Inioluwa Deborah Raji

Inioluwa Deborah Raji studies the distance between a model’s measured performance and its consequences in use. Across auditing, multilingual testing and public policy, she has pressed for evaluation systems that examine institutions, deployment conditions and the people affected—not scores in isolation.

Inioluwa Deborah Raji has spent much of her career asking what an AI evaluation is actually for. A benchmark may describe performance under controlled conditions; an audit must connect evidence to responsibility, remedy and real-world decisions. As a researcher at the University of California, Berkeley and an Academic Fellow at the Leadership Conference on Civil and Human Rights, Raji works across computer science, institutional design and public accountability. In 2025, she contributed to the writing group behind the first International AI Safety Report and to California’s expert report on frontier-model policy. Her peer-reviewed work that year examined how predictive systems interact with human decision-makers and proposed ways to aggregate individual reports into evidence of systemic harm. In 2026, multilingual functional evaluation co-authored with Victor Ojewale and Suresh Venkatasubramanian showed that model performance could fall unevenly when static tests were converted into more demanding functional tasks across six languages. The result reinforces a principle running through Raji’s work: an evaluation is credible only when its construct, context and consequences are clear. Her influence lies in moving accountability from ethical aspiration toward inspectable practice.