AI Snake Oil: What Artificial Intelligence Can Do, What It Can't, and How to Tell the Difference by Arvind Narayanan, Sayash Kapoor — Summary & Review

“AI Snake Oil” is like that brutally honest friend who tells you your new gadget is mostly smoke and mirrors—except this friend is a Princeton professor and a PhD student who actually know their AI. Arvind Narayanan and Sayash Kapoor don’t just vent about AI hype; they meticulously unpack what AI can actually do and where it falls flat, especially when the tech industry dresses up mediocre systems as revolutionary breakthroughs.

The central thrust is simple but rarely spelled out plainly: not all AI is created equal, and a lot of what’s marketed as AI is just hype with a fancy label. They introduce the term “AI snake oil” to describe these overhyped systems that promise a lot but deliver little, sometimes causing real harm. The authors are particularly concerned with the societal impact of these misleading claims, showing how AI hype can lead to bad decisions in hiring, policing, and beyond.

One eye-opening example is predictive AI in recruitment. Companies love to tout AI tools that supposedly identify the perfect candidate by analyzing past hires or resumes. But these systems often bake in existing biases—like favoring candidates who look like the current workforce—and produce shaky predictions at best. The result? Hiring decisions that might seem scientific but are actually flawed and unfair. It’s a classic case of technology dressed up as impartial but actually just replicating human prejudices.

Facial recognition technologies get a similarly skeptical treatment. The book highlights how these AI systems have led to wrongful arrests, disproportionately affecting Black individuals. It’s a grim reminder that AI isn’t neutral; it inherits and amplifies societal biases, especially when deployed without critical oversight. Narayanan and Kapoor argue that these failures aren’t just technical glitches but symptoms of hype-driven deployment without adequate scrutiny.

The book doesn’t just bash AI, though. It carefully distinguishes between predictive AI—tools that analyze data to make forecasts—and generative AI, which creates new content like text or images. While the authors are solid on predictive AI’s limitations, their take on generative AI is more tentative, reflecting the fast-moving nature of that field. This cautious stance might frustrate readers looking for definitive answers on generative AI’s capabilities and risks.

What’s refreshing is the authors’ accessible style. They avoid jargon and talk directly to readers who want to make sense of AI’s promises without getting lost in technical weeds. Their arguments lean heavily on research, including their own reproducibility studies, lending weight to their skepticism. Yet, they’re not dismissive of AI’s potential; instead, they push for a more realistic, evidence-based conversation.

That said, the book isn’t a deep dive into the nuts and bolts of AI algorithms. If you’re after a technical manual, this isn’t it. It also spends less time on solutions, focusing more on diagnosing the hype problem than prescribing fixes. And given how fast AI evolves, especially generative AI, some parts risk feeling outdated sooner than others.

Published amid a frenzy of AI breakthroughs and debates, "AI Snake Oil" is a timely reminder that not all that glitters in the AI world is gold. Narayanan and Kapoor’s reputations—spotlighted by their inclusion in TIME’s 2023 list of influential AI figures—add credibility to this no-nonsense critique.

In short, this book is for anyone who’s had enough of the AI hype circus and wants a grounded look at what AI can actually do, where it stumbles, and why it matters. It arms readers with the critical tools to spot snake oil and think twice before buying into the next AI miracle pitch.