Augmenting human decision-making and learning via physiologically-informed AI

About the Project

Recent advances in machine learning, particularly the integration of deep learning and reinforcement learning into deep reinforcement learning (DRL), have created powerful AI systems capable of learning state–action policies in complex environments. However, major challenges remain in integrating AI agents into teams with humans, where effective collaboration requires sensitivity to human cognitive, emotional, and physiological states.

This project will develop a closed-loop, physiologically informed AI platform that uses neural, systemic physiological, and affective signals—such as EEG, arousal, and stress markers—to infer latent cognitive and emotional states during naturalistic decision-making tasks. These inferred (latent) states will be detected online through a closed-loop brain–computer interface and converted into rich reinforcement signals for training DRL agents. By dynamically learning from these signals, the AI agent will begin to model human intentions – an emerging form of machine “theory of mind” – and ultimately help inform upcoming human behaviour. The ultimate goal is to close the human–AI interaction loop: enabling the AI agent to feed predictions and adaptive support back to human users in real time, thereby augmenting decision-making, performance, and social coordination.