Markov Blankets, Modularity, Enactivism

Welcome everybody to the fourth episode of Active Inference Insights. I'm your host, Darius Parbizzi-Wayne, and today I have the pleasure of hosting Inez Hippolyto. Inez is a lecturer of philosophy of artificial intelligence at Macquarie University in Australia. Her research focuses on 4E cognition, which holds that cognition is embodied, embedded, enacted, and extended beyond the brain. And she studies how this framework can help us understand and build other systems that exhibit a form of augmented cognition, including brain-computer interfaces, neurotechnologies, and smart environments. In addition to this, she co-founded the International Society of the Philosophy of the Sciences of the Mind, Euglartiero Puccini, and she is currently the ethics advisor in artificial intelligence versus the cognitive computing company, which implements active inference into artificial intelligence agents. Inez, thank you so much for joining me. It's an absolute pleasure. Thank you so much for having me. The pleasure is mine. And it's, we were just discussing off air that, you know, we're doing this across many oceans. So Inez is in Australia, I'm in London. But we managed to sync it up, which is fantastic. Excellent. So I wanted to start with Markov blankets. All things start with Markov blankets. Markov blankets really play a fundamental role in active inference and in the framework that we apply to artificial intelligence, to cognition more generally, and also just the underlying physics and massive active inference. And a lot of your work has focused on Markov blankets and the scaling up of Markov blankets. So maybe a good place to start for our audience would be if you could just give a simple overview of exactly what a Markov blanket is, and then we can go from there. Yeah, fantastic. You're absolutely right when you say that everything starts with a Markov blanket. Everything that is a thing has got a Markov blanket. And that does not necessarily mean that a Markov blanket is going to correspond to a certain kind of like physical boundary that one can see. Right. A Markov blanket is a theoretical formal construct that what it does is it tells you how or what things are going to be influencing a certain state to be the case. Right. So it is a statistical tool precisely to understand something really important, which is how a certain thing is coupled with within or with its environment. So everything does start, everything has its Markov blanket, does start with the Markov blanket, because it is a way in which we can capture two very important aspects. And these two aspects are that a thing, typically an open system as the thing, is both thermodynamically open and operationally closed. So this is quite crucial because it is what allows the thing to persist, not to perish, not to succumb to the second law of thermodynamics. So Markov blankets really come up very useful because, to my understanding, they allow us to capture simultaneously these two features, being both operationally open, sorry, operationally closed and thermodynamically open. Because what happens is the system is situated in the environment in a way that is open to the environment. It must be coupled with the environment, otherwise it is perishing. It will not maintain itself. But it also has a certain boundary that allows it to maintain its coherent form. So the Markov blankets become really interesting because they allow us to capture those two dimensions of an open system. And what is an open system? An open system is mostly everything that we see. We don't really come across closed systems. Excellent. Yeah, that's great. I'm wondering whether you could give us a sort of overview of exactly how that coupling works. So people may have seen that the Markov blanket is essentially composed of active states and sensory states. That makes sense if we're talking kind of about physical boundaries like actuators, arms, legs and so on. But on the understanding of the Markov…