Thursday, September 27, 2007

Nonhuman primates perceive human goals

Hauser and collaborators do a clever experiment to demonstrate that several primates can make inferences about a human experimenters goal that cannot be explained by simple associative learning. This means that our capacity to infer rational, goal-directed action derives from capabilities present in monkeys ~40 million years ago. Here is their abstract and a figure showing the basic idea of the experiment.
Humans are capable of making inferences about other individuals' intentions and goals by evaluating their actions in relation to the constraints imposed by the environment. This capacity enables humans to go beyond the surface appearance of behavior to draw inferences about an individual's mental states. Presently unclear is whether this capacity is uniquely human or is shared with other animals. We show that cotton-top tamarins, rhesus macaques, and chimpanzees all make spontaneous inferences about a human experimenter's goal by attending to the environmental constraints that guide rational action. These findings rule out simple associative accounts of action perception and show that our capacity to infer rational, goal-directed action likely arose at least as far back as the New World monkeys, some 40 million years ago.

Figure: During each trial, an experimenter presented subjects with two potential food containers, performed an action on one, and then allowed the subject to select one of the containers. In the intentional condition, the experimenter reached directly for and grasped the container. In the accidental condition, the experimenter flopped his hand onto the container with palm facing upwards in a manner that appeared, from a human perspective, accidental and non–goal-directed (13). If non-human primates fail to distinguish between intentional and accidental actions when making inferences about others' goals, attending to the mere association of the hand and container, then they should show the same pattern of searching in both conditions—that is, approach the experimenter-contacted container. However, if they distinguish between intentional and accidental actions, then they should selectively inspect the container targeted by the experimenter'sintentional action but not that targeted by accidental action.

No comments:

Post a Comment