Emotion Recognition in Hiring: The Pseudoscience Deciding Who Gets Work

You sit alone in a spare room, laptop propped on a stack of books to get the camera level with your eyes, and you wait for the first question to appear. There is no interviewer. There is a progress bar, a countdown timer, and a small green light beside the webcam that tells you the machine is watching. A prompt fades onto the screen. You have thirty seconds to think and two minutes to answer, and there will be no second take. So you talk. You describe a time you handled conflict, a moment you showed leadership, the reason you want this job, and the whole while you are performing not for a person but for a piece of software that is measuring something you cannot see and were never told about. It is not weighing your words alone. It is parsing the cadence of your voice, the micro-movements at the corners of your mouth, the length of the pauses where you gather your thoughts, the steadiness or flicker of your gaze. From these signals it is assembling a verdict: a number for your conscientiousness, a number for your emotional stability, perhaps a quiet flag against your honesty. You will never see those numbers. You will receive, days later, a templated rejection that thanks you for your interest and wishes you well in your search.

This is the invisible audition, and millions of people now sit through some version of it every year, mostly without realising what is happening on the other side of the glass. The asynchronous video interview, in which a candidate records answers for a machine to score rather than a human to watch, has become a standard gate in high-volume hiring. The market that builds these systems is growing at a clip that industry analysts put in the region of fifteen per cent a year, with one estimate projecting the AI video interview sector to climb from under a billion dollars in 2024 towards nearly three billion within the decade, and the largest vendors process tens of millions of interviews. Most of that scoring is uncontroversial enough on its…