"The AI Con" Con

Reading Emily Bender and Alex Hanna’s The AI Con is like time-travelling back to 2020. Many passages are a stark reminder of the world of peak woke—they described Searle’s famous Chinese room argument as “an extremely othering way of making the argument.” More importantly, the picture they paint of AI capabilities would have been an accurate enough description of the AIs of 2020, but seemingly has not advanced since then.

Perhaps it is no surprise that the authors admit to not using the technology they discuss.

The Bender and Hanna (B&H) thesis is that AI is massively overhyped. That it’s a mostly useless technology with little upside and massive downside. It will devastate the environment, use up enormous quantities of water, exacerbate inequality, and so on, all without having any real benefit. It will not assist seriously with scientific advancements, nor will it be economically viable. The authors seem to curiously maintain the position that it will replace a number of jobs but won’t improve the productivity of jobs.

This position was not plausible in 2020, and has grown less plausible since, having had a head-on collision with the facts. AI can invent novel math proofs that impress the best mathematician in the world. It can automate away big chunks of coding. They can assist writing far better than a secretary. In seconds, by asking LLMs, you can get a detailed and well-researched answer with sources. They can almost instantly complete tasks that take people hours.

The AI Con is what you get when a thesis you’ve been stochastically parroting for years is decisively disproven by the evidence: it’s a desperate and error-filled attempt to rescue a deeply implausible thesis. Nearly everything in the book is poorly reasoned or misleading. It is also written with the confidence of someone expounding trivialities, rather than arguing for a contentious thesis. The authors rarely saw the need to respond to objections.

The best thing I can say about the book is it’s very well written. B&H are engaging writers. Some bits are quite funny (e.g. “Meta’s LeCun beclowned himself.”) I rarely found myself bored when reading (though this is perhaps because it is hard to be bored when encountering outrageous falsehoods on nearly every page).

The case for AI being a big deal is pretty straightforward. AI is advancing rapidly. It can already write better than most undergraduates, and possesses a great degree ...