Companies that rely on facial recognition software to “read” human emotions may run the risk of making ill-informed decisions, according to one of the world’s preeminent experts on the psychology of emotion.
Why? Because of outdated and unreliable science.
Northeastern University’s Lisa Feldman Barrett, a distinguished professor of psychology, warns that many facial recognition technologies ignore the growing consensus that such expressions are not universal. Due to a flawed fundamental assumption, the technology is inaccurate and may further lead to discrimination, says Feldman Barrett.
An Uncommon Stance
Hefty opposition has met Feldman Barrett’s cautionary words. Companies such as Unilever have claimed to save over 100,000 hours of interview time … all in a single year thanks to AI.
American company HireVue developed the cutting-edge technology. Utilized by industry giants like Unilever and Goldman Sachs, the software analyzes playback of video interviews. Scanning candidates’ facial expressions, word choice, and body language, the system cross-references data with traits linked to job success.
More companies are exploring potential applications for AI with successful, cost-saving rollouts already realized. The EU is testing software it claims can detect deception through the analysis of micro-expressions. Amazon touts its facial recognition software, Rekognition, as being able to distinguish between seven fundamental emotions: sadness, anger, confusion, disgust, calmness, happiness, and surprise.
However, Feldman Barrett disagrees. “Our judgment is that [these technologies] shouldn’t be rolled out and used to make consequential decisions about people’s lives,” says the psychologist. She rests her case on published scientific evidence. Several studies confirm the human emotional range is far more complex than scientists originally thought.
According to Feldman Barrett, the idea that all humans express the same, basic seven emotions originated with scientist Paul Ekman. Ekman conducted research in 1960s Papua New Guinea. The result of his findings centered around members of an isolated tribe. The tribespeople selected similar answers to Americans when asked to match pictures of different facial expressions with certain scenarios.
Not So Simple
Despite Ekman’s findings, a growing body of evidence suggests that human emotions encompass a far greater range, both throughout and across cultures.
Feldman Barrett highlights the lack of reliability that results from a simplistic interpretation of facial expressions. For example, “People scowl when they’re concentrating really hard, when you tell a bad joke, when they have gas.” The reasons are complex and often varied. Certain facial recognition technologies, however, may always equate a scowl to … anger.
On the other hand, some expressions thought to be universal may, in fact, not be. In Malaysia, for instance, the expression widely associated with fear is taken as a threat or sign of anger. Variations exist similarly within cultures, while body language and who a person is talking to remain crucial for context.
Feldman Barrett asserts, “We conclusively show across hundreds of studies that this common-sense belief—that each emotion has its own bodily fingerprint—is just false.”
Indeed, it is not so simple when “Variation is the norm.”