原文
Employers are increasingly using "emotion AI" software to monitor and analyze workers' feelings during meetings, customer calls, and job interviews, a report in The Atlantic shows. Experts say these tools rely on flawed science, can exhibit racial bias, and unfairly penalize employees. The EU has banned emotion AI in the workplace, but the global market is projected to triple by 2030, raising serious concerns about invasive worker surveillance. A feature in The Atlantic by Ellen Cushing explores how software that claims to read emotions using AI is quietly becoming a fixture of everyday work life. Cushing tried out the service MorphCast on herself: the program analyzed her facial expressions during a meeting with her boss and determined she was generally "amused," "determined," and "interested" - though occasionally "impatient." The self-experiment serves as a jumping-off point for a broader look at a fast-growing industry operating under labels like "emotion AI" and "affective computing." These products analyze everything from job interview video to call center audio to chat transcripts. MetLife monitors its call center agents' pitch and tone of voice, Burger King is piloting a headset chatbot called "Patty" that evaluates employee interactions for friendliness, and furniture maker Framery has tested office chairs equipped with biosensors that can measure heart rate, breathing rate, and nervousness. Slack integrations like Aware and Microsoft's Azure offer sentiment analysis of internal communications, while providers like Imentiv market emotional analysis tools for the hiring process. Ad In her report , Cushing also questions the scientific foundation of these tools. Many products draw on Paul Ekman's theory of six basic emotions, which has been "widely challenged as oversimplistic and methodologically flawed" for decades. The neuroscientist Lisa Feldman Barrett is quoted as saying that movements "whether it's on your face or in your body or the tones that you emit, don't have inherent emotional meaning. They have relational meaning." In the U.S., she notes, people scowl when angry only about 35 percent of the time. A study by Lauren Rhue, the article reports, found that emotion-recognition AI judged Black NBA players to be angrier than their white teammates, "even, in some cases, if they were smiling." Ad DEC_D_Incontent-1 The questionable science and unchecked surveillance are already causing real harm. According to a 2022 NYT investigation , social workers at UnitedHealth were downgraded for keyboard inactivity while they were talking to patients. The ACLU has alleged that the screening platform HireVue and its client Intuit denied a deaf accessibility-team employee a promotion, with the rejection email advising her to "practice active listening." Both HireVue and Intuit have disputed the claims. The EU has already banned this kind of emotion AI in the workplace under the AI Act , with exceptions only for medical and safety purposes. MorphCast responded by moving its headquarters from Florence to the Bay Area. The global market is expected to triple to nine billion dollars by 2030, according to an estimate cited in The Atlantic. Ad Cushing closes with a dystopian twist: more troubling than flawed systems would be a future where the software actually works - and workers would have, on top of their actual job, "the work of making the emotion robot think that I'm sufficiently cheerful." Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.