Techmeme Ride Home (Bonus) Building AI We Can Trust With Gary Marcus
Sep 15, 2019
Gary Marcus, author and AI researcher, delves into the pressing need for AI we can trust. He discusses the skepticism surrounding AI, emphasizing how it hasn't yet earned public trust. Marcus shares insights on the limitations of current AI models and the necessity for a hybrid approach that merges data-driven techniques with classical methods. He also explores the competitive AI landscape between the U.S. and China, alongside advancements in AI technology that challenge traditional models, calling for innovative integration.
AI Snips
Chapters
Books
Transcript
Episode notes
Tesla Autopilot Accidents
- Drivers rely too much on Tesla's Autopilot, leading to accidents in unusual situations.
- Current AI struggles with unexpected events, like a semi-truck making a left turn, as seen in fatal Tesla accidents.
AI Lacks Judgment
- AI excels at rote tasks but lacks judgment and deep understanding of the world.
- While AI can make simple decisions, it cannot handle complex judgments requiring balancing various considerations.
AI's Reliance on Pattern Matching
- Self-driving cars rely on pattern matching, making them vulnerable to unfamiliar situations.
- AI cannot intuit solutions; it relies on labeled examples and struggles with novel scenarios.
