Understanding credibility and deception

Can a computer help people tell when someone is lying, and should it? Across a series of studies we built systems that analyze facial expressions, language and prosody as they relate to credibility, and we proposed ethical limits on technology that could read a person's credibility without their knowledge.

From the 'Mental Trespass' framework: where AI that infers truthfulness or hidden thoughts should and should not be allowed.
1st placeIARPA Credibility Assessment Standardized Evaluation (CASE) challenge
9papers across IMWUT, IEEE FG, ACII, IEEE TAC, EMNLP and IEEE TTS
Onlinea framework for recording honest and deceptive conversations at scale

Recording and analyzing deception at scale

The Automated Dyadic Data Recorder lets pairs of people hold honest or deceptive conversations online, so deception can be studied at scale. Follow-on work found facial cues of lying, gender differences in deceptive expressions, and how facial features change the meaning of words.

A mental trespass? The ethics of AI lie detection

Non-invasive AI lie detection could threaten civil liberties. We proposed a framework that separates truth metering from exposing a person's thoughts, and argued for limits on both.

More

← All research