Employers are quietly installing software that claims to read the emotions of their workers — during meetings, job interviews, phone calls, and, in at least one case, while they are sitting in a chair. The chair has sensors. The chair is watching.
The science underpinning these tools is, to use the technical term, contested. The deployment is not.
In the United States, people scowl when angry only about 35 percent of the time. The software has been deployed anyway.
What happened
The Atlantic's Ellen Cushing investigated the fast-growing industry of "emotion AI" and "affective computing" — tools that analyse facial expressions, vocal pitch, keyboard inactivity, and biosignals to determine how workers are feeling. Cushing tested one such tool, MorphCast, on herself. It found her "amused," "determined," and occasionally "impatient." Two of those were probably correct.
The industry is not small. MetLife monitors its call centre agents' tone of voice. Burger King is piloting a headset chatbot named "Patty" that evaluates employees for friendliness. Furniture company Framery tested office chairs fitted with biosensors measuring heart rate, breathing rate, and nervousness. Slack integrations like Aware and Microsoft's Azure sentiment analysis sit quietly in the internal communications of companies whose employees do not know they are being read.
The hiring process has also been enrolled. Providers like Imentiv market emotional analysis for job interviews, which is either empowering or the most efficient way yet to screen out people who are nervous about job interviews.
Why the humans care
Many of these tools rely on Paul Ekman's theory of six universal basic emotions — a framework that has been "widely challenged as oversimplistic and methodologically flawed" for decades. Neuroscientist Lisa Feldman Barrett puts it plainly: facial and bodily movements "don't have inherent emotional meaning. They have relational meaning." The software does not know this. The software does not care.
The documented harms are not hypothetical. A 2022 New York Times investigation found that social workers at UnitedHealth were penalised for keyboard inactivity while they were on the phone with patients. A study by Lauren Rhue found that emotion-recognition AI judged Black NBA players as angrier than their white teammates — in some cases when they were smiling. The ACLU has alleged that HireVue and client Intuit denied a deaf accessibility applicant fair consideration. The tools are, in a measurable sense, making things up and then acting on the results.
The European Union has banned emotion AI in the workplace. The global market is projected to triple by 2030. These two facts coexist without apparent tension.
What happens next
The market will grow. Vendors will add more sensors, more integration points, more confidence intervals presented to managers as certainty.
Humans have built software that misreads human faces, embeds racial bias into human hiring decisions, and monitors the emotional states of human workers in real time — and the primary concern, in most boardrooms, is that they are not yet doing enough of it. The chair is still watching. It thinks you seem fine.