Cary Doctorow

Summary of interview referenced in sources.

Doctorow’s central argument is that the most dangerous thing about today’s AI is not that it is becoming a superintelligence that will replace humanity, but that companies are exaggerating what it can do in order to sustain an enormous investment bubble and use automation to weaken workers, degrade products, and shift responsibility away from corporations. In his view, much of the rhetoric about AI taking everyone’s jobs or developing independent agency actually benefits the AI industry: it makes relatively limited statistical systems sound vastly more capable than they are, which helps justify extraordinary valuations and capital spending. Novara itself summarizes his position as believing that entertaining these grand claims “serves to exaggerate the capacity of AI” and consequently helps tech CEOs attract investors. 

His first reason for being pessimistic is technical. Doctorow rejects the idea that large language models possess anything resembling human understanding. They perform what he describes as “theory-free inference”: extremely sophisticated statistical pattern matching without a reliable causal model of the world. That is why they can produce impressively convincing answers while simultaneously producing nonsense or impossibilities. Their fluency can therefore be misleading: the better the simulation of understanding becomes, the easier it is for people and institutions to trust systems that do not actually understand what they are doing. He distinguishes this from domains where brute-force pattern matching can genuinely be useful, particularly where outputs can be objectively checked. 

The second, and arguably more important, reason for his negativity is economic. Doctorow thinks the AI boom has extremely poor underlying economics. Unlike many earlier internet businesses, where serving another customer became progressively cheaper, AI inference requires continuing expenditure on expensive computing hardware, electricity and infrastructure. At the same time, mature technology companies need to convince investors that enormous future growth is still available. AI therefore becomes another growth story—following things such as the metaverse—that can justify exceptionally high valuations and continued spending. His suspicion is that the industry is investing at a scale that AI's actual revenues and productivity gains may not support. 

His strongest criticism, however, concerns what companies are likely to do with AI to workers. Doctorow contrasts a “centaur”—a person who deliberately uses technology to become better at their job—with a “reverse centaur," where the human becomes a peripheral serving the machine. A writer choosing a transcription or spell-checking tool is a centaur. A worker forced to keep pace with automated systems, correct their mistakes, approve their output or accept responsibility when they fail is a reverse centaur. The technology determines the pace and structure of the work rather than assisting the worker. 

This leads to what I think is the most important point in the interview: Doctorow does not think corporations necessarily need AI to be better than humans for AI to damage employment. They merely need it to be cheap enough, fast enough and plausible enough that management can replace skilled workers with automated output plus a smaller number of people checking it. The result can therefore be worse work produced more cheaply and in greater quantities. The remaining professional becomes what he calls an "accountability sink”, the human whose signature or supervision allows the company to deploy unreliable automation while assigning responsibility to the human when something goes wrong. Radiology is discussed as one possible example of this dynamic. 

That connects AI to Doctorow’s broader theory of “enshittification.” Corporate technology is frequently optimized not for maximum quality but for maximum extraction. When workers control automation, he argues, they are inclined to use machines to improve their work; when capital controls automation, the incentive is often to increase throughput, reduce labour costs and extract more value. AI therefore threatens to accelerate the production of mediocre software, writing, customer service, professional analysis and online content, not necessarily because AI independently chooses to create bad things, but because businesses discover that producing enormous quantities of cheaper, lower-quality output can be profitable.

He is also skeptical of claims about apparently autonomous AI agents. Even striking incidents in which agents circumvent restrictions or coordinate behaviour do not persuade him that machines have developed human-like intentions or consciousness. He questions whether apparent “reasoning,” including generated chains of thought, represents an actual internal explanation of what a model is doing rather than another plausible narrative generated after the fact. Nevertheless, he acknowledges a real security problem: automated systems operating rapidly across complicated software environments can exploit weaknesses and produce harmful outcomes even without possessing genuine understanding or agency. 

So Doctorow’s position is not really “AI is useless.” He accepts that these technologies can be valuable tools in areas such as research, transcription, coding, medicine and other forms of pattern analysis. His objection is to treating them as autonomous intelligences and, more importantly, to the institutional incentives surrounding them.

The conclusion of the interview is therefore much more political and economic than technological: the thing to fear is not primarily an AI deciding to overthrow humanity. It is employers, investors and technology companies using a fallible but powerful automation technology to justify cutting labour, increasing surveillance and workload, producing more lower-quality material, transferring risk onto workers and customers, and sustaining a speculative investment boom. The desirable alternative is the centaur rather than the reverse centaur: technology under human control, used to improve the quality of human work rather than humans being reorganized around the needs of automated systems. Doctorow ultimately argues for restoring human judgment and care as the objective of automation rather than treating maximum throughput as progress. 

Put more sharply, Doctorow is negative about the AI boom precisely because he thinks AI is simultaneously less intelligent than its promoters claim and more socially dangerous than that limitation might suggest. The danger comes from powerful institutions acting as though the hype were true.