Birhane's PhD thesis catalogues 67 values embedded in ML

Abeba Birhane's PhD thesis, posted to arXiv, argues that machine learning's core predictive premise is "impossible in principle" and names 67 values embedded in ML research.

"Machine learning (ML) and artificial intelligence (AI) tools increasingly permeate every possible social, political, and economic sphere; sorting, taxonomizing and predicting complex human behaviour and social phenomena," Birhane writes in the abstract. "They remain opaque and unreliable, and fail to consider societal and structural oppressive systems, disproportionately negatively impacting those at the margins of society while benefiting the most powerful."

The thesis audits large-scale datasets behind current AI, saying they "embed societal historical and structural injustices." It reviews the "historical and cultural ecology from which AI research emerges." It studies "the seemingly neutral values of ML research," and puts forward 67 prominent values underlying the field.

Birhane's prediction claim is the sharpest line in the abstract: that "predicting complex behaviour with precision is impossible in principle." The thesis closes by putting forward a framework for ML failures and "alternative ways forward," without listing which alternatives in the abstract itself.

Two of the AI researchers we track posted the thesis link to our feed recently, pulling an older piece of FAccT-adjacent work back into current circulation.