A normative account of human temporal structure learning

Abstract

People rapidly recalibrate their expectations about the world in the face of surprising observations. This recalibration should depend on the temporal structure of the environment, however how people should and do learn temporal structures remains unknown. To examine this gap, we developed a Bayesian model that infers temporal structure of the environment directly from observations and makes accurate predictions across qualitatively different environments. We tested predictions of our model in an online behavioral study in which participants predicted outcomes generated according to various temporal structures, and demonstrated that people learned to exploit these structures in a qualitatively similar manner to that of our structure learning model. Furthermore, we show that normative structure learning accounts for puzzling asymmetries previously observed in sequential effects of temporal structures on human learning. Taken together our model and empirical data provide the first general account of how people use temporal structures to interpret unexpected observations in the service of learning.