The Role of Affective States in Computational Psychiatry

Hello, welcome, it's May 21st, 2025. We're in Active Inference Guest Stream 107.1 with David Benrimo, the need for a common language to unite levels of explanation in mental health, the potential of computational psychiatry. So, thank you for coming, David. Looking forward to the presentation and to anyone's questions in the live chat. So, thank you. To you. David Benrimo, Ph.D.: Perfect. Thank you so much. It's a pleasure to be here. I don't think I've done a live stream before, so this is a new experience for me. Looking forward to any questions that come from the audience. So, today I'll be talking a little bit about some work we've been doing, much of which is relevant, either directly involving or peripherally involving active inference around levels of explanation in mental health research and some new and more recent theoretical work that we've done in incorporating affective states into understanding things like psychosis development, which is a particular interest of mine. I am a psychiatrist at the Douglas Mental Health University Institute, which is a mental health hospital in Montreal, Canada. I did a master's degree with Carl Friston at UCL where we did some work that I'll be talking about today on modeling hallucinations. I have a lab at McGill and I also happen to be the chief science officer of a mental health company, which is not really relevant to the talk today. It's not active inference focused at all. So, just keep in mind the disclaimer. So, my journey towards active inference and towards computational psychiatry in general really was motivated by this difficulty I had as a medical student, as a pre-medical student, with the idea that everything seemed to be about cutting things from bigger boxes into smaller boxes. It's nice to know how many boxes you have to deal with, but that doesn't necessarily tell you anything about what the boxes are actually doing. So, if you start with humans, we can split those into individuals and societies, and then individuals have a bunch of things that are phenomena, and then those phenomena we think are rooted in the brain, and then the brain has regions, and then those light up with different tasks and what have you on fMRI and other things, and then those regions have circuits, and then those circuits have neurons and astrocytes and glia and etc. Those have genes which are regulated by genetic regulators, etc. And so, it all just kind of felt like we were just breaking things into smaller boxes. And this is not a dig against a reductionist approach at all. It just didn't feel like we were quite getting at mechanism. And I always wondered, how does all this work? What is the mechanism? What are we actually trying to achieve? What is the thing that is being computed? And of course, for all those of you who are interested in computational work, this is by no means a foreign concept, but since I'm sure many people listening aren't necessarily thinking about this from a medical or mental health standpoint, I just sort of wanted to give you an insight into how I think about this in the utility of computational psychiatry. So, the way I see it is that it's using new and old computational techniques to understand the healthy and pathological mind or the environment. No reason why we can't use it to explain the environment. And it comprised of different methods that focus on answering different questions. What is thought and behavior, which we think is the manipulation of information, and what answers are useful. So, and those are the ones that can provide predictive tools for clinical or preventative use. Sometimes the useful answers are not the ones that teach us anything about the brain and vice versa. So, we'll be focusing today mostly on the what is thought and behavior piece. The predictive stuff is, I have another talk for that. Not really super relevant to active inference per se. But it gave us these two camps of computational psychiatry, the modelers, which is where…