According to Lisa Feldman Barrett, a psychology professor at Northeastern University, Artificial Intelligence (AI) systems that companies claim can “read” facial expressions are based on outdated and “outdated” science. She also said that these systems tend to ignore a large number of factors that undermine the notion that basic facial expressions are the same across cultures. As a result, AI-based systems are considered inaccurate, unreliable, and biased. Some of them are already being deployed in real-world environments.
In addition, AI systems are being developed for an ever-increasing number of applications. In late 2019, Unilever said that by developing such software, which was used to analyze video interviews, it saved 100,000 hours of work that would have been required by human resources to perform this function. The Artificial Intelligence (AI) system, developed by HireVue ,“scans” the facial expressions, body language and word choice of candidates and links them to characteristics that they believed were associated with job success. Amazon also claims that its own facial recognition system, called “Recognition”, has the ability to detect seven basic emotions. These are happiness, sadness, anger, surprise, disgust, calm and confusion.
Notably, the EU is testing software that could detect fraud by analyzing microexpressions to help improve border security.
Feldman Barrett, speaking at the annual meeting of the American Association for the Advancement of Science in Seattle, said the idea of “universal” facial expressions for emotions such as happiness, sadness, fear, anger, surprise and disgust was particularly popular in the 1960s. This was after an American psychologist, Paul Ekman, conducted research in Papua New Guinea. The research showed that members of an isolated tribe gave Americans similar answers when asked to match photos of people displaying facial expressions in different scenarios. One such scenario was “Bobby’s dog has died.”
However, a large body of evidence has shown that beyond these basic stereotypes there is a huge range in how people express their emotions, both within their own culture and in their interaction with other foreign cultures.
In Western cultures, for example, people frown only about 30 percent of the time when they are angry. This means they “move” their faces in different ways about 70 percent of the time. However, according to Feldman Barrett, these results are not reliable, since people can frown even when they are not angry. For example, they can frown when they are concentrating and focused on something or when they hear a bad joke.
At the same time, regarding the expression that is supposed to be common to everyone in moments of fear, threat or anger, she said that there are large variations within cultures in terms of how people express their emotions. She also stressed that it is very important to take into account body language and who is speaking. Finally, Feldman Barrett underlined that Artificial Intelligence (AI) systems are trained to a large extent on the basis that everyone expresses their emotions in the same way, while at the same time very powerful technology is used to answer simple and simplistic questions.
