Tim Scarfe: belief

Transcript context

…And then it's very easy to us versus them, other groups. And you start seeing things break down. And that's where 1 mechanism for solving this is very rigid hierarchies. That's what the military does. Another mechanism is embrace the chaos, which is sort of a little bit more Google does. Another mechanism is to have, is to effectively make it a satellite of different startups, which is what Amazon does, where each group is a kind of autonomous and has very clean interfaces to other groups. So it can kind of, there's many different sort of management approaches to this. But the breakdown I do think emerges from humans did not evolve to interact with 100,000 people. We did evolve in an environment where we interacted naturally with about 100 people. And that's why that comes very we don't need as much sort of process to make that work, but we do when we scale it up. Yeah. Yeah. It's fascinating. I mean, as you say, you could argue that Amazon has 1 overarching goal to make money. But as soon as you increase the autonomy in the organization, it's a very human trait, isn't it? So you were talking about the the Machiavellian behaviors and the deceptive behaviors, and you just wonder how much energy is just wasted with infighting. And even though I've made the comment that in the military, you know, they might be doing quite simplistic jobs compared to Google. But even at Google, there's an obsession with job level. That's the I mean, if you go on war or if you go on blind, that's the only thing people talk about is their total conversation and job level. So, you know, mean, maybe we should save the sin the cynicism. But so come coming coming back to the to the chapter. So we're telling the story basically of of how this metacognition and this predictive apparatus, you know, gave rise to an entire suite of complex social behaviors that that we see in primates, which is fascinating. And maybe we should just talk a little bit about, I call it, why bootstrapping? So it's like why why why. I I think there was a, 1 guy at Toyota Research was quite famous because he would get people to ask why 5 times. So he'd say why why why. And it's almost as if there's some magic number, you know, like, everyone is only a certain number of degrees of separation. And it's a similar thing that you only need to ask why a few times and you'll always get to some kind of base reason. And maybe that's why evolutionarily, we have, you know, 2 levels of of causation metacognition in our brain. You know, we we have the the agranular, prefrontal cortex, and we have the granular prefrontal cortex. So I guess, like, 1 potential question there is why is there not a third level of asking why, and what would that look like? And can you just kind of sketch out that metacognition picture in general? When we think about what, when we speculate what agranular prefrontal cortex does, a reasonable framework for it is it generates explanations of an animal's behavior. It models an animal's own behavior. So 1 sort of cognitive tool to reason about that is if it observes a rat wake up and have certain hypothalamic activations and run to it in a certain direction to drink water, it produces a representation that could be interpreted as I am explaining this behavior by I am hungry as an animal. And so that can be useful in a variety of ways. It can trigger simulations to find alternative solutions to satiate the same need. Right? So if I if you think about this, if you put a rat in a novel situation, but the agranular prefrontal cortex infers that right now I am hungry, now we can start triggering a bunch of simulations to try and satiate the same desire to fulfill, this is a little bit in active inference land, to fulfill sort of what I believe about myself through alternative means. It enables an animal to be flexible. Okay. So this is sort of the, this is an explanation of an animal itself. Why would it be the case? So what I sort of argue in the book is the granular prefrontal cortex builds a model of that model. So instead of a simulation, it's a simulation of the simulation. And so what that would mean is if could, as a thought experiment, if you could ask, let's let's go 1 step further if we or 1 step back. If we could ask the basal ganglia, which, you know, is the sort of reinforcement learning system, why did you turn left to go, in this direction to drink water? It would just say because that turning left maximizes reward. The answer would always be the same. If you ask the agranular prefrontal cortex, why did you turn left? You would say, oh, because, I'm thirsty. There's a specific thing that me as an entity, this animal I'm modeling, wants to achieve. But if you ask the granular prefrontal cortex, it would say, I turned left because I am thirsty and that made me think about ways to satiate my thirst. And I simulated going to the left and I remembered water being there because last time I was there, there was water. And so I went to the left. And so in other words, it enables you to simulate different types of simulations and reason about what you would think in a new setting, which of course enables you to think about what someone else might be thinking. And we do this all the time. I mean, someone doesn't respond to a text message. Someone makes an odd facial expression and social interaction. And we're immediately trying to figure out what is this person thinking, why would they do this, etcetera. So the question first is why do we even need this new level at all? And sort of I think 1 of the main adaptive values is it enables survival in the politicking arms race. Because now if I can simulate a simulation, I can infer why you might do a certain behavior, how to manipulate someone's knowledge, your intentions behind things.…

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