There Auto Be A Law Artificial Intelligence and Auto Safety with Phil Koopman - Part 1
Jun 26, 2025
Phil Koopman, a professor at Carnegie Mellon University with expertise in autonomous vehicle safety, discusses the intersection of AI and auto safety. He shines a light on the common misconceptions about AI, emphasizing its limitations and potential errors in self-driving systems. The conversation navigates the legal hurdles in updating frameworks for autonomous vehicles, alongside the challenges of maintaining safety in AI-driven cars. Through real-world examples, Koopman distinguishes the capabilities of human drivers from robotic systems, urging for improved standards in safety engineering.
AI Snips
Chapters
Transcript
Episode notes
Computers Fail Differently Than Humans
- Computers make different, sometimes inhuman, mistakes compared to humans.
- No one knows whether AI will overall be safer than human drivers.
Hotel Check-In Name Parsing Bug
- Philip Koopman shared a hotel check-in bug caused by poor name parsing.
- The system booked his last name as 'J.R.' because it assumed the last token was the surname.
LLMs Fake Understanding
- Modern LLMs produce statistically plausible text but lack true understanding.
- They excel when factual accuracy is unnecessary but can hallucinate confidently.