Zoom URL: https://mit.zoom.us/j/96991479605

Humans flexibly choose appropriate approximate solutions to planning problems, which are notoriously hard to solve exactly. Which mental algorithms do people use to achieve this? I will present a behavioral Maze Search paradigm to study sequential decision-making under uncertainty in a context of a spatial search, and present computational analysis of human behavior in this task. This paradigm abstracts natural behaviors that people perform in daily life, such as, searching for lost keys, shopping, or looking for an object in a cluttered office.

Using this paradigm I will demonstrate how (1) a variety of descision-making principles studied in the context of simple laboratory tasks, such as bandits and gambles, can generalize to naturalistic spatial behaviors; (2) model-based prediction of individual behaviours in spatial search are significantly improved by combining utility maximization, utility discounting, and probability scaling; (3) Failure cases of utility maximizing models provide important insights into computational principles and representations that people may use for planning.