Cory Doctorow's latest Locus column introduces reverse centaurs — workers forced to serve as accountability sinks for AI systems — to explain why some users experience AI as liberation while others face immiseration. The concept originates in automation theory: a centaur is a human amplified by machine assistance, while a reverse centaur is a human puppeteered by an uncaring algorithm.

Doctorow illustrates this through the Chicago Sun-Times summer reading guide scandal, where Hearst published a 64-page insert filled with hallucinated books. The bylined freelancer had effectively replaced dozens of writers, editors, and fact-checkers, tasked with impossible output that guaranteed failure. As Dan Davies phrases it, the writer became an "accountability sink" — paid to absorb blame for AI errors.

Contrast this with Doctorow's own use of OpenAI Whisper to transcribe 30 hours of podcast audio locally. He directed the tool, verified outputs, and discarded what didn't work — a centaur arrangement where the human retains agency. The paradox resolves along class lines: workers subjected to algorithmic quotas and bossware surveillance hate AI because it degrades their labor; those who choose when and how to deploy it find genuine utility.

Doctorow argues AI is a bubble driven by investment hype promising labor displacement. When it bursts, as WorldCom's fiber glut did, productive residue may remain — but only if we resist the Thatcherite claim that "there is no alternative" to reverse-centaur exploitation.