

Cass R. Sunstein, "Imperfect Oracle: What AI Can and Cannot Do" (APS Press, 2025)
Sep 29, 2025
In this engaging discussion, Cass R. Sunstein, a Harvard law and public policy scholar, explores the intricacies of AI outlined in his book, Imperfect Oracle. He reveals how AI can help combat human biases and noise that lead to poor decision-making. Sunstein discusses the limits of AI predictions, the importance of quality training data, and how algorithms can aid in reducing discrimination in legal contexts. He also highlights the potential risks of over-reliance on AI and the need for thoughtful regulation to mitigate manipulation.
AI Snips
Chapters
Books
Transcript
Episode notes
Practical Frame For AI
- Cass Sunstein wrote Imperfect Oracle to cut through extreme claims about AI and focus on practical use and limits.
- He argues AI can improve judgments but cannot eliminate fundamental limits on prediction.
Human Heuristics Create Predictable Errors
- Humans use heuristics that often produce predictable biases and errors in judgment.
- Understanding these biases shows where AI can outperform humans by reducing systematic mistakes.
Hayek's Limit On Prediction Still Holds
- Hayek's argument about dispersed information limits applies to prediction even with powerful tools like AI.
- Many outcomes remain unpredictable because we lack access to all relevant, dynamic variables.