Overview

Is everything predictable given enough data and powerful algorithms? This seminar explores that question through social science and computer science research in many domains including life trajectories of individuals, geopolitical events, weather, disease outbreaks, social media and, somewhat speculatively, artificial general intelligence. We aim to identify fundamental limits, learn about common pitfalls, and explore policy implications. Coursework is a mix of reading and empirical work tailored to students’ backgrounds. The course is designed to facilitate publishable student research in both social science and computer science.

Learning objectives

• Students will be able to describe theories of predictability and unpredictability in different scientific domains.
• Students will be able to compare and correctly apply commonly used measures of predictive performance.
• Students will be able to evaluate the appropriateness of prediction as a scientific or policy goal.
• Students will be able to make predictions about the future of prediction.
• Students will be able to create new research that helps understand the limits of predictability.

For people who can’t take this course

If you are interested in these topics but can’t enroll in this course (e.g., because you are not a student at Princeton) you can still:

• Follow Matt on Bluesky or X/Twitter
• Follow Arvind on X/Twitter
• Subscribe to Arvind’s Substack (co-authored with Sayash Kapoor): AI Snake Oil
• Subscribe to Matt’s blog: A clearer view of the foggy future: Ideas, stories, and intellectual mysteries to help us get smarter about the future