Computational Modelling of Infant Gaze Following in Cluttered Environments and Reduced Caregiver Gaze Reliability

Humans’ ability to share attention with others plays an important role in social competence, language acquisition, and action understanding. Gaze following, an indicator of shared attention, occurs when an individual shifts their gaze from looking at another individual to the object that the other is focused on. Previous work showed that infants’ gaze following can be formulated using a reinforcement learning model, in a simplified environment with one visual target. However, real-world infants’ social interactions are typically far more complex, involving multiple competing stimuli and unpredictable attentional demands. To have a more ecologically valid context compared to previous work, we extended the established reinforcement-learning model for infants’ gaze following by introducing visual clutter consisting of multiple visual objects and by varying the reliability of the caregiver’s gaze direction. The extended version of the model showed that gaze following still develops in the presence of distracting objects and that variability in caregiver gaze reliability substantially affects this development. Our findings lay the groundwork for further ecologically valid computational studies and can be adapted to simulate conditions in which the emergence of gaze following may be disrupted by atypical developmental processes.