The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Service Cards and ML Governance with Michael Kearns - #610
Jan 2, 2023
In a fascinating discussion, Michael Kearns, a UPenn professor and Amazon Scholar, delves into the vital topics of AI governance and fairness. He describes the innovative service cards introduced at Amazon, emphasizing their holistic approach compared to traditional model cards. Kearns also tackles the ongoing debate surrounding algorithmic versus dataset bias, reflecting on current challenges in fairness within large language models. His insights shed light on the importance of responsibly addressing these issues in AI development.
AI Snips
Chapters
Transcript
Episode notes
Early Machine Learning Career
- Michael Kearns's career began in the late 1980s when machine learning was a niche field.
- He worked at Bell Labs with other machine learning luminaries during a golden era of research.
Responsible AI Focus
- Michael Kearns's research focuses on responsible AI, initially from a technical perspective.
- He now recognizes the need for non-technical solutions like diverse teams and policy input.
Service Cards vs. Model Cards
- AWS launched service cards, inspired by model cards, but for entire services, not individual models.
- This is because AWS AI services often use many models, and users experience the whole system.