A stimulus-computable rational model of visual habituation in infants and adults

Abstract

How do we decide what to look at and when to stop looking? Even very young infants engage in active visual selection, looking less and less as stimuli are repeated (habituation) and regaining interest when novel stimuli are subsequently introduced (dishabituation). The mechanisms underlying these looking time changes remain uncertain, however, due to limits on both the scope of existing formal models and the empirical precision of measurements of infant behavior. To address this, we developed the Rational Action, Noisy Choice for Habituation (RANCH) model, which operates over raw images and makes quantitative predictions of participants looking behaviors in a classic visual habituation paradigm. In a series of pre-registered experiments, we exposed infants and adults to stimuli for varying durations and measured looking time to familiar and novel stimuli. We found that these data were well captured by RANCH. Using RANCHs stimulus-computability, we also tested its out-of-sample predictions about the magnitude of dishabituation in a new experiment in which we manipulated the similarity between the familiar and novel stimulus. By framing looking behaviors as rational decision-making, this work identified how the dynamics of learning and exploration guide our visual attention from infancy through adulthood.