The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) “Fairwashing” and the Folly of ML Solutionism with Zachary Lipton - TWIML Talk #285
Jul 25, 2019
Zachary Lipton, Assistant Professor at CMU's Tepper School of Business, dives into the intersection of machine learning and healthcare. He highlights the importance of human expertise in AI decision-making and critiques the concept of 'fairwashing' in tech. The conversation touches on the challenges of applying machine learning in medical contexts, discussing the necessity for robust models that account for real-world complexities. Additionally, Lipton explores the ethical dimensions of algorithmic decision-making and the gap between theoretical fairness claims and practical realities.
AI Snips
Chapters
Transcript
Episode notes
Unplanned Path to Machine Learning
- Zachary Lipton's career path was unplanned, starting in math/economics, shifting to music, then unexpectedly landing in machine learning.
- He credits following his interests and being open to new experiences for his unique interdisciplinary career.
From Startup to PhD
- Despite lacking formal ML training, Zach pursued a PhD, driven by his interest in algorithms and computational thinking.
- He moved to California, worked at a startup, and eventually got accepted into a PhD program.
Limitations of Supervised Learning
- Supervised learning, the dominant approach in ML, assumes future data will resemble historical data, which is often unrealistic.
- Real-world problems involve distribution shifts and feedback loops from ML-driven decisions, making this assumption problematic.