A reinforcement learning model of social referencing

Abstract

We present a novel computational model of social referencing. The model replicates a classic social referencing experiment where an infant is presented with a novel object and has the choice of consulting an adultpsilas informative facial expression before reacting to the object. The infant model learns the value of consulting the adultpsilas facial expression using the temporal difference learning algorithm. The model is used to make hypotheses about the reason for a lack of social referencing found in autistic individuals, based on an aversion to faces. Comparisons are made between this reinforcement learning model and a previous model based on mood contagion.