Princeton UP Ideas Podcast On Bullshit in AI
Jul 31, 2025
In this engaging discussion, Arvind Narayanan, a computer science professor at Princeton and director of the Center for Information Technology Policy, dives into the pitfalls of AI hype. He addresses how AI chatbots often 'hallucinate' and the rampant misinformation in AI marketing. Narayanan distinguishes between generative and predictive AI and critiques their applications in fields like hiring and education. The conversation also evaluates extreme views on AI's future, urging a balanced understanding to avoid policy missteps.
AI Snips
Chapters
Books
Transcript
Episode notes
AI Is Not One Technology
- AI is an umbrella term for loosely related systems, not a single technology. - Predictive AI often only slightly outperforms random guessing but impacts important decisions like jobs and justice.
How AI Learns and Its Limits
- Machine learning AI copies patterns from data instead of using fixed rules. - Progress in generative AI stemmed from scaling training data, but biases persist and affect system behavior.
Cheating Detectors Are Snake Oil
- Cheating detectors purport to identify AI-generated homework but perform poorly, causing false accusations. - These detectors exist because educational institutions struggle to address the root causes of cheating.
