the argument that I want to make. [00:05:17] Speaker 3: The idea of having to traffic in squishy people in order to make our systems go is not immediately appealing. Let's put it that way. [00:05:25] Speaker 4: This episode is sponsored by Prolific. Let's get few quality examples in. Let's get the right humans in to get the right quality of human feedback in. So we're trying to make human data or human feedback. We treat it as an infrastructure problem. We try to make it accessible. We make it cheaper. We effectively democratize access to this data. [00:05:45] Speaker 2: What do you think about these broad sort of metaphorical idealizations? You know, the big one is that the brain is a computer. The probably the more popular one is that the brain is a prediction machine. [00:05:58] Speaker 1: It will always be the case that our explanation for how the brain works will be by analogy to the most sophisticated technology that we have. Is that how's that for a non answer? Right. Right. So. So, you know, you know, a couple thousand years ago. Right. How'd the brain work? It was like levers and pulleys, man. I mean, duh. Don't be ridiculous. Why? That was, you know, at some point in the middle ages, it became humors. Right. Because fluid dynamics was like the, you know, was the kind of technology, you know, the technology that was like the most advanced or technology took advantage of of water power was like the most advanced technology that we had. Now the most advanced technology is computers. So, duh, that's exactly how the brain works. [00:06:39] Speaker 2: Philosophers used to think that the universe was a machine. Mm hmm. And we interviewed Chomsky about this as well, you know, because he talks about the ghost in the machine, you know, and the ghost is all of the bits in the machine that we don't understand. But do you think now that we can think of the universe as a machine? [00:06:55] Speaker 1: I think that that that is a very convenient way to think of the universe. Right. So when we model the universe as having like causal structure, right, do we do so because it actually has causal structure or because that's a really convenient class of models with which to work? I think that it's, you know, it has causal structure. Right. Right. But also it's a convenient class of models. Right. So like a good example is large language models. Right. So they're all, you know, most, most, but not all are auto regressive in terms of their predictions. Right. Well, why? Like, why is it auto? Oh, it's because it's mathematically convenient. It's a compact way to like take the past and make a prediction about, about the future. Does it mean that that's actually the way language works? No, I don't think it's actually the way language works, but it's a, it's a computationally convenient model. In physics, we have like, there are in fact, like very, like momentum is a good example. Like why, why do we need momentum in order to describe that? We don't, you know, we don't observe momentum directly. Right. We only, you know, you're just looking at videos. You just, you know, the position of the ball. Right. You know, you want to infer the velocity. Well, you just take the difference between two adjacent positions. And then that gives you, but you don't ever directly observe like the momentum. And this is, you know, in a mechanical, in a mechanical setting. So why did we choose momentum? Well, we chose momentum because that's the variable that, that, you know, that if we knew what, if we knew what momentum was, now everything is Markovian. Right. Everything is, it's a, now there's like a simple, like causal model that describes how the world works. We picked that model because, we picked that particular hidden variable because it's what rendered the model causal. Does that mean that's how the universe works? Or was that just a, a, a computationally convenient choice? I'm going to stay agnostic on that one. But I do like that. It's a computation that ended up working out. Right. So. [00:08:52] Speaker 2: And just quickly riff on the, the benefits of having models that preference causal relationships. So the nice thing about cause. [00:08:59] Speaker 1: So when you have a causal relationship, it reduces the number of variables you have to worry about and track. That's the beauty of having a cause. It's like, it's like a Markov. It's the same. It's the same argument with, with momentum and Markov models. We chose to have that hidden variable because it's the thing that made the model simpler. Right. It made the calculations easy. Now we can just like go forward in time, just make predictions in a, in a, in a totally like iterative fashion. That's what makes causal models. Great. The other thing that makes causal models. Great is if you do ever intend to serve, you know, act or behave. Right. Then you still need to be, you know, you need to be able to predict the consequences of your action. The, the more tightly linked your actions or your affordances are to the things that causally impact the world. The more effective those actions are with respect to your model, but hopefully also with respect to reality. And so we, we prefer causal models, you know, in part because they are relative, you know, relatively speaking, simpler to execute. Right. And in, in a simulation form, but also because they, they point directly to, well, where should I intervene? You know, where should I go in and, and, and, you know, and how should I choose my like series of actions that will give me the desire that would lead me to the desired conclusion or goal. [00:10:13] Speaker 2: What's the difference between micro, micro causation and macro causation? I think the difference between micro and macro is a single letter. No. Um, so. We could just model the light cone at the particle level. Oh yeah. So that's how that'd be expensive. Yeah. Yeah. That's the way physicists see the world and, and we see the world in terms of populations and people and all these macroscopic things. And, and we still reasonably do experiments and we do interventions and we do randomization. [00:10:43] Speaker 1: To truly identify a causal relationship, you have to do an intervention. Right. Um, you know, the classic example, this is also in the, in lung cancer. Right. It was like, I forget how long ago this was, but at one point there was this belief that alcoholism caused lung cancer, but it was actually because they were in poor health because they were alcoholics and they smoked a lot more than the rest of the population. Right. So you do need to do that kind of intervention to discover a causal relationship. However, right. The causal relationships that we care about are the ones that mesh with our affordances. Right. If you know, identifying a microscopic causal relationship is super, that's great. Right. But unless you have really tiny tweezers, it's not very helpful. Right. What you need to do is you need to identify the causal relationships that are present in the domain in which you are capable of acting. We care about the causal relationships at the macroscopic level because that is where we live. We live in the macro, at the macroscopic level. Most of our actions are at them. Now, one of the best things about humans is our ability to extend the domain of our affordances with technology. Right. We have like nuclear power because what we did was we acquired the ability to take tweezers, you know, at that, at that scale and like, you know, make these things happen. Right. We've figured out how to take advantage of causal relationships at that level, not because we have those abilities, but we were able to create the tools that gave us access to that space. It all depends on what the problem it is that you're trying to solve. And the causal relationships that you always care about will be the ones that are related to the actions that you are capable of performing. Now, that said, there's clearly a great advantage in understanding the microscopic causal relationships. Right. If for no other reason, then that might lead to us discovering a way to expand our affordances into the, you know, into another aspect of the microscopic domain. [00:12:37] Speaker 2: Is this just instrumental, you know, is this just something that it's a little bit like we say that agents have intentions and representations and it's just a great way of understanding things. But for all intents and purposes, it's not, it's not actually how it works. [00:12:52] Speaker 1: Well, I think that that sentence ended on a rather definitive statement with which I don't think we could, I would agree. But the rest of it, is it all, you're asking like the, you know, the scientific anti-realist if it's all instrumental. So, yeah, yeah, it's all instrumental. Right. I mean, we, you know, we, the, the things that we care about are the things that, that, you know, again, back to affordances. Right. So, you know, we need to understand causal relationships at the scale that we can manipulate. Right. That's what, that's what matters most. Right. Because that allows us to have effective actions in the world in which we actually live. To the extent that we care about other scales. Right. It's, it is because simply, simply wish to expand, you know, our domain of influence. Right. [00:13:37] Speaker 2: The mind is quite an interesting example. So let's say, um, I want to move my hands and my, my mind willed it. So it's top-down causation. Now I can't act in the world of my mind, but it seems, it seems macroscopically intelligible. You know, we think about our minds. So maybe the mind is a special case. I don't know. [00:13:56] Speaker 1: Well, the mind is a special case. I'll agree with that. I think of like downward causation from, well, I guess from an instrumentalist perspective. Right. It's