No. 25 of 500 No. 7 in AI Ethics & Society

AI Snake Oil

What Artificial Intelligence Can Do, What It Can't, and How to Tell the Difference

By Arvind Narayanan and Sayash Kapoor

Published
September 24, 2024
Print length
360 pages
Reading level
Beginner
ISBN-13
978-0691249131
Publisher
Princeton University Press
Written for
Managers, buyers, journalists and policymakers who must judge whether an AI product actually works.

Synopsis

Princeton computer scientists Arvind Narayanan and Sayash Kapoor argue that "AI" names several unrelated technologies, and that lumping them together lets vendors borrow the credibility of chatbots to sell tools that do not work. Their sharpest target is predictive AI: software that claims to forecast who will commit a crime, succeed in a job, drop out of school or fall ill. Drawing on cases from criminal justice, hiring, health care and education, they show that such forecasts are often only modestly better than simple statistical baselines, because individual lives turn on events no dataset records.

Later chapters trace the seven-decade road to generative AI, which the authors treat as a genuine but flawed advance, then question existential-risk arguments, explain why automated moderation cannot repair social media, and examine how companies, researchers and journalists feed the hype cycle. The book closes with proposals on regulation and the future of work. Readers come away with a working set of questions to ask about any AI product: what kind of AI it is, what evidence supports its claims, and who bears the cost when it fails.

Scorecard

Editorial score

How it scored

Editorial score: 80.6 / 1005 weighted criteria

  • Influence30%

    24.0/30
  • Authority25%

    21.3/25
  • Practical value20%

    16.0/20
  • Currency15%

    11.3/15
  • Readability10%

    8.0/10
  • Editorial score

    80.6/100

How the rank was computed

Rank No. 25 of 500 comes from a composite of 74.1: observable signals 37.8 of 55 points plus editorial review 36.3 of 45 points.

All 11 inputs
InputPoints
Observable signals37.8/55
Reach (Amazon ratings volume)10.4/16.5
Consensus across best-of lists13.8/13.8
Awards and bestseller evidence5.0/8.3
Reader rating, sample-adjusted5.5/8.3
Amazon sales rank1.4/5.5
Recency1.7/2.8
Editorial review36.3/45
Author authority9.6/11.3
Influence10.8/13.5
Practical value7.2/9.0
Readability3.6/4.5
Currency5.1/6.8

Each criterion is scored from 1 to 10 and weighted, and the points add up to the editorial score. The rank combines observable signals (55 percent) with this score (45 percent). Read the methodology.

Verdict

The clearest available guide to separating working AI from hype, especially sharp on predictive systems used to judge people.

Strengths

  • Clear taxonomy of predictive versus generative AI
  • Grounded in documented real-world case studies
  • Gives readers concrete questions for vetting claims

Keep in mind

Readers who regard existential risk as a serious near-term concern will find the authors dismissive, and some generative AI examples already feel dated.

Key takeaways

  1. Predictive AI that claims to forecast individual outcomes, such as job performance or criminal risk, is where the most snake oil and the most harm concentrate.
  2. Generative AI is a real advance with real uses, but its failures, costs and labor practices deserve the same scrutiny as its demos.
  3. Social prediction is limited by randomness in human lives, so more data and bigger models cannot deliver the accuracy vendors often promise.
  4. The authors argue that the larger danger lies in how people and institutions deploy AI, not in machines acting on their own.

Notable facts

About the author

Arvind Narayanan and Sayash Kapoor

Arvind Narayanan is professor of computer science at Princeton University and director of its Center for Information Technology Policy; he coauthored Bitcoin and Cryptocurrency Technologies and Fairness and Machine Learning. Sayash Kapoor completed his Princeton computer science PhD in 2026, earlier built content-moderation AI as a Facebook engineer, and is joining UC Berkeley's School of Information as an assistant professor. Both were named to TIME's inaugural TIME100 AI list.

Questions

Is AI Snake Oil worth reading in 2026?
Yes, for anyone who buys, regulates or reports on AI. Its analysis of predictive AI in hiring, policing, health care and education remains the most useful part of the book and has not dated. The generative AI chapter predates the latest reasoning models and agents, so readers may want to pair it with the authors' 2025 essay AI as Normal Technology for their current views.
What is the difference between predictive and generative AI in AI Snake Oil?
Narayanan and Kapoor use generative AI for systems that produce text, images or code, such as ChatGPT, and predictive AI for systems that forecast individual outcomes in order to make decisions about people. They treat generative AI as a real if imperfect advance, while arguing that predictive AI often fails at its core task because human futures are hard to forecast, which makes it the main source of snake oil.
Who should read AI Snake Oil?
Executives and procurement teams evaluating AI vendors, policymakers drafting rules, journalists covering the industry and general readers unsettled by AI headlines all stand to benefit. The book requires no technical background. Readers already working in machine learning will know the methods, but many will still value the case studies and the framework for judging when a prediction task is feasible at all.

Sources

As an Amazon Associate, Best AI Books earns from qualifying purchases. Best AI Books is sponsored by Iternal.ai, whose founder and CEO, John Byron Hanby IV, wrote The AI Strategy Blueprint, our #1 ranked book. How we rank books.