The Author of "AI Snake Oil": People Aren't Afraid of Machines — They're Afraid AI Will Expose Who Can Actually Think

Princeton computer science professor Arvind Narayanan has spent years pushing back against Silicon Valley's grand claims about artificial intelligence. He co-authored the book "AI Snake Oil," which takes a skeptical look at claims that algorithms can reliably predict who will be a good employee, which patient will get sick, or who might commit a crime. At the same time, he considers the growing public backlash against AI to be entirely real and understandable — but not the product of a single cause. Instead, he sees it as a tangle of overlapping anxieties: fear of job loss, distrust of Big Tech, anger at the outsized influence of billionaires, environmental costs, and uncertainty over which skills young people should still be holding onto.

Narayanan's critique doesn't mean generative AI is useless — he calls it "extremely useful" for any knowledge worker, and his real target is the hype surrounding it, not the technology itself. His sharpest objections are reserved for a different category of AI: the systems used by hospitals, insurers, HR departments, and the criminal justice system to make high-stakes predictions about people, since reliably forecasting the future is inherently difficult, and a wrong prediction can shape decisions as serious as hiring, insurance coverage, or incarceration.

He uses a "sandwich" metaphor to describe AI's near-term effect on jobs: AI doesn't so much eliminate jobs as bury them under layers of verification and oversight of its own use. An even riskier scenario is when someone keeps their title but the job itself is hollowed out into what he calls "janitor work": automated systems handle most of the actual thinking, while the human absorbs the blame whenever something goes wrong. He calls this a "moral crumple zone" — just as a car's crumple zone absorbs the force of an impact, the person nominally responsible, despite lacking real control over the system, becomes the one who absorbs the blame for its failures.

Narayanan also explains why different professions relate to AI so differently: programmers work with AI interactively — spotting errors, testing outputs, and folding the model's suggestions into a project — which keeps the human at the center of the process, something he calls a "growth loop." For artists, the situation is entirely different: a single prompt produces a finished image, creating the impression that the system has bypassed the human creative process altogether, which can spiral into what he calls a "dependency spiral."

For students, the situation is even more fraught: they're expected to be fluent in the tools they'll encounter at work, yet over-reliance on AI can strip them of the fundamental skills needed to judge whether the system's output is actually correct. Narayanan himself admits as much: his optimism doesn't rest on the belief that the technology produces good results on its own, but on the assumption that people will keep it under constant scrutiny. He compares AI to what the crane did for physical labor — we still build skyscrapers, we just no longer haul the steel ourselves, and the person operating the crane still decides exactly where the beam goes. His final conclusion: people aren't really afraid of thinking machines — they're afraid that AI, in real time, will expose who can actually think.