like, I'm not saying downward causation is the thing. I'm saying that downward causation is one of, is, is like how it all works. I would take it from more from the perspective that, um, downward causation. If discovered downward causation is what justified your macroscopic assumption. So what do I mean by that? I mean that like, suppose I'm in the following situation. I got a bunch of microscopic elements and they're all doing stuff. And I'd like to draw a circle around them and call that a macroscopic object. Now I am justified in doing so. If that particular description of the macroscopic at the macroscopic level, right. Has the downward causation property. Right. It's, it's, it is a way of sort of saying, oh, that was a good, you drew that circle you drew. That was a good circle. Right. Because it's summarized the behavior of the system as a whole. Right. In a way that rendered the microscopic behavior irrelevant to further, for further consideration. Right. [00:15:05] Speaker 2: Yes. I can think of some situations where we do this. I mean, we might identify an aspect of culture or a meme, and we might say, well, that is responsible for violence or something like that. Yeah. [00:15:16] Speaker 1: You still have to show that it has that property. Right. And I think, you know, intentionality is a tough one. Right. Because, you know, it's, it's a variable that has a lot of explanatory power, but it's not, but, but it's not one that evolves. So when I think of a good macroscopic variable, it's one that I understand how it evolves over time. That's what makes it a good macroscopic. I can just write down a simple equation and it says, you know, pressure, volume, temperature, right. They are going to do this over time. And like taking any little microscopic measurement becomes like totally irrelevant. Right. But what made it useful wasn't just that the, the microscopic measurements are irrelevant. Right. Right. It's that I had an equation that describes how it would have behaved, you know, that's also fairly accurate. So I have a nice determined, you know, relatively deterministic model that's, you know, that, that is at the macroscopic level. Right. And so when you talk about like intentionality, I think it's, you know, yes, it can be used as an explanatory variable, but it's only good to the extent that we understand how that intentionality changes over time. Right. It's a long-term prediction. And this is why like, you know, the jurisprudence example made me really uncomfortable because it's sort of like saying, well, you know, what you're kind of doing is you're saying this is a bad person. Right. And I don't know how we would necessarily like identify like that intentionality, except in a very indirect way. Right. That is that, that then they're stuck with, but then, you know, because it's only good as a macroscopic variable, if we can make predictions about the, how that variable changes over time. And we're not doing that. We're saying that you're stuck with it. Right. And I just, that's why it sort of makes me a little uncomfortable. [00:16:59] Speaker 2: I did, I did actually notice that, um, the active inference community has quite a ragtag. It's, it's got very diverse. Yeah. So in, in a way you see people rubbing up against each other that you normally wouldn't. And that can create arguments. Yeah. [00:17:17] Speaker 1: Well, I think, you know, this was, this was, this was Carl's influence. So what did Carl actually discover, right? He's got this link between information theory and, and, and, you know, and statistical physics that in some way gives you this sort of uniform mathematical framework that's widely applicable to a huge number of situations. It has a lot of sort of things that are baked into the, how we think about the world is kind of like baked into it. And so it can be applied in a whole bunch of different areas. Hmm. And Carl spent a lot of time basically evangelizing various different aspects of the scientific community. It's like, oh, look, you can apply this to epidemiology. You can apply this to the social sciences. You can apply this to physics. You can apply, you know, and just sort of in, you know, wrote a series. This is one of the reasons I think he's so prolific is because he's basically, you know, written variations on the same paper. Right. But just applied in different domains. And he did this, and this was intentional, right? Because he wanted to show that this is a, is a nigh uniformly applicable mathematical framework. And I think he's largely right about that. Um, as a result, right, there's all these people from all these different communities that have been pulled into his sphere that think about the world very differently. And it makes for some very entertaining conversations at the pub. Yes. [00:18:31] Speaker 2: Even in our discord server, you know, we, we've got people were thinking about it in terms of crypto, even in terms of Christianity, phenomenology, um, psychology. It's, it's really interesting, but yeah, it's, it's, uh, but that's the beauty of constructing [00:18:45] Speaker 1: like a, a nearly uniformly applicable mathematical framework, right? Yes, exactly. You get to, you get to suddenly, this is one of the things I love. I mean, this is what I love about the community. In fact, is that we now have a relatively common language to discuss a huge variety of different things. Yeah. Um, now of course that means we often end up talking cross purposes, but that's half the fun, right? So I often ask people in the business, like what, what, what changed? Like what's, you know, what's what, you know, why did we have this like massive explosion in, um, you know, in AI development over the last several years? Um, and I get three, there, there are three common responses and I agree with every single one of them. Autograd, right? The transformer, but why the transformer is something that I, I often disagree with, with people about, uh, the transformer architecture, um, and just the, the, the, the ability to scale things up in a manner that we haven't, haven't really seen before. I actually, uh, the reason why I say transformer comes with an asterisk is because a lot of the things that transformers have, that, that people believe that the transformer enabled, um, I think really resulted more from scaling. And my, the point, you know, the point of evidence that I like the site is like Mamba. Mamba, which is a state, which is a traditional state space model. It's basically a common filter, but like on steroids. So they scaled it way up and yet, and now it's, you know, got, they've, you know, Mistral has their very nice, like coding agent and it works pretty darn well. Right. They got a lot of the same functionality with a completely diff with a, you know, a completely different architecture simply by virtue of scaling. So transformers get a, get an asterisk. I think that the biggest thing was auto grad, right? And auto grad turned, um, the development of artificial intelligence, um, from being, uh, something that was done by like carefully constructing your neural networks and then writing down your learning rules and going through all that painful process. It was tick, tick for, and they turned it into an engineering problem. It made it possible to experiment with different architectures, different networks, different nonlinearities, different structures, different ways of like getting your memory in there and different ways and all this fun stuff that allowed people to just start trying things out in a way that we couldn't do it before. And then we, what did we did? We, we suddenly discovered, oh, it turns out back prop does work. I mean, when I was a young man, like back prop was considered a non-starter for two reasons, right? One is, is it is not brain-like, which is true, right? Yeah. The brain does not use back prop. And the other one was a vanishing grades. Oh, you'll never solve the vanishing gradients problem. And it's like, oh, it'll always be unstable. And they, and yet nonetheless, once we turn into an engineering problem, start playing around. With tricks and hacks and certain kinds of knowledge and values and this and that. We discovered that, oh no. In fact, like there are ways around this. You just, you know, we just, you know, weren't going to discover them by like playing with equations. We had to actually start. So we turned it into an engineering problem. As soon as it got turned into an engineering problem, you know, that's what enabled the hyperscaling, which is what led to all of this, all of this, you know, these great developments over the last several years. What got lost in the mix though, was the notion that, that, that, that there's more to artificial intelligence than just like function approximation. We got really good function approximators, but that's not the only thing you need to develop like proper AI, right? You need models that are structured, like the brain is structured. You need models that you need, you need models that are structured, like how we conceive the world is structured. Certainly if you want to have models that think the way we think and that, that got lost in the shuffle. And we're starting to see, you know, as, as, as we're starting to see the limitations and the faults and flaws of, of, of these approaches and starting to see them not living up to the hype, which I think is like now it's standard that like, like AGI is no longer, I don't know if we read the other day, at least according to, you know, the experts in the field at the top of, of, of, of the best companies in the business, like AGI is no longer like a huge priority. Right. And that they're, they're, they're dialing back the rhetoric surrounding that. Um, in part, because I think that they've begun to realize that like