FHI’s Owain Evans just released an online book describing and implementing models of rational agents for (PO)MDPs and Reinforcement Learning in collaboration with Andreas Stuhlmüller, John Salvatier, and Daniel Filan. The book aims to educate its readers on the creation of richer models of human planning, capturing human biases and bounded rationality.
Leading from one-shot decision-problems to a complex sequence of problems, the book aims for a generative model predicting human behaviour based on given preferences and beliefs, with unknown features in the environment.
The production of the book was supported by a grant from the Future or Life Institute, and a grant from Alexander Tamas.
If this subject area is of further interest to you, FHI is currently advertising internship positions in AI safety and reinforcement learning; we especially encourage female candidates to apply.