Early roots of information-seeking: Infants predict and generalize the value of information

Abstract

Humans face the challenge of making sense of a complex world. Learning where to find information is crucial to filter through the abundance of stimuli, distinguish relevant from irrelevant sources, and optimize our learning. Here, we examined the developmental roots of information-seeking by testing whether 8-month-old infants can predict where to find information. We presented infants with visual cues indicating whether they will later receive information about the location of a rewarding stimulus. We analyzed the dynamics of pupil dilation when the cues were presented, but before the actual information was delivered. By combining additive Bayesian models with reinforcement learning, we show that infants learn to successfully predict what cues have a greater informational value and that they generalize these predictions to novel cues that share the same perceptual features. These results reveal the fundamental learning processes that support information-seeking from early in life.