just function approximation, isn't going to deliver that was just hype. Right. We do need to do something different. We do need to start bringing in what we know about how the brain works. Right. If we're ever going to get to something that is a human like intelligence. And that was the starting point for us, you know, about a year or so ago is that we were sort of like, yes, let's do the same thing for cognitive models. Like let's talk about, let's take what we know about how the brain, the brain actually works. Let's take what we know about how people actually think about the world in which they live and start building an artificial intelligence that thinks like we do by incorporating these principles. And this means, this means basically creating a, you know, a modeling and coding framework for building brain-like models at scale. And that's like the critical element because obviously scaling was a, was a big part of the solution. And right now, most of the work in the active inference space, as I'm sure you're aware, is not at scale. There's very little like active inference work that is active inference at scale. Most of the models are like relatively small toy grid world-y type models. And part of the reason for that is that, you know, it is in fact difficult to scale Bayesian methods. Now that also has now begun to change, right? We now have a lot of great mathematical tools and a lot of great frameworks for approximating Bayesian inference. You'll never do it exactly or approximating Bayesian inference, which I believe is how the brain works, right? Bayesian brain and all that, that allows us to build these kind of structured models that, that, that, that are structured both after the brain, how the brain is structured and how the, the world that we live in is actually structured. Hence the, the, the, this notion that what we need to build, get the net to the next layer of, of AGI. And I also don't like that term and don't intend to use it very often. Um, but what we need to get to the next level, right? Is, is, is this, um, uh, is this framework, um, that allows us to build, uh, the kinds of models that we know people actually use and just make them bigger and more sophisticated and, and, and, and so on. And then take advantage, like hyper-scaling Bayesian inference is part of it, but also like it's, you know, constructing models. Um, of the world as it actually works, the way the world actually works. Right. Is what is, is, is it, you know, provides us with the structure of our own thinking, right? The atomic elements of thought is how I like to phrase it. Um, are models of the physical world in which we live in the physical world, which we live is a world of macroscopic objects that, you know, um, that have specific relations and interact in certain ways that we understand. Right. Um, you know, uh, looking around the room for a good example, right? You sit on a chair, right? That's an example of a relationship. It holds you up and all that fun stuff. Um, and those are the kinds of the, you know, that understanding of the physical world was necessary for us in order, you know, for us to have in order to survive. Dogs have it too, right? It's language isn't what makes, you know, it isn't, isn't all that special. Right. Well, it's, it's actually quite special, but, but, um, uh, those are the models that form that, that, that understanding of the world in which we live is where we get our, the models that form the, the, the, the, the, the models that form the atomic elements of our thoughts out of which we have composed more sophisticated models that have allowed us to do all this great systems engineering, build this great technology that we've got. So that's what we want to do, right? Is we want to, is, is we're focused on building cognitively inspired models that are based on our understanding on, on, on the way the world in which we live actually works because we believe intelligence must be embodied. Building a framework for, for putting those models together and experimenting with them at scale, all in approximately Bayesian way, because we believe that's how the brain works. It's not just about putting your AI into a robot. It's about giving that, giving the robot a model of the world that is like our model of the world, a model that is object centered. It's dynamic. It's, uh, it's largely causal. Right. Um, it's, it, you know, that's, that's, that's the big difference. And I think that the, the, the, the sort of sparse structured models is another sort of key differentiating component. Like when you think about how, like a transformer and LLM work, a transformer takes every word in the document and says, now, how does this word work? relate to every other word that does it many, many, many, many, many times, right? It's a, it, it, you know, it's, it's very much word. It's the same thing with like your, your, your generative, um, uh, vision language action models. They operate in pixel space. They are microscopic models. Now, yes. Do they have an implicit notion of sort of macroscopic? Yes. They must because they work. Right. But it's implicit and it's not implemented with the kind of sparse structure that actually exists in the real world and in our conceptualization of it. And that's the thing that we are, that we are, we're saying, no, no, no. Look, like if we want an AI that thinks like us, right. Then we are going to build models that are structured, like both like the real world is structured. They have this sparse causal macroscopic structure to it. Um, and so should our models and social. And, and, and the only way to do that is not just to like put a robot in the real world, but to put a robot with a model that is structured in that fashion into the real world. [00:28:11] Speaker 2: No one's using the XLSTM. Not many people are using Mamba because why not? All you need to do is just scale the transformer as much as possible. So, um, you know, many people just really think you just magically get these things for free. Right. [00:28:25] Speaker 1: So I think you could argue that, that with enough data, that's the right kind of data. One of these like really big, super scaled models will, uh, obtain an implicit representation of the world. That is more, more or less correct. Now having an implicit representation is great. If, if your only goal is to just represent the world. If your only goal is to just predict what's going to happen. But it turns out people do something, which is very different. People are creative. People can solve novel problems. They can't, it's not just about mining old problems and figuring out where I can move some words around and get a pro and get an answer that looks more or less right. Right. We actually are capable of creating. We're capable of inventing new things. The way that we invent, I think is exemplified by, by like systems engineering. Right. How does systems engineering work? Well, I, I, you know, I know how, you know, I, I know, I know how an air foil works to create lift. I know how a jet engine works to create thrust. Right. And I can take those two bits of information to invent something brand new, which is an airplane. Right. Right. That kind of systems engineering was predicated upon having this sort of model of the world that was relational. Right. Here's the wing. I can put a jet engine. I can like, I don't know. You don't staple it on. I'm sure you use rivets or something. Right. I know how to put things together. I know how to construct new relationships and new objects. Right. An AI that, that is designed for systems that is designed to do systems engineering will have a object centered or system centered understanding of the world. And we'll know how those, all of the objects relate so that it can sort of start experimenting with different ways to combine them. It's absolutely, you know, without that, the only thing you will ever be able to do, right, is just retool old solutions for new purposes. And it won't even, and even that is, I think it's a generous interpretation of what a purely predictive model is going to do. Right. So this is how I like to think about like, you know, the, the principal advantage of taking this object centered approach, right. Is that it enables systems engineering. What is a grounded world model? That is a, so, so I, I feel like that's a trick question. I was, I actually had this conversation with, with, with one of my friends, Max the, uh, and co-conspirators the other day. Um, uh, in some sense, every model is grounded. It's grounded in the data that it was given now. [00:31:05] Speaker ?: Okay. [00:31:05] Speaker 1: So that's like a true statement. It's like, yeah, okay. [00:31:07] Speaker ?: Yeah. [00:31:07] Speaker 1: But that's not what we want. And when we often use the word, like a grounded world model, it's, we say that it's grounded in something and that something is not just the data that it saw. So example vision language models, a vision language model is like, is a way of grounding the visual model in the linguistic space. And this is the approach that we're doing. This is what lane chain does, right? It's all about taking, you know, models and everything becomes a blank language model, right? Uh, you know, vision, you know, odd, you know, whatever, everything becomes. And what, what you're, when you do that, what you're doing is you're saying that, that you're grounding all of your models in like a common linguistic way. So that they can communicate with one another, right? Via language. Now, why did we choose language? Well, we chose language because like, honestly, I think it's because we wanted models that we could talk to, right? We wanted, we wanted a model that like, you know, it was really all about the, making the interface convenient for us. And which is