First-Person Fairness in Chatbots - Understanding Bias and Fairness in AI Interactions

Watch a 48-minute research lecture from the Simons Institute where Adam Tauman Kalai of OpenAI explores the concept of first-person fairness in chatbot interactions. Delve into how fairness principles apply differently to chatbots compared to institutional decision-making systems, with a focus on fairness toward individual users across diverse applications like resume writing, technical support, and entertainment. Learn about a new methodology for analyzing fairness in open-ended text generation systems and discover how post-training reinforcement learning has been effective in reducing harmful biases in ChatGPT. The presentation shares findings from collaborative research with team members from various institutions, examining the unique challenges of measuring and implementing fairness in conversational AI systems that generate unrestricted text output.