Perception of Artificial Agents


No longer encountered only in science fiction, artificial agents such as humanoid robots and interactive animated characters are rapidly becoming participants in many aspects of social and cultural life. With applications in a range of domains that concern all of us, such as education and healthcare, we need to understand human factors guiding our perceptions of and interactions with these agents.

Our lab studies how humans perceive other individuals and their body movements using psychophysics, neuroimaging and neuropsychological patient studies (e.g., Saygin et al., 2004; Saygin, 2007). We have recently expanded our focus to artificial agents such as robots. Robots can perform body movements, but these motions do not have the dynamics of biological motion. Robots can also have a range of appearances; they can be more or less anthropomorphic. So what are the effects of biological motion and biological appearance in the perception of others? We are currently exploring this question using behavioral and neuroimaging studies.

It may seem like a good idea to make artificial agents look as human-like as possible. However, we soon encounter the "uncanny valley": As an agent's appearance is made more human-like, people's disposition toward it becomes more positive and empathetic, until a point at which the increasing human-likeness leads to the agent being considered repulsive, disturbing, or "zombie-like". This phenomenon was first described in robotics (Mori, 1970), although there are earlier references to the uncanny or unheimlich concept (Freud, 1919; Jentsch, 1906).

Our research is helping to demystify the "uncanny valley" using methods from experimental psychology and cognitive neuroscience. We reframe the problem in terms of processing in the brain's action perception system. More generally, we aim to both improve our understanding of how the human brain enables action perception and social cognition, and to help develop interactive agents that are well-suited to their application domains, as well as to the brains of their creators.

Find out more at: http://sayginlab.org/