Currently, in human-AI collaboration, intelligent systems model their users as passive sources of data which limits the effectiveness and efficiency of collaboration. The effectiveness of human-human collaboration is shown to be related to the collaborators' ability to model each others' minds. This research project focuses on developing better models for human-AI collaboration by treating the users as agents in a multi-agent learning system. Theory of mind is described as the human ability to attribute beliefs, desires, intentions, and mental states to others. The developed models will be based on the theory of mind, and will be learnt from the interaction data. The project aims to demonstrate that interactive intelligent systems which use theory-of-mind based models of their users are more efficient in terms of data, and perform better. This requires bringing together ideas from sequential decision-making, behavioural economics, and machine learning into developing a mathematical framework which will be the end result of the dissertation

2022

2023