great. That's totally something you want, but it begs the question, what's the right domain in which to ground your models? Yes. Now I like grounding models. Like, so we also use the phrase like, you know, one of those ground truth. And of course, ground truth is the thing you made up and a priori said was ground truth, right? So what's ground truth? What is the, what is the right, you know, domain in which to ground models in order to get them to think like we do? That's the relevant question. And so my, my view is, is that if, you know, again, if you want AI that thinks like we do, you need to have it grounded in the same domain in which we are ground. [00:32:47] Speaker 2: Hmm. [00:32:48] Speaker 1: And we are grounded in this domain, right? This is why the embodied bit is such an important thing. Um, we want models that, um, that are, are grounded in the physical world in which we evolved. And the reason for this is because that is the world that provides us with these atomic elements of thought. Your single cell like lives in a, in a soup, right? And it has, uh, you know, and it, it, it, it's, you know, whatever model it has of the world to the extent that it has one, or it behaves as if it has one. Um, that model is, is the model of its environment, right? If it didn't understand the environment in which it lived to some extent, right? Then it would, it wouldn't be able to continue to exist and function in that environment. So you can sort of say that a cell has a model that's grounded in chemistry, right? Of the chemistry of the soup in which it lives. You know, when we talk about like that, that is a prerequisite for its survival. Now we talk about like mammals and bigger animals and things that live in the macroscopic world that includes other animals. right and you know and all that so what's that model what's what's what's the world the the you know well at the very least we can say that whatever models we have a significant subset of them are grounded in that world right and that world we know has properties that we can understand it is object-centered it's relational it's all this you know all this stuff um and so the ground the the the the the grounded bit is more about like properly grounded grounded in the domain in which in in which we are grounded as a route to con to creating you know AI you know an AI models that in fact think like we think right that's the grounding that we that we're particularly focused on if you had to choose the domain in which to ground your models what would you choose right I don't think language is the right one language is an incredibly poor description of both our thought processes and reality I tell the story all the time right so you ever you ask any cognitive scientists or psychologists who's done some experimental work with humans right the you know you you put them in a chair you make them do some tasks you carefully monitor their behavior you look at what they did right and then you have a nice way of and then you you know that informs your theory of that behavior or however that works then you and you know and if you do the experiment well you have a very good model of how they made whatever decisions they made throughout the course experiment and then you go back and you ask them what why did you do what you did and they give you an explanation it sounds totally reasonable it also is completely inconsistent with an accurate model of their behavior self-report is the least reliable form of data right that one gets out of a cognitive or psychological experiment and so we don't want to rely on that we don't want to ground our models in what we know is an unreliable representation both of the world and of our thought processes right we want to ground it in something that's a good model of our world and that's why we we've chosen to focus on like mac you know models that are grounded in the domain of macroscopic physics as opposed to language can you speak a little bit more to the the limitations with current active inference a nearly uniformly applicable information theoretic for describing objects and agents right it's it really is inspired by statistical physics and its links to information theory and when you take those two mathematical structures throw in a little like markov blankety things so you can talk about macroscopic objects you kind of have a very generic widely applicable mathematical framework that you can throw up many problems and a lot of what has gone on in the active inference community over much of the last 20 years has been demonstrating that um it's a it's it's like uniformly applicable so there's been a lot of breadth breadth and not a lot of depth right and part of you know and and i think that you know of course like that's you know that that's appropriate right given you know you if you really want to make the the argument that everyone should be using this you so see in this in this domain it works on your like toy examples but the people doing that right it's kind of you know the active image community has had that has this habit of showing like like see like oh this basically like i'm i can handle this like psychological phenomenon i can model this cognitive phenomenon oh and look like it's a good post hoc description of this neural network's behavior and things like that right there they've been showing that but they've they've never really sat down and and and like tried to tackle any really big really hard problem because the emphasis has been on evangelism you couple that with the fact that there is this strong bias within the active image community towards being as bayesian as possible and so of course they also like shun the really hard problems because bayesian inference you know is it has been historically challenging to scale there have been a lot of developments over the last few years that you know um that have come you know out of the machine learning community as well you know but most of the bayesian machine learning community um that that have really made it possible start scaling bayesian inference in ways that we we we really weren't able to do it before um uh and you couple that with a desire you know to sort of stop the evangelizing and start solving really hard problems with these methods um and you've got a way to prove that like active inference really can live up to its promises [00:38:03] Speaker 2: yeah it was a similar thing with um constraint satisfaction you know but in the 1970s there was that light hill report and people said symbolic ai will never work and they wrote it off apparently just that there are all these empirical methods that have been discovered in the last 20 years that just make it massively more scalable and tractable and is it the same thing here right are there some specific techniques that have dramatically improved the tractability of active inference uh well i would just [00:38:26] Speaker 1: sort of i would lump it all into the into the bayesian inference category there have been a number of developments over the last um i would say eight eight eight yeah eight years or so um that have made uh bayesian inference significantly more tractable than it used to be um some of it had to do with you know work in the sort of um gaussian process space uh my my current favorite trick is is you know normalizing flows um which is a great way of ensuring that you have like access to sophisticated likelihoods but nonetheless result in tractable probability distributions um uh uh there's the work um you know i mean i've been using i've been using like natural gradient methods for a very long time which allow you to like massively speed up gradient inference and in some situations completely eliminate the need to do gradient inference and instead like you know do coordinate descent and allowing you to take massive jumps in parameter space and not actually lose the the ability to do learning of sophisticated model in a sophisticated modeling scenario um i also like the fact that like the the natural gradient stuff has been getting some great acronyms recently like bayesian online natural gradient for bong for short i just think these guys these guys get me every time i i wish i was that clever honestly but like it's it's but there's been a lot of developments in that space as well um you know making you know uh in addition additionally there's been a lot of developments in like rapid sampling methods uh conditional sampling methods constrained methods like that that that that that have really improved things and i think that like one of the problems again with the active awareness community you know historically that that i think is now starting to change has been a hesitance to use these sort of certain approximate methods there's been this this focus on like straight up old school message passing and you know as soon as you sort of you know you know if you relax the desire to be as bayesian as possible it opens up a lot more possibilities for for [00:40:29] Speaker 2: scaling this stuff up when we're now talking about agents that are you know interacting with the world around them and that that still presumably needs a lot of data so so we've we've got a couple of tricks um [00:40:43] Speaker 1: one of the nice things about taking an explicitly object-centered approach is that you can you don't have to train all of your models you don't have to train just one model at a time right this is this is my favorite trick and i think that we'll you know this is one of those things i think we're going to be seeing a lot more of in the near future but um you know so if you want to train you know a vision model to understand like youtube videos or something you know really complicated like that you basically take one big model and you train it on a ton of data right you