Emotion AI researchers say exaggerated claims give their work a bad name

Emotion recognition, a subset of affective computing, is still a nascent technology. As AI researchers have tested the limits of what we can and cannot quantify about human behavior, the underlying science of emotions has continued to develop. There are still several theories, for example, about whether emotions can be discretely distinguished or fall on a continuum. Meanwhile, the same phrases can mean different things in different cultures. In July, a meta-study concluded that it is not possible to judge emotion by simply looking at a person’s face. The study was widely covered (including in this post), often with headlines suggesting that “emotion recognition cannot be trusted.”

Emotion recognition researchers are already aware of this limitation. Those we spoke to were careful to make statements about what their work can and cannot do. Many have pointed out that emotion recognition cannot actually assess an individual’s internal emotions and experience. It can only estimate how that individual’s emotions might be perceived by others, or suggest general population-based trends (such as one movie getting, on average, a more positive reaction than another). “No serious researcher would claim that you can analyze opposite action units and then you actually know what people are thinking,” says Elisabeth André, an affective computing expert at the University of Augsburg.

The researchers also note that emotion recognition involves much more than just looking at someone’s face. It can also involve observing body posture, gait, and other characteristics, as well as using biometric and audio sensors to collect more holistic data.

Such distinctions are beautiful but important: they disqualify apps like HireVue, which claim to assess an individual’s inherent competence, but support others, such as technologies aimed at making machines smarter collaborators and companions for people. humans. (HireVue did not respond to a request for comment.) A humanoid robot can smile when you smile, a mirror action that humans often use to make interactions feel more natural. A wearable device could remind you to rest if it detects levels of cortisol above baseline levels, the body’s stress hormone. None of these apps require an algorithm to assess your private thoughts and feelings; they only require an estimate of an appropriate response to cortisol levels or body language. They also don’t make major decisions about an individual’s life, unlike untested hiring algorithms. “If we want computers and computer systems to help us, it would be positive if they had some idea of ​​how we feel,” says Nuria Oliver, chief data scientist at the nonprofit DataPop Alliance.

But much of that nuance is lost when emotion recognition research is used to create lucrative business applications. The same stress-monitoring algorithms in a wearable device could be used by a company trying to make sure you’re working hard enough. Even for companies like Affectiva, founded by researchers who talk about the importance of privacy and ethics, boundaries are hard to define. He has sold its technology to HireVue. (Affectiva declined to comment on specific companies.)

A call for regulation

In December, research institute AI Now called for a ban on emotion recognition technologies”in important decisions that impact people’s lives.” It is one of the first calls to ban a technology that has received less regulatory attention than other forms of artificial intelligence, even though its use in job screening and classrooms could have serious consequences. effects.

On the other hand, Congress has just held its third facial recognition hearing in less than a yearand he has become an issue in the 2020 election. Activists are working to boycott facial recognition technologies and several representatives acknowledge the need for regulation in the private and public sectors. For affective computing, there have not been as many dedicated campaigns and task forces, and attempts at regulation have been limited. An Illinois law regulation of AI analysis of job interview videos went into effect in January, and the Federal Trade Commission has been asked to investigate HireVue (though it’s unclear if it intends to do so).

Although many researchers think a ban is too broad, they agree that a regulatory vacuum is also harmful. “We have clearly defined processes to certify that certain products that we consume – whether it’s the food we eat, whether it’s the medications we take – we can take them safely, and they really do everything that they claim to do,” says Olivier. “We don’t have the same processes for technology.” She thinks that companies whose technologies can significantly affect people’s lives should prove that they meet a certain level of security.

Rosalind Picard, a professor at the MIT Media Lab who co-founded Affectiva and another affective computing startup, Empatica, echoes that sentiment. For an existing regulation model, it points to the Employee Polygraph Protection Act limit the use of polygraphs, which she says are essentially affective computer technology. For example, the law prohibits most private employers from using polygraphs and does not allow employers to request polygraph test results.

She suggests that any use of these technologies should be voluntary and that companies should be required to disclose how their technologies have been tested and what their limitations are. “What we have today is that [companies] can make these outrageous claims that are just plain wrong, because right now the buyer is not that well educated,” she says. “And we shouldn’t require buyers to be well educated.” (Picard, who says he left Affectiva in 2013, does not support HireVue’s claims.)

For her part, Meredith Whittaker, research scientist at NYU and co-director of AI Now, insists on the difference between research and commercialization. “We’re not challenging the whole field of affective computing,” she says. “We particularly denounce the unregulated, unvalidated and scientifically unfounded deployment of commercial affect recognition technologies. Commercialization is harming people right now, potentially, because it makes claims that determine people’s access to resources.

A ban on the use of emotion recognition in applications such as job screening would help prevent commercialization from overtaking science. Stop deploying technologies first, she says, then invest in research. If the research confirms that the technologies work as the companies claim, then consider relaxing the ban.

However, further regulations would still be needed to keep people safe: there is ultimately more to consider, argues Whittaker, than mere scientific credibility. “We need to ensure, when these systems are used in sensitive contexts, that they are contestable, that they are used fairly,” she says, “and that they do not lead to an increase in power asymmetries between the people who use them and the people they are used on.

James G. Williams