Do robots have goals? How agent cues influence action understanding in non-human primates

The capacity to understand goals and intentions emerges early and universally in humans and is a basic precondition for the interpretation and prediction of others’ actions, be it other humans, animals, or even robots. It is unclear, however, how this goal attribution system is acquired, in particular with regard to the role of prior experience with the actor and visual characteristics that are necessary. In four preferential looking time experiments we examined how familiarity, appearance, and movement of different agents influence the capability of marmosets to perceive the behavior of these agents as goal directed. To this end we compared the monkeys’ reactions to the same goal-directed actions performed by four different agents: a human actor, a conspecific, a monkey-like small robot, and a black box. The results showed that monkeys attributed goals to the human actor, the conspecific, and the robot, but not the box. Thus, the monkeys extended their capacity for goal attribution not only to familiar agents, but also to agents not previously encountered, provided that they had some conspecific-like features. Our results suggest that in non-human primates, the system for goal attribution does not require previous experience with a specific agent or agent-category, as long as it exhibits certain visual characteristics like face, body or legs. Furthermore, the results suggest that the capacity to attribute goals emerged very early during evolution and, at least in marmoset monkeys, does not necessarily require pre-learned associations in order to fulfill its function when dealing with unfamiliar agents.

► Marmoset monkeys attribute intentions to a human actor and unfamiliar monkey-like artificial agents.
► Monkeys’ system for goal attribution cannot be extended to inanimate abstract agents.
► The capacity to attribute goals may have developed early during primate evolution.