just keep training keep training and eventually it sort of gains this implicit and it doesn't get an implicit sort of sort of object-centered understanding another way to go right is is to is to you know train objects in specific domains that are so these are smaller data sets like i'm only going to worry about like the zillow problem like the inside of people's houses right and that's going to have a much smaller set of objects right that it has to that it has to learn an implicit distribution over and you can do this with one big neural network right you can train and you know there's a really great like you know gaussian splatting paper where they trained a massive neural network that is able to like you know sort of make predictions about what's going on inside people's houses and some nice language models but obviously it has it has an understanding that's limited to a house and the objects that are inside a house if you have an explicitly object-centered model then you end up not just with one model that understands a house you end up with one model that's actually thousands and thousands of little models each of which right you know um uh sort of explains like a single object or object class within the house right so you got like a book model so all books come in different shapes and colors right but there's just one like book model and the beauty of doing this is that is that that book model you have to be a little clever about the how you structure the interactions between these things but if you're a little bit clever um about um how you describe the relationships between objects within this modeling framework you gain the ability to train a model just on the insides of houses a model like just on you know just on like um you know parks and park benches and take the objects that were discovered in this space and the objects that were discovered in this space and put them into a combined environment that has objects of both of those kinds and it still works right that's the advantage of the of taking an object-centered approach or a what i like to refer to as the lots of little models approach some of [00:43:12] Speaker 2: these things are a little bit weird you know some cultures have you know maybe one culture doesn't have the notion of time and some cultures might see two objects as one so is is there a potential problem here that there's some ambiguity that we need to overcome i'm not going to say that there's [00:43:29] Speaker 1: not the potential problem for ambiguity that we need to overcome what i will say instead is that is that the the additional constraint that we're imposing it's not just about objects it's also about their relationships now think about physics this is this is why we this is why the physics discovery stuff is such a big part of it right in physics um is in particular like you know newtonian mechanics um you can you know let's pretend we're living in a world of rigid bodies right so all i need to worry about is like weight and shape of things and that defines a particular object type but i also need to know how they interact and so in in in natoni mechanics we have like you know what we can do is we can take these objects we watch them like bouncing off of each other and doing all these sorts of things and we can quickly infer that like oh like their interactions are all governed by us in a single language which is the language of forces and force vectors right um that like that that language of interaction right is really what makes it work right otherwise we just have like you know pictures of things that's all we would have got what you're empirically discovering is sort of a generalized notion of forces that describe the relationships between things and you can the constraint that you place in order to avoid the problem of things being too brittle right is that well they all have to use the same class of forces together in order to interact we're stuck with that but by being flexible about our definition of what a force is and the having the ability to discover new kinds of forces not just like literal force vectors right gives us the ability to sort of generalize without be without becoming [00:45:10] Speaker 2: too brittle you're talking to this interaction dynamic so there's a graph of interactions which might possibly represent affordances in the macroscopic domain and by doing analysis on the interaction graph you and and sort of simplifying the the analysis as much as possible you get a principled way to [00:45:29] Speaker 1: partition the world up that's right and so so there's there there it's it's uh it it's all about having um interactions and interaction classes so it's not like there's not just one adjacency matrix right there's an adjacency matrix that also specify there's there's one for every type of interaction that that uh that that's possible um that's what gives you the additional flexibility the other thing that gives you the additional flexibility is being is being a little bit bayesian about things right it may very well have been that all of your observations of this object you know when it was in a house were like really simple it was all just it sits on a shelf right and so what do you know well what you know is that that object sits on a shelf but you have to be you know which is one kind of interaction right that's just the you know it has a force pushing down there's a force pushing out you don't know anything about like the weight on this you have nice but if you keep error bars about that if you keep error bars about the other kinds of interactions that you have seen but are agnostic right about the specific deals for this particular object it gives you the flexibility to say well i'm going to put it in this environment i can make some predictions about how it's going to behave right but if i throw a bowling ball at it i'm going to be making some you know assumptions about how it might behave but i but once the bowling ball hits it right i might have to revise those assumptions this is the other critical elements of the approach we're taking which is continue which is you have to have some kind of continual learning element this is something that really doesn't exist in contemporary ai right and you know you know when you build your big model you've spent millions of dollars training it and then you're done right yes someone else can come along and fine tune it a bit for a particular task right which is great but at the end of the day when you're at the deployment phase you turn learning off um whereas in this approach we're saying no no you've one of the things that's critical a critical aspect of the way we think about the world the way we learn about the world is is that it's continual and it's interactivist right so there's you know and that needs to be true of the objects that we're discovering as well we've learned classes of interactions but just because we haven't seen a particular class of interactions previously doesn't mean we say the others never happen right we still allow for that possibility right and then do continual learning the quick with with rapid updates when we see something happen we see you know a new interaction now the what makes that work right is the fact that you specify that there are certain set of kinds of interactions some of which you previously or some of which you still don't know about and might observe soon and then you could update your posterior beliefs about whether or not that object interacts in that way what would the architecture of such a system look like i'm [00:48:09] Speaker 2: imagining it'll be distributed right so you know we have all of these have all these different agents and and then we have the consistency problem because maybe this agent has empirically learned that these two things are a book but the agent over there you know just thinks this one thing is a book and then there's how many objects are there would it become intractable you know like realistically i [00:48:26] Speaker 1: don't know what so so from a simulation perspective the way that this gets simulated is remarkably the way the remarkably like the way a video game engine simulates the world the only difference being is this abstract notion of forces so how does a video game represent the world well you have all these assets right and each asset is basically a shape maybe a texture color or something like a fork is an asset right or a little like three-legged stool is an asset um and it has a bunch of properties but it's basically you know that you know that are associated with um its shape color mass all of this stuff and then it has a set of interaction rules which are like newtonian forces force vectors then you've got other things like water and sand that have like special rules for them because if you just try to because otherwise you know you need a macroscopic rule to describe them otherwise the compute would be insane and stuff like that so it's very similar to that right we you know what we do when you take this lots of little models approach what you end up with is the moral equivalent of a giant list of video game assets right and then when it goes to modeling a particular environment right when you find the agents that you're talking about that has this lots of little models model in its head what it does is is it it um it sort of looks at the scene and says oh okay i need to worry about these 10 000 little models right now and that's it i don't need the rest of it right and then it just sort of operates in that space running something that looks a lot like a video game simulation right so it's it's that sparsity um is what makes is what makes this little there's lots of little models approach right you may have a million little models but at any given time you only need a tiny fraction of them and you just [00:50:09] Speaker 2: instantiate those the thought occurs though that in a game engine um all of these particles are they're in they're in the engine i can say what what are the forces between these two parts yeah it's called cheating well yeah because you know when you deploy an agent in the real world you can't just ask well what's the force vector between jeff and the light that's right yeah you have to learn those [00:50:30] Speaker 1: you know does this model disc you know if you take a video game engine as ground truth are we capable of discovering the video game the assets and their properties that were in that in that in that game engine so what would your input be would it just be the pixels yeah why not make it hard like if you know it it would be cheating to sort of like start out with something that already segments the image for you right if if you can't solve the hard problem from you know from from the bottom up then [00:51:00] Speaker 2: like it's not a hard problem why'd you do it if i understand correctly a successful implementation of the technology you're talking about would be let's start with a game engine and we almost treat the ai like a black box so it has input you know like i can move left i can move right pan up down i can interact with objects and and then maybe there's some kind of a score function i'm not sure but you know it can learn inside the game engine and it will build up this internal model library that represents things in the world in the game engine and if it's learned a sparse robust model library you could in principle take the same learned model and apply it to a robot in the real world and it would generalize [00:51:40] Speaker 1: that's the idea and that's it and that's that's the problem that we're trying this is like one of the critical missing elements in in the robotic space is that you know if you you know training models in simulated environments does not translate really very well to real world environments um this could have this could be a result of of a situation where the um you know the simulated environment is just too impoverished but it could also be a result of a situation where the artificial environment just isn't actually a very accurate representation of the real world right and i think it's i think it's largely the latter right is that is it these you know these um i mean also coupled with the fact that the robot that the artificial agents internal model looks nothing is not structured like the the world that it actually is being trained to function in i think those are the two big those are the two biggest problems but it could also be the case you know what do you need in order to address those well one is you need a good model for the robot's brain that has the structure of the world in which it lives the other thing is is you need a mapping from real world data to simulated data and right now what we're typically using is video game engine video game engines are great i know i certainly enjoy them on a 10 hour a week basis um the uh the the problem with them though is that they weren't trained to be realistic physics right most of them were were designed to be plausible they were designed to look good to the user um part of this has to do you know and and there's a lot of like tricks and hacks and things that are thrown in to deal with the fact that the equations of newtonian mechanics um are very stiff right when collisions happen if you're just a little bit wrong about that you know then things can you know then weird stuff can happen and and non non-physically realistic things can occur um so if if you if you had the ability to construct an environment that had good enough physics that accurately represented the real world and trained your robots in that domain where they have these models in their heads so they're actually capable of learning you know the quote-unquote ground truth that you've implemented in the in the simulated world then i i i believe that they will uh generalize better to functioning in the real world and this is this is absolutely critical i think for robotics going forward if for no other reason than right now i mean like large language models all these self-supervised models um the way that we're currently training robots to like put your put your groceries away and things like that is all by training them to mimic human behavior right it's expert trajectory learning they're not really learning the physics of their environment they're learning to mimic human behavior without like crushing the eggs right and so with you know you know if you want them to be able to generalize across domains across tasks you need to get rid of reliance on expert trajectory learning and so that's the and that only happens when you move to something that is explicitly model based with a model that accurately [00:54:36] Speaker 2: represents the world in which they live once you've got a core set of models that work in the world is [00:54:41] Speaker 1: that the value of the ai yeah so once you have a core set then then you have the ability to like deploy your agent out there in the world world and it can handle situations that it couldn't previously it handle one of my one of uh my co-conspirators likes to talk about um the uh cat in a warehouse problem right so what do we have so now we've we've we've we've we've got an ai agent that has been trained to like manage a warehouse right and so it understands things like forklifts and boxes and workers hopefully you know and and only and it knows how to and then one day one day something comes along that's never seen before it's cat right cats don't belong in whereas the cat comes along right and so the the model has no eye has never seen a cat before right because that's the environment in which it was trained this is the one of the beauties of this approach so it's the cat comes in the warehouse and it's like what the hell is this and like it's you know it's screwing with my system and um because we're taking this sort of like you know free energy based approach right one of the critical elements is that is is tracking surprisals so when a cat comes along doesn't know what a cat is the surprisal signal goes crazy and then it says okay stop right don't run over the cat right let's figure out what's going on and what it can do is it can take a picture of the cat and it can fire it off to a server somewhere that has a huge bank of models and has been pre-trained on on model selection to to a small extent um and say and it says what the hell is this and then the big bank of models says oh i think it's you know here's like seven or eight things it could possibly be and it's different kinds of cats maybe there's a dog thrown in whatever and then it ports those little models over to the to the to to the warehouse model and then it sort of does some proper hypothesis testing watches the cat behave for a little bit ah it's a cat puts the other models sends the other models back because it doesn't need them anymore right it's figured out that this is what it is and now it's incorporated understanding of the cat into the system this is the beauty of taking an explicit is another beauty of taking an explicitly object-centered approach it gives the model the ability to be um to to to to know what it doesn't know that comes from the active inference component know what it doesn't know when it doesn't know it it can go phone a friend that's another way to describe it the friend will respond by saying oh it's a cat and then and it can take the model account incorporate it into its warehouse model right and now it understands that this is really great from a compute there's a huge compute advantage to this right if we had started with one big model that already knew what all cat was think how many parameters that would have it'd be huge this model is very frugal in the sense that it only knows it only needs to know two things what it needs to know about the environment in which it in which it exists right and when it sees something it doesn't know and then it can just go pull so that's the idea is that you have this massive bank of models but when you instantiate a particular if for a particular use case you don't need them all yeah right you just need the ones that are relevant to that environment but you could but these cape these models are continuously tracking surprise or uncertainty and when it sees something doesn't know before it's [00:57:47] Speaker 2: smart enough to say i don't know what that is how and when should deep learning be combined with this i mean my my naive perception of bayesian inference is right now if you have a photograph from a camera and it's like you know 300 pixels squared or something you know that that would be a challenge for bayesian inference something and could you could you uh just use a like a vision language transformer or something and use that as part of the bayesian framework or could you even use deep learning models as a way of bootstrapping the knowledge acquisition in the in the bayesian framework so [00:58:18] Speaker 1: the reason why i mentioned normalizing flows is because that's technically that's a deep learning tool it just happens to be a deep learning tool for which the output that that takes in an image and turns it into something that is easy to deal with from a probabilistic reasoning perspective are we going to use deep learning tools yes the ones that are fit to purpose for sure um and you know that's that that's a that's a great example of of one where we're taking sort of like oh well like why wouldn't we use this if it's compatible with our framework many folks in the audience won't know what a normalizing flow is can you just give us a quick update on that well so okay