AI Snake Oil: What Artificial Intelligence Can Do, What It Can't, and How to Tell the Difference

Two computer scientists separate what AI can actually do from what vendors, journalists, and venture capitalists claim it can do. Required reading for anyone who teaches, studies, or makes decisions about technology.

Arvind Narayanan and Sayash Kapoor have spent years watching artificial intelligence get oversold: hiring algorithms that claim to predict job performance, predictive policing tools marketed as objective, facial recognition systems promoted with accuracy figures that dissolve under scrutiny. In AI Snake Oil, they document the gap between real machine learning capability and the hype that surrounds it, with enough technical fluency to be authoritative and enough clarity to be genuinely useful to non-specialists.

This is the book to assign in courses on technology policy, data ethics, science and technology studies, or any undergraduate seminar where students are expected to think critically about algorithmic systems. It is also the book to keep on your own desk when a vendor pitches you a new tool or a dean circulates a proposal to automate something that shouldn't be automated. Narayanan and Kapoor are precise without being dry, and their skepticism is grounded in how these systems actually work rather than in generalized technophobia.

Princeton University Press, 2024. Paperback.