ML Pioneer Since 1980: Why AI Still Breaks on Anything New | Tom Dietterich

Tom Dietterich has worked in machine learning since the very first ML workshop in 1980. He explains why today's AI is brilliant until it meets something new, why we still can't guarantee its outputs, and who should be watching it. CHAPTERS 00:00 Where "Machine Learning" Came From 10:06 Neural Networks vs Kernel Methods: The Rise and Fall 20:57 Where AI Research Is Heading Next 28:51 What We Still Don't Understand About Deep Learning 30:12 Why AI Safety Is a Moving Target 32:59 Why Humans Still Need to Watch AI 35:02 AI's Biggest Weakness: Anything New 39:30 Why AI Forgets What It Learned 40:46 When AI Fails, Who's to Blame? 43:17 Who Monitors Superhuman AI? 49:20 AI Agents, Regulation and What Comes Next #AI #MachineLearning #AISafety