so well we all we we've got a pretty good handle on how diffusion models work these days right it's you know you take your image you just add a bunch of noise to it make it gaussian and then you learn an inverse transformation it's the same thing right what you're doing is you're learning a mapping from a probability distribution that is easy to deal with like a gaussian distribution and you're learning a mapping from that distribution onto the thing you actually are observing the thing you care about so in this case it could be an image in fact i actually don't think we should call them diffusion models it's a normalizing flow the diffusion is it should be referred to as a diffusion training protocol for a normalizing flow so to some extent we will be using some of those tricks as well you could say yeah we're going to use diffusion models if you're going to make me roll my eyes and say that jeff what is your approach to alignment well i i typically like to talk to people about their beliefs and values and figure out what it is how it is that they come came to form them and then try to convince them to adopt my values the beliefs that these systems have right the belief that our artificial artificial systems have are not the same as our beliefs and they're and they're the reward functions that we specify for these artificial agents um are definitely not the same as our reward functions now there's a few exceptions it's like like like go chess right any game where you either win or you lose the reward function is obvious um but in general like in complicated situations reward functions um it's not so obvious what the reward function should actually be you know i know that there's this definite belief that like reward is all you need and i and and and there's some truth to that but the question is well where'd your reward function come from now from a philosophical perspective you know there there is no normative solution to the problem of reward function selection i always say barring divine intervention um and that which is just another fancy way of saying that like your values and my values might be different and it's really difficult to say whose are better right um and it from a practical perspective right um a situation that i like to point out is like if you're talking about self-driving cars obviously you'd like to penalize your self-driving car if it drives over a squirrel right but if it had to choose between a squirrel or a cat you probably most people would want it to choose the squirrel and the way you would do that in rl model is you say minus 10 points for a squirrel minus 50 for a cat where those numbers come from it's completely ambiguous right they're that's relatively arbitrary they're they're kind of sort of made up um and so relying on on arbitrarily selected reward functions like seems like a terrible idea we also know that like things can go horribly wrong and i know everyone's sick of this example but like you know you all you know when when when you rely on reward you're you're effectively like you know making wishes from like a malevolent genie or you know you run the risk of saying hey hey skynet end world hunger and it's like no problem kill all humans right those are if you don't specify your reward functions very carefully you can get very degenerate behavior so the goal of alignment in an rl setting is to get you is is to get it would be to somehow get my reward function or perhaps humanity's collective reward function right into the ai agent this is really really really hard it's really really really hard because measuring reward functions is really really really challenging the approach that we're taking is we're saying well so how do people actually do this how do how do we as humans construct alignment well the first thing we do is we try to figure out what other people's reward functions are the problem of reward of reward function identification is conflated by the fact that people have different beliefs action which is what we can observe other people doing is a combination of of their beliefs right and their and and their and their reward function or their values um and so just sort of like taking those you know so the problem of course is is that you only observe people's actions like there's a difference of opinion about what to do right and so you want to figure out why and it could be because your beliefs are different or it could be because your values are different the only the but but it's ambiguous you can't tell it's mathematically it's not even possible to separate these two belief and value are fundamentally conflated when all you observe is action or decision the way that we solve this problem as people is we talk about our beliefs right i tried to i ask you well why do you think this is the action and then you tell me like oh well it's because you know this fact this fact and this fact suggests that if i do this then this will happen right and then i can go in and say ah i see so maybe the reason for the disagreement in our beliefs or in our in our decision right is because you're not aware of this fact and i'd forgotten about this fact and so what we do is say well let's incorporate all of these things together and then sort of see and then sort of and then you would still say like well i still think we should do x and i'm like no it's still definitely y and we continue this conversation until we both until each of us has a very reasonable model of the belief formation mechanism that the other person has at which point the only cause for disagreement is a disagreement [01:04:09] Speaker 2: about the reward function ai systems are completely illegible and that's almost a good thing because if we actually understood how flawed they were they would be banned right well they're amoral it's we [01:04:23] Speaker 1: have no idea to put morality into them the the smart safe thing to do is is is is to remove decision making from their capabilities and to simply use them as oracles or prediction engines right and then we can just say hey like you know what would happen if i did x y and z and then it just sort of tells you well this is the this is the ultimate outcome and then we're like oh okay well then maybe a b and c were better choices right and things like that um that eliminates them from the considerate from the considerate from from uh participating in the val in the you know in the actions sorry that that prevents them from using their reward function right and you can get that just by like training them to just just do good prediction that's totally great but that doesn't give us the kind of automation that we really want right we really want our decision you know artificial agents that are decision makers that can act on our behalf and so it's either going to be human in the loop or it's going to be something like what i propose where we figure out how to like solve the alignment problem in in that fashion [01:05:25] Speaker 2: but jeff you're an old school cognitive guy so for someone like you would you always think that in the absence or in the loo of explicit cognitive models that we would never be able to say that [01:05:37] Speaker 1: these things actually had beliefs or intentions i think that what what allows us to currently say that they don't have beliefs or intentions actually stems a lot from from our knowledge of how they actually work right i for example have no problem concluding that you have beliefs and intentions though it may very well be that that conclusion is drawn from the fact that i really don't know how you work i have an intuitive feel for it i assume you work the way i work i have beliefs and intentions you know that's my perspective of myself and so i conclude the same about you um it's uh it's kind of like emergence emergence is such a funny concept right is that um there's this whole class uh this whole like branch of the emergence literature that defines an emergent phenomenon as anything that i didn't predict right which is a remarkably anthropocentric and i would argue ignorance-based definition of of emergence and i don't like it that's for that for those reasons um the same sort of thing you know goes with you know you know i think this the the sort of converse of that is what's going on here it's like we we know that these that these algorithms do not have like the capability to do anything other than predict and so we don't believe they have intentions but something like strong [01:06:46] Speaker 2: emergence usually means like you know causal irreducibility whatever definition of emergence you end up going [01:06:52] Speaker 1: with it shouldn't be ignorance based it shouldn't be based on like oh well there's you know and so you you know that includes explanations of emergence in that that that involves things like well the only way i could have discovered this was by simulating it therefore it is an emergent phenomenon i don't even like that i don't even like that i am more sympathetic to that but i i prefer definitions of emergence that are sort of more pragmatic right that are sort of like oh no an emergent this is why i like downward causation as like a fundamental feature of emergent behavior mostly because downward causation that is it it's not only a nice explanation of when you can you know a nice sort of fairly rigorous definition of when you can say a phenomenon is emergent it also comes with a practical tool it tells you you don't need to model the microscopic phenomenon last time we spoke about linear and game of life didn't we yeah that was oh that's so i'm still playing with that by the way some of the lenius one of my one of my favorite lenius simulations this is not particle lenio this is the traditional lenio and so what they do is they have a a field and they're obstructions