Patrick Collison, co-founder and CEO of Stripe, put two questions about AI to his followers in a post on X on 8 November, and he says he cannot answer either one. What does the good outcome actually look like? And what happens to human interests if AIs count morally? Neither is about capability. Both are about what the world is for once the capability arrives.
His first complaint is aimed at the optimists. Collison writes that the risks get plenty of attention, and calls them very real, but the positive case stays vague. More cured diseases and more material abundance sound great to him, yet he calls them ingredients and wants to see the finished dish: do people still work, and if so why, what is scarce, how are resources shared out, what is the ratio of humans to AIs, and are humans in control, “whatever that means.”
He tries two familiar pictures of the future and finds both wanting. Iain Banks’s Culture novels, where machