Research Areas

Eliciting interpersonal judgments

Human Perception of Facial Features and Proportions

The uncanny valley hypothesis predicts increased sensitivity to human norms in figures approaching human likeness. Norms are the standards by which we evaluate the appearance, performance, and interactivity of entities in our environment, and they give meaning to circumstances based on an individual's perspective. They derive from biological, developmental, and cultural factors, among others.

An entity may seem uncanny or eerie because it violates norms related to appearance, movement quality, or contingency during interaction. In considering appearance, presumably, people should be more sensitive to deviations from perceptual norms in humanlike entities. The purpose of this study is to determine whether human beings vary in their sensitivity to the facial proportions of people, robots, and humanlike characters. It specifically considers what facial proportions are ideal and what ranges are acceptable with respect to a figure's degree of human likeness.


Robert D. Green, Clinton T. Koch, Sandosh Vasudevan, Karl F. MacDorman

Rapid Evaluations of Human Faces

Images of realistic androids and computer-generated people can elicit powerful negative reactions from human observers. The neural origins of this uncanny valley effect are unknown. Moreover, the extent to which uncanny faces trigger judgments more commonly associated with perception or cognition is unclear.


Himalaya Patel, Karl F. MacDorman

The Influence of Human Presentational Factors on Ethical Decision-Making (Persuasion)


Joseph Coram, Adam Burton, Karl F. MacDorman

Implicit Associations about Robots and People


Sandosh Vasudevan, Karl F. MacDorman

Implicit Associations about Human and Machine Speech

The implicit association test (IAT) measures differential associations of two concepts with an attribute. Measurement is implicit, based on time-on-task performance. Performance is faster if a more strongly associated attribute-concept pair has the same response key than if a less strongly associated attribute-concept pair does. The IAT is intended to measure automatic evaluations that may differ from self-reported preferences. To the extent that these preferences are exclusively subconscious, their causes are not available to introspection. The IAT was developed by Anthony G. Greenwald (University of Washington), Mahzarin Banaji (Harvard University), and their colleagues.


Wade J. Mitchell, Karl F. MacDorman

Other topics

  • Mining telepresence data for contextualized interactions
  • Developing an android baby for eldercare
  • Studying eye contact
  • Improving motion quality