Synthesizing the Born rule with reinforcement learning

All right, hello and welcome to Active Imprance Model Stream number 13.1. It's August 23rd, 2024. We're here with two of the authors of the paper Synthesizing the Born Rule with Reinforcement Learning. So thank you both for joining for this presentation and discussion. To you, Jacques, for the presentation. Great, thanks for inviting me, Daniel. And yeah, it's really fun to learn about the Active Imprance community and to speak to you guys. So I think there'll be as many questions from us to you as there will be from you guys to us. So I'm not going to take up too much time, I hope, with my slides. I wanted to just give a kind of high-level overview of this article, which hopefully many of you have already had time to look over. But I'll give you a high-level overview and that shouldn't take more than about 30 minutes, I guess. And after that, we'll have plenty of time for discussion. There will be many points where you might feel that you want more information, more technical details. But since I couldn't really predict which points might strike you as the most interesting, I'll leave it to you to ask at the end. And as necessary, I might be able to pull up the article and point to the relevant parts. Yeah. Let's dive into it. Oh, and one last thing. You might hear some background noises of either drilling or baby screaming. That's just because I'm at home at the moment and it's part of the domestic atmosphere. Okay, let's go. So this is a very busy slide. Sorry for that. But we do have a lot of people involved in this from many different institutions. I'll put this slide also at the end of the talk in case you're curious. But it's important to mention we've got five people based at the Institute of Physics in LabQ Rio, Rio de Janeiro. And that's where I'm currently based. Although before this, I was, you know, during the time that this paper was mostly written, I was at the University of Massachusetts, Boston, together with two other co-authors on this paper, are from there as well. And we're all part of the Cubism group run by Chris Fuchs. So I'll speak a little bit about Cubism, which is an interpretation of quantum theory based on subjective Bayesian decision theory. A bit of a mouthful, but if you're curious about that, we can talk about it at the end. And John has since gone on to the University of New Mexico. And the lead author of this work is Rodrigo Piero, who is a PhD student at the time of this work. And he's now at the Technology Innovation Institute in Abu Dhabi. So this is another paper that was published some years ago now, just during the pandemic or towards the end of the pandemic. And this was by us at the Cubism group. So you can see John and I are there. Chris Fuchs is there. This was a purely theoretical paper, but it contains the seeds for the article that I'm going to talk about. So I'm going to just take a little diversion to explain what we did in this paper as a motivation for the article that I'll talk about. Here's a little excerpt from the abstract. The title of the paper is Born's Rule as a Quantum Extension of Bayesian Coherence. And explaining that a little bit from the abstract, the subjective Bayesian interpretation of quantum mechanics that we're calling Cubism, asserts that the Born rule is a normative rule in analogy to Dutch book coherence. I'll explain what Born rule and Dutch book coherence are, don't worry. But with the addition of one or more empirical assumptions characterizing the particularities of the physical world. So the idea is that we're taking this rule called the Born rule from quantum physics. And we're saying that you can understand that rule as coming from decision theory. Decision theory plus certain empirical assumptions. Okay. So let me explain what that means. A bit of terminology. What is this interpretation, the subjective Bayesian interpretation of quantum theory? Well, there's a lot I could say about it. But for the purposes of this paper. Decision theory. In…