Reinforcement Learning for Cooperative AI

Résumé: Reinforcement learning (RL)—both single- and multi-agent—now underpins much of cooperative AI, whose aim is to create agents that can collaborate with humans and with one another in open-ended tasks. This talk begins with a concise overview of the field and my prior work, then turns to two current research thrusts: (1) multi-agent RL for human–AI cooperation and (2) model-based RL in complex stochastic processes with transformers and variational flow matching. I will close by highlighting how these directions intersect with MIAT’s interests and outlining concrete opportunities for collaboration.