and they're squares and circles and things like that and then they have these little creatures that are sort of like a little amoeba like swimmers like they've got like fins in the back and everything and they swim and they'll hit one of these obstructions which will cause them to deform and kind of look like oh it's gonna die that's so sad then it reforms and becomes itself again and so we thought of this as like a really nice abstract environment in which to to test the um you know some of the properties of the physics discovery algorithm because one of the nice things about the approach we've taken is is that as the thing as a little swimmer goes and hits hits something it's possible that it loses its identity when it like deforms into something new and then reforms into itself and we wanted to see if the approach we've taken captures that and it more or less does right it hits the obstruction right the you know it it it changes its identity into an object of a different type and then reforms and comes out the other side and then [01:08:48] Speaker 2: regains its identity back fascinating quick aside um you know we spoke last time about alex mord vincef and he had this um uh you know convolutional cellular automata with the gecko you know the self-healing gecko and he has now written a new paper with his friends at google and it's using logic gates so it's like an emergentist logic gate thing that draws a google logo and i haven't read it in detail but it looks amazing so i definitely look at that and now you're taking your system and you're applying it in like something like a game of life basically but you still expect it to work yeah well well so i i there are [01:09:23] Speaker 1: forces in in in linia right it's there you know there's the the the the rule that causes the pixels to change right it has a few properties right well it's it's radially symmetric yeah right and um um and it can flip sign but it's like any radial symmetry can work um and then and so it has like a polarity it has like you know and you can think of that i mean it is a force you know in in a sense right and it's even a force that's like kind of like real forces it's like a weird kind of charged particle thing um and so i i still think that you can you know it's you know the the approach that we're taking is basically just discovering what are the effective forces between it's you know we're not worried about the microscopic forces we don't care right that's the whole point of a macroscopic physics you know that there are microscopic forces that govern the behavior of the system as a whole but what you're interested in are the things that that that make predictions on the scale you care about and so what you're doing is you're discovering the effective rules that describe the interactions between not just the particles that make or the pixels that make up the the little floater or whatever flyer um but the rules that govern its interactions with other floaters or physical objects like obstructs they put obstructions in the domain and things like that and uh keith really loves cellular automata [01:10:43] Speaker 2: because they are too incomplete and they have this miraculous ability to arbitrarily expand their memory so you can have a grid size that's this big and you can just add more memory and you can add more memory and you don't have to train the thing from scratch and using some of these approaches we've just been talking about you can actually train the update rules learn the update rules with stochastic gradient descent so do you think in the future we might actually have an ai system [01:11:05] Speaker 1: which is running inside the cellular automata that is a very good question so the snarky response is to say don't we all don't we already i mean what's you know we got it running on a computer and at the end of the day a computer is just a whole bunch of logic gates so isn't it already [01:11:23] Speaker 2: well it's in the same class of algorithms yeah but there seems to be you know a cellular automata that has this emergentist thing so if what it what it does is not how it's programmed and it feels that it feels like there's a trick that the the way it's programmed is an order of magnitude less complicated than the thing it does right so it feels like a magical bridge to do stuff which is more complicated than we could explicitly program or learn i agree but that also sounds a lot like a [01:11:53] Speaker 1: computer you know it's like well what can you do with like a lot of stuff now i guess the difference between a cellular automata is that a computer you know you program it you tell it exactly like you're you specify something whereas in a cellular automata the way that you're training if you're training to do something in particular like so for example find a bunch of discrete objects that go in a certain direction and then you're allowed to like tweak the rules that govern the local interactions until you get something that more or less does that um that's just programming in a sort of backhanded way yeah um but uh uh i think that's i mean it those systems are very interesting because it is remarkable that really dumb simple rules right can lead to like really interesting sophisticated behavior but i've the thing that i find interesting though isn't the fact that like like complicated you know complicated stuff can result from like simple local rules what i find interesting what i'm more interested in rather is like well what are what are the properties of the resulting large-scale objects right is that how is that related to the small-scale objects what's the mathematical description of those big things the things that the things that are that have emerged i'm less concerned less interested in how they precisely emerge this probably is because of my bias for taking a human cognitive approach most of the people don't act you you know when you look at the game of life when most people think oh that's really cool like look at these pretty pictures and all these little creatures and they're doing fun things they don't really care about the low level rules right the thing that captures their imagination is the macroscope the the high level the macroscopic level behavior of these things um though it is cool that [01:13:28] Speaker 2: you can get them from simple rules yes yes no as you say we can program computers but it there's the legibility ceiling we can do program synthesis doesn't work very well it will they're gonna they're gonna i [01:13:42] Speaker 1: i have confidence that that's that's not one of those things that i'm gonna like outright poo poo yes i do have confidence that that there's a lot of that you know that is a rich that is a new area it's a you know a relatively new and they haven't really you know there's a lot and there's you know there are a lot of there's a it has a lot of promise that's what i'm gonna say um and to some extent the approach that we're taking is compatible with program synthesis right we're taking this object-centered description of the world um and the reason we're doing that is because we want to automate systems engineering well what systems engineering oh that's like taking this object and attaching to this one attaching this until you get something that does something really cool right program synthesis right is an abstract way of doing that right is that you start with one program you attach it to another program attach it to another program and so on and so forth [01:14:29] Speaker 2: there is this problem of just understanding the program i mean i'm going back to dreamcoder and i'm sure kevin and josh have put other ones out more recently some of the programs which i learned are just really complicated you know they had examples of like i think drawing towers and drawing graphs and stuff like that and you just saw this you know huge um confection of rules that are being composed together and it's great it has many good properties that it's a program but it doesn't really make sense to us [01:14:58] Speaker 1: yeah to a large extent i suspect that there are ways around that that are related to how it is that like your ai coding agent actually works so for example right when they when they're doing this program synthesis what they don't currently have access to is the is you know is the kind of data set that like github has access to they don't have access to a whole bunch of really well-written programs that were that do exactly what they were intended to do um there was a paper in nature this is actually one of those situations where um you know neuroscience is is making interesting statements about machine learning uh from tony zador and what he had done is they'd taken a whole bunch of neural networks that did a variety of different things and then they came up with a way of genetically encoding them uh for the purposes of seeing if like okay so what's this so it's like oh i had to have a layer that did this and then a layer that did this and then and then and then what i'm going to do is i'm going to like compactly represent the weights in each layer and come up with a representation of that and then i'm just going to look at a whole bunch of different neural networks to solve a whole bunch of different problems and say are there any like patterns that are present in these neural networks such that when i have a new problem i'm interested in i can sort of just you know take something that understands this genetic code maybe mutated a little as a way of sensibly you know traversing the space of possible neural networks until i find the best one program synthesis could in principle exploit the same trick they just need the data set to do it yeah yeah what are what are humans in a world where everything can be done by a robot yeah