New Books Network Cass R. Sunstein, "Imperfect Oracle: What AI Can and Cannot Do" (APS Press, 2025)
Sep 29, 2025
Cass R. Sunstein, a Harvard legal scholar and author, delves into the intriguing world of AI in his new book, discussing its potential and limitations. He highlights how AI can reduce human biases and noise in decision-making but emphasizes that it cannot predict every outcome due to inherent uncertainties. Sunstein also addresses the importance of quality training data for AI performance, the dangers of overreliance on technology, and the need for regulations to protect consumers from manipulation. His insights promise to reshape how we view AI's role in society.
AI Snips
Chapters
Books
Transcript
Episode notes
Heuristics Create Systematic Human Biases
- Human judgment uses heuristics that work but create systematic biases.
- AI can avoid many of those biases and reduce costly mistakes in judgment.
Hayek Explains AI's Predictive Limits
- There are intrinsic limits to prediction because many important variables are dispersed and unknowable.
- Hayek's calculation problem applies to AI: access to more data doesn't always make certain predictions possible.
Coin Flip Dialog With An LLM
- Sunstein asked a large language model to predict his coin flip and the model explained it lacks access to necessary variables.
- He used this exchange to illustrate deterministic events that remain unpredictable in practice.
