During the pandemic, technology companies sought out their emotion recognition software to monitor workers and even children remotely. Take, for example, a system called 4 Little Trees. The program was developed in Hong Kong and claims to assess children’s emotions while doing classwork. It maps facial features to assign the emotional state of each learner into a category such as happiness, sadness, anger, disgust, surprise and fear. It also measures the motivation and predicts the degrees. Similar tools have been marketed to provide supervision for remote workers. According to one estimate, the emotion recognition industry will grow to US $ 37 billion by 2026.
There is deep scientific disagreement about whether AI can detect emotions. A review from 2019 found no reliable evidence for it. “Technical companies may be asking a question that is fundamentally wrong,” the study concluded (LF Barrett et al. Psychol. Sci. Public interest 20, 1–68; 2019).
And there is growing scientific concern about the use and misuse of these technologies. Last year, Rosalind Picard, co-founder of an artificial intelligence (AI) company called Affectiva in Boston, and head of the Affective Computing Research Group at the Massachusetts Institute of Technology in Cambridge, said she was regulating support. Scholars called for mandatory, careful auditing of all AI technologies employed, along with the disclosure of the findings. In March, a civilian panel convened by the Ada Lovelace Institute in London said an independent, legal body should oversee the development and implementation of biometric technologies (see go.nature.com/3cejmtk). Such oversight is essential to defend against systems driven by what I call the phrenological impulse: to draw erroneous assumptions about internal conditions and abilities from external appearance, with the aim of extracting more about a person than they prefer to to reveal.
Countries around the world have legislation that requires scientific precision in the development of medicines that treat the body. Tools that claim our minds should enjoy at least the same protection. Scholars have been calling for years for federal entities to regulate robotics and facial recognition; it should also extend to emotion recognition. It is time for national regulatory agencies to guard against unproven applications, especially those targeting children and other vulnerable populations.
Lessons from clinical trials show why regulation is important. Federal requirements and subsequent advocacy made much more clinical trial data available to the public and subject to strict verification. It becomes the basis for better policy and public confidence. Regulatory oversight of affective technologies will bring similar benefits and accountability. It can also help set norms for achieving too much by corporations and governments.
The polygraph is a useful parallel. This ‘lie detector’ test was invented in the 1920s and used for decades by the FBI and the U.S. military, with conflicting results harming thousands of people until federal law was largely banned. It was not until 1998 that the U.S. Supreme Court concluded that “there is simply no consensus that polygraph evidence is reliable.”
The psychologist Paul Ekman is a formative figure behind the claim that there are universal facial expressions of emotion. In the 1960s, he traveled the highlands of Papua New Guinea to test his controversial hypothesis that all people display a small number of ‘universal’ emotions that are innate, cross-cultural, and consistent. Anthropologist Margaret Mead disputes this idea early on, saying it discounts context, culture and social factors.
But the six emotions that Ekman describes fit perfectly into the model of the emerging field of computer vision. As I write in my 2021 book Atlas of AI, his theory was accepted because it fits what the tool can do. Six consistent emotions can be standardized and automated on a scale – as long as the more complex issues are ignored. After the terrorist attacks on September 11, 2001, Ekman sold his system to the U.S. Transportation Administration to determine which airline passengers show fear or tension, and that they may be terrorists. It has been strongly criticized because it is not credible and that it is racially biased. However, many of today’s instruments, such as 4 Little Trees, are based on Ekman’s six emotion categorization. (Ekman claims that faces transmit universal emotions, but says he saw no evidence that automated technologies work.)
Yet companies continue to sell software that will impact people’s opportunities without clearly documented, independently audited evidence of effectiveness. Job seekers are judged unfairly because their facial expressions or vocal tones do not match those of employees; students are marked at school because their faces look angry. Researchers have also shown that facial recognition software interprets black faces as having more negative emotions than white faces.
We can no longer allow technologies for the recognition of emotions to be unregulated. It is time for legislative protection against unproven use of these instruments in all areas – education, healthcare, employment and criminal law. These precautions will give more careful science and reject the mythology that internal states are just another data set that can be removed from our faces.