Maximising expected outcomes—framed as expected utility in behavioural science and expected reward in reinforcement learning—has long dominated models of decision-making. In this seminar, Ollie Hulme (DRCMR) asks if people maximise time averages instead. He presents data showing that time-average models better account for both neural and behavioural data in risky-choice experiments, with participants operating surprisingly close to the theoretical optimum. Ollie also discusses arguments for the biological and cognitive plausibility of time averaging as well as its consilience with a diversity of theories in psychology and neuroscience.

The seminar was hosted by Emilie Rosenlund Soysal (London Mathematical Laboratory).