Perceptual Crossing

What makes an interaction social?

Historically, investigating social cognition has been focused on studying the effects on an individual. From fMRI studies attempting to tie neural correlates to social action, to observing changes in behavior in different social circumstances, much of the previous research on social cognition has focused more on how it affects the individual. An alternative approach is to shift the focus toward investigating the dynamics of the interaction process itself.

Earlier work in developmental psychology began this line of reasoning by asking a deceivingly simple question – at what point in an infant’s development does it begin to distinguish “live” social interactions with its mother? In the original 1985 experiment, Murray and Trevarthen would place infants in front of a TV monitor that would either play a live video of their mother actively engaging with them through the screen, or a previously recorded video of their mother . They found that the infants were able to tell the difference between the two scenarios – coordinating with their mothers when the interaction was live and becoming upset when the interaction was previously recorded. This was more systematically investigated and replicated in subsequent work that also more specifically defined the infant’s sensitivity to “social contingency” , i.e. that there was something necessary about the high degree of sensitivity within the mutual feedback between social agents.

Within the field of artificial life, we often implement minimal instantiations of life-based dynamics to test a hypothesis. The Perceptual Crossing task, introduced in by Auvray, Lenay, and Stewart in 2009, applies that technique to investigation of social cognition . In this work, they recruited human participants and set them in front of a computer with minimal information on the screen. The study found that even with extremely limited information, the participants were able to differentiate between other participants, mobile objects, or fixed objects. They found also that the participants preferred to interact with the the other participants even within the limited information space.

This task was then translated into non-human agents in several artificial life works: