With a game show as his guide, researcher uses AI to predict deception

Utilizing information from a 2002 recreation present, a Virginia Commonwealth University researcher has taught a pc how one can inform in case you are mendacity.

“Human behaviors are a rich source of deception and trust cues,” mentioned Xunyu Chen, assistant professor within the Division of Info Methods in VCU’s College of Enterprise. “Using [artificial intelligence methods], akin to machine learning and deep learning, can higher exploit these sources of data for decision-making.”

In one of many first papers that examine high-stakes deception and belief quantitatively—”Trust and Deception with High Stakes: Evidence from the ‘Friend or Foe’ Dataset” appeared in a current situation of Determination Help Methods—Chen and his group use a novel dataset derived from an American recreation present, “Friend or Foe?” which relies on the prisoner’s dilemma. That game theory explores how two folks may gain advantage from cooperating, which is difficult to coordinate, or endure from failing to take action.

Lab experiments which have been generally used to review belief and deception have limitations when it comes to realism and generalizability. In contrast with low-stakes fictitious instances, high-stakes deception present in recreation exhibits calls for larger cognitive assets for behavioral administration. Additionally, the numerous achieve or punishment related to a high-stakes choice may trigger stronger emotional and behavioral variation in cues akin to facial, verbal and motion fluctuations.

“We found multimodal behavioral indicators of deception and trust in high-stakes decision-making scenarios, which could be used to predict deception with high accuracies,” Chen mentioned. He calls such a predictor an automatic deception detector.

This analysis extends the understanding towards deception and belief behaviors that would result in substantial penalties from a scientific and quantitative perspective. Researchers and practitioners can use findings from this analysis to investigate human behaviors in high-stakes eventualities, akin to presidential debates, enterprise negotiations and court docket trials, to foretell deception and the safety of self-interest.