Active Agents: An Active Inference Approach to Agent-Based Modeling in the Social Sciences

hello and welcome this is active model stream 15.0 with Andrew Pasa from ABM to active inference inference-based action selection so Andrew thank you for joining looking forward to this presentation and overview so go for it awesome thank you so much Daniel uh yeah again I'm Andrew a and I uh I am a full-time data analyst SL scientist engineer uh many roles to fill when working in the nonprofit world and uh so a couple months ago I gave a uh a workshop like a full uh talk on constructing active inference agents using the pmdp library as we'll be using here today um and uh during that talk I described how we might be able to employ active inference agents uh whenever putting together agent-based models uh in the context of computational social science and so you'll see some links in the description referring back to that tutorial that I gave including my GitHub which has all uh tutorial materials uh on that page as well as some in browser Google collab scripts that run python code and so you can actually follow along um a lot of these uh learning resources including this talk I'm giving are sort of part of a a personal initiative I've taken to try and uh create and develop and then share open- Source learning resources for applying active inference um I see so much potential as someone coming from more of a social sciences economics oriented background uh you know for for applying active inference agents in areas where maybe they haven't been employed too much before um and as active inference is beginning to be increasingly applied to uh social science research such as uh you know attempting to study and and simulate uh something like human behavior and decision making uh I thought that that conference was a was a really ripe opportunity to try and uh make those kinds of connections for people and and just kind of get those kinds of dialogues moving um and so for me as someone who came to active influence with that kind of background but also strong interest in cognition um I just really wanted to uh kind of do some from scratch sort of tutorials on how to construct agents um you know as someone who's been interning with active inference Institute for nearly a year now um and I'll also be giving a talk at her upcoming posium which itself will be a kind of extension of that tutorial I gave a couple months ago as well as the talk I'll be giving now um yeah I would just like to uh really make it plain for people sort of the general mechanics that go into constructing active inference agent uh I would also refer people back to the textbook that was published a couple of years ago by uh pulo par and Carl friston um it's freely accessible you'll find links to it in my uh slides as well as in the collab scripts if you follow the GitHub link um a lot of material that I'm using to construct these tutorials is actually coming from that textbook so anyone involved with the Institute uh reading group uh where we cover the textbook will uh kind of feel hopefully a little at home whenever they follow these tutorials um so we'll start off now um we'll be constructing uh the longer span of the tutorial is one uh make the connection between active inference and agent-based modeling agent based modeling is a very common Paradigm in computational social science uh that said and this is not doing it justice but I'm saying this just to keep it brief uh in some ways it's very traditional in the uh kinds of assumptions that it makes more often than not about human behavior uh often times people rely upon a kind of homoeconomicus or or self-centered uh utility maximizer kind of a description of what defines human decisionmaking and um I think that that loses a lot of the sort of richness that we find in cognitive science neuroscience and and Behavioral Studies uh of how people might actually make decisions and active inference supplies us with you know um a rather you know simple assumption it's a very complex assumption that being free energy…