Infants' teleological and belief inference: A recurrent connectionist approach to their minimal representational and computational requirements

This paper explores the minimal representational and processing requirements for teleological and mentalistic inferences. These inferences are already present in 6- to 12-month-old infants when they judge the goals of moving agents like inanimate shapes, without any cue about human body motion. This precludes the mirror system as a potential explanation for these early inferences. I propose a recurrent connectionist network that represents a spatial grid in which objects are moving along trajectories that avoid potential obstacles. This network can simulate young infants' (a) teleological sensitivity to goals and obstacles blocking these goals, as well as their (b) mentalizing sensitivity, like attributing to an individual a false belief that differs from the child's reality. The model demonstrates how infants choose the most rational or efficient way through the grid towards the goal, a rational choice that emerges from prior familiarization with and generalization from trajectories and situational obstacles. In addition, mentalizing is accomplished by explicitly representing in the network an individual who watches (or not) where an object moves and keeping track of this individual's experiences and knowledge. This model helps to clarify that these rudimentary characteristics seem to be essential ingredients for an early teleological and mentalizing system.