PERSON
Abeba Birhane
The cognitive scientist who opened the black boxes feeding the world’s largest AI models and found, inside them, a vast unaudited inheritance of human prejudice—demonstrating through empirical audit rather than argument that scale does not dilute harm, it concentrates it.
Abeba Birhane came to artificial intelligence from an angle almost nobody else takes: through cognitive science, by way of the embodied and enactive tradition that treats mind as an activity of a living body coping with a world rather than computation running on a chip. That starting point determined what she noticed when she turned to the systems marketed as artificial minds. What she noticed, repeatedly, was that the field had skipped a step. It was building tools to sort, classify, and predict human beings at planetary scale while almost nobody was checking what those tools were trained on. Her landmark audits of
large-scale training datasets—exposing racist and misogynistic labels in ImageNet and finding that hateful content in LAION grew measurably as the dataset scaled from 400 million to two billion samples—converted diffuse worry into documented fact and forced one canonical resource offline. But her contribution is not merely forensic. Her framework of
relational ethics