Beyond the Truth Machine: Emerging Technologies for Lie Detection

Deception is a universal phenomenon in human communication, but it is also a source of moral indignation that is sanctioned in all cultural systems. Therefore, for centuries, there have been several techniques to detect lying. One of the first methods of deception detection comes from ancient China, ca. 1,000 BC. A suspect would be made to chew dry rice while being questioned. Immediately after, the rice was examined. If it was dry, the suspect was assumed to be guilty of fraud. The underlying assumption was that a lie is accompanied by “honest” signals of the body (salivation), which can reveal the anxiety experienced by the suspect. Interestingly, the principle on which this ancient method relies—the somatic expression of emotions accompanying deceptive communication—is still reflected in contemporary, machine-based lie detection techniques. The scientific groundwork for the development of such devices was carried out by famous Italian philosopher and psychologist Vittorio Benussi, who first discovered that lying causes an emotional change within a subject that results in detectable respiratory changes that are indicative that the person is not telling the truth. Drawing on these findings, in 1921, John Augustus Larson, a medical student and a police officer with the Berkeley Police Department in Berkeley, California, invented the first polygraph. The device worked by simultaneously measuring changes in blood pressure, pulse, and respiration in response to questioning. A few years later, Larson's assistant, Leonarde Keeler, perfected the device by making it portable and including the measurement of galvanic skin response. Thanks to the simplification of the technology, the use of the polygraph became widespread. However, the scarcity of academic, peer-reviewed studies carried out in the following years generated mounting doubts on the reliability of this instrument, which culminated in the following conclusion by the empirical review carried out in 1965 by the U.S. Committee on Government Operations: “There is no lie detector, neither man nor machine. People have been deceived by a myth that a metal box in the hands of an investigator can detect truth or falsehood.” ¹ Despite this and other pitfalls, the applications of the polygraph have continued to grow, and more recent tests have produced more convincing evidence on its reliability, ranging between 81% and 91%. ²

The emergence of new technologies, such as artificial intelligence, may further extend the potential of lie deception beyond traditional methods based on face-to-face interviews and polygraphs. For example, in a recent study published on arXiv, ³ a team of researchers from the University of Maryland and Dartmouth College reported the development of an AI system called Deception Analysis and Reasoning Engine (DARE; https://doubaibai.github.io/DARE/), which has been designed to automatically detect deception in real-life courtroom trial videos. Specifically, the AI system was trained to recognize five micro-expressions (frowning, eyebrows raising, lip corners turning up, lips protruded, and head side turn) that are regarded as behavioral proxies of lying. After the training, DARE was then tested on whether it could detect deceptive communications. Results showed that the AI system reached an accuracy of 87% when evaluated on subjects who were not part of the training set. However, when combined with human annotations of micro-expressions, the accuracy improved to 92%.