Talk Date and Time: June 18, 2024 at 6:30 pm - 7:15 pm EST followed by 15 minutes of Q&A in IRB-5105 and on Google Meet
Topic: Hypothetical Minds: Scaffolding Theory of Mind for Multi-Agent Tasks with Large Language Models
Abstract:
Multi-agent reinforcement learning (MARL) methods struggle with the non-stationarity of multi-agent systems and fail to adaptively learn online when tested with novel agents. Here, we leverage large language models (LLMs) to create an autonomous agent that can handle these challenges. Our agent, the Hypothetical Minds model, scaffolds the decision-making process around an LLM-mediated Theory of Mind module. This module enables the agent to synthesize hypotheses about its opponent’s strategy and goals. It then iteratively evaluates and refines these hypotheses, drawing on its memory of prior events, and leveraging intrinsic rewards derived from the LLM’s own predictions. Hypothetical Minds significantly improves performance over RL baselines on the challenging Running With Scissors scenario in the Melting Pot MARL benchmark. In contrast to RL methods that are trained with a large number of samples, the Hypothetical Minds agent succeeds in a zero-shot fashion, learning to identify and exploit strategies purely from in-context learning.
Bio:
Logan Cross is a postdoctoral fellow at Stanford University in the Department of Computer Science under the joint supervision of Prof. Dan Yamins and Prof. Nick Haber. His research combines interdisciplinary approaches to understand how intelligence emerges in humans and can be built in machines. This work is centered on modeling complex decision-making processes in dynamic, naturalistic tasks with cutting-edge deep learning tools, including deep reinforcement learning and large language model based agents. Currently, he is investigating how to endow artificial agents with theory of mind and social cognitive processes such that they can predict and interact with other agents in multi-agent/social domains. He received his Bachelor’s in Neuroscience at University of Southern California and completed my PhD in Computation and Neural Systems at Caltech with John O’Doherty.