Between Model Misbehavior and AI Harms: Red-Teaming in the Public Interest - Bobinski Lecture

Description:

As generative AI (genAI) systems become more integrated into daily life, concerns about AI safety drive the need for rigorous evaluations. GenAI red-teaming has emerged as both a stress-testing strategy and a way to engage the public in critical discussions. This talk explores the challenges and opportunities of red-teaming, questioning its role in addressing complex AI outputs. Drawing on collaborative research, I examine how AI and society shape each other, who defines “harm,” and why evaluation methods must evolve. I propose red-teaming as an ongoing, collective inquiry that balances experimentation with addressing AI risks.

Location:

UB Libraries Special Collections (420 Capen Hall, North Campus)

Event Date:

-

Presenter Information:

Ranjit Singh, PhD

Ranjit Singh is a senior researcher at Data & Society’s AI on the Ground team, conducting qualitative research for the Algorithmic Impact Methods Lab (AIMLab). His work focuses on evaluating and regulating algorithmic systems’ impact on daily life while promoting research ethics and equity. His research explores AI infrastructures, majority world scholarship, and public policy using qualitative sociology and ethnography. At D&S, he has examined AI’s conceptual vocabulary, algorithmic impact assessments, and datafied governance. Ranjit earned a PhD from Cornell, studying Aadhaar’s role in shaping Indian citizenship and inclusive development.

Register here: