New evidence for learning-based accounts of gaze following: Testing a robotic prediction

Gaze following is an early-emerging skill in infancy argued to be fundamental to joint attention and later language. However, how gaze following emerges has been a topic of great debate. The most widely-accepted developmental theories suggest that infants are able to gaze follow only by understanding shared attention. Another group of theories suggests that infants may learn to follow gaze based on low-level social reinforcement. Nagai et al. [Advanced Robotics, 20, 10 (2006)] successfully taught a robot to gaze follow purely through social reinforcement, and found that the robot learned to follow gaze in the horizontal plane before it learned to follow gaze in the vertical plane. In the current study, we tested whether 12-month-old infants were also better at gaze following in the horizontal than the vertical plane. This prediction does not follow from the predominant developmental theories, which have no reason to assume differences between infants’ ability to follow gaze in the two planes. We found that infants had higher accuracy when following gaze in the horizontal than the vertical plane (p = .01). These results confirm a core prediction of the robot model, suggesting that children may also learn to gaze follow through reinforcement learning. This study was pre-registered, and all data, code, and materials are openly available on the Open Science Framework (https://osf.io/fqp8z/)