Cory Doctorow is sharp, funny, technically literate, and quite good at talking about patterns many of us already kind of feel. His critique of AI is more interesting than the lazy version where the technology is “fake” or “useless.” He uses AI himself, in modest ways, and his real target is the ideology around it: the investor hype, the managerial fantasies, etc.
Doctorow’s idea of the “reverse centaur” is useful as well, I think: instead of a human using a machine to extend judgment, the human becomes a servant of the machine, clicking, checking, correcting, and absorbing blame at whatever speed the system demands.
Still, I find myself getting weary of Doctorow’s union-style poor-workers shtick. That is not insensitivity to the issue. The rhetoric can start to feel like punk Marxism. You got your bosses, oligarchs, monopolists, bubbles, workers, exploitation … rinse/repeat. Of course that story seems reasonable. Hasn’t it been part of the cultural and political playbook for much of the twentieth century?
But as a sociologist (thunk), I have to say, the world is a heck of a lot more complicated than that.
Managers are not all cartoon villains, even though they often seem so from a distance. And workers are not always noble victims. Users sometimes want the tools that later constrain them, right? Institutions adopt bad systems for many reasons, which I have seen firsthand. Yeah, cost-cutting is part of that, but let’s also not forget panic, imitation, compliance theatre, labour shortages, professional insecurity, procurement, and plain old, well, confusion. Power just isn’t a master key that opens every door.
Sam Harris offers a useful comparison.
Doctorow is worried about a familiar kind of power. Your political elites. He sees AI as an extension of institutional power or elite power or whatever. The danger is AI that does exactly what firms, states, platforms, and administrators want it to do.
Harris worries about a different kind of power, if I understand him correctly. More … cognitive power.
For Harris, the deeper issue is what happens when we build systems that are more capable than we are. Systems that can reason, code, persuade, strategize, deceive, or eventually improve themselves in ways we may not understand or control.
Doctorow’s critique can make it sound as though the danger comes mainly from the social order around AI, and I think that’s intentional given the little political slice of life he occupies. Harris would say, perhaps, no, the danger also comes from the nature of the technology itself.
Even if we solved capitalism, monopoly, oligarchy, DRM, and labour exploitation, the alignment problem would remain. A benevolent research lab could build a dangerous system. A democratic government could build a dangerous system. A university consortium could build a dangerous system. A worker-owned AI collective could build a dangerous system (there’s a short story idea for you).
This also changes how we hear Doctorow’s “AI bubble” argument. He is right that high demand does not prove profitability. If companies are selling hundred-dollar bills for one dollar, then adoption tells us bloody little about the sustainability of the business model. Yup. Fair.
But from a Harris-style perspective, it is also beside the deepest point. A lab leak, synthetic pathogen, cyber weapon, or autonomous military system does not really need good unit economics to matter.
What if capabilities keep improving anyway? Even a collapsed bubble could leave behind powerful models, open weights, automated research tools, cyber capabilities, and lowered barriers to dangerous forms of misuse.
Doctorow’s boss psychology also raises some issues. His line about bosses wanting to fire workers and replace them with machines is memorable, and there is truth in it, sure. Many managers do dream of a workplace with fewer difficult humans in it. I’m not a manager and part of me fantasizes about that because I am anti-social. But this is way too morally theatrical.
Other pressures include competition, uncertainty, national-security concerns, prestige, and fear that someone else will move first.
Good people can race because they fear worse actors.
Cautious labs can become reckless under pressure.
Governments can militarize AI because adversaries might.
Open-source idealists can release dangerous capabilities in the name of democratization.
Ordinary users can demand tools that make society less governable.
This sits near Teaching festival, Getting the AI Details Right, and Radical AI defaults.