

On Bullshit in AI
Jul 31, 2025
Arvind Narayanan, a Princeton computer science professor and director of its Center for Information Technology Policy, joins to discuss the nuanced realities of AI. He critiques the rampant hype and misinformation surrounding AI, emphasizing its limitations and ethical concerns, especially in critical areas like healthcare and justice. The conversation dives into the deceptive nature of AI technologies, urging listeners to critically evaluate AI tools and understand the broader implications of AI solutions, while addressing the dangers of polarized perspectives on the tech’s future.
AI Snips
Chapters
Books
Transcript
Episode notes
AI Types and Limitations
- AI is an umbrella term covering loosely related technologies, including generative AI and predictive AI.
- Predictive AI often used in consequential decisions has limited accuracy, just slightly better than random guessing.
How AI Learns and Progresses
- AI learns by mimicking patterns in large datasets rather than using fixed rules.
- The progress in generative AI slowed as models exhausted internet data for training.
AI Cheating Detectors Fail
- AI-based cheating detectors in education often fail and falsely accuse many students, causing serious consequences.
- These tools don't work well due to adversarial conditions where improved detectors lead to better cheating AI.