Can Computers Catch You Telling a Lie?

A popular school of thought, dramatized in the recent TV drama Lie to Me, is that a careful study of facial expressions—especially eye movements—tells investigators if a perp is dissembling. Reality is neither as dramatic nor as decisive. Even experienced investigators average only about a 65 percent success rate, according to researchers. Could computers do a better job?
Researchers at the University at Buffalo, The State University of New York (U.B.), claim their video-analysis software can analyze eye movement successfully to identify whether or not a subject is fibbing 82.5 percent of the time. The researchers, who first presented their (still unpublished) results at the 2011 IEEE International Conference on Automatic Face and Gesture Recognition a year ago, believe they have laid the foundation for a more extensive study that will include a larger sample and take into account body language in addition to eye movement to determine whether new technologies can help interrogators in their search for the truth.
The 40 interviews were conducted by Mark Frank, a U.B. professor of communication and a study co-author, and included a diversity in age, gender and ethnicity. Prior to the interview, each subject was given the opportunity to "steal" a check made out to a political party or cause that the volunteer strongly opposed. Afterward, subjects sat down with a retired law-enforcement interrogator. The interviewer first posed conversational questions unrelated to the possible theft, to establish a baseline of normal eye movement, and then asked whether the interviewee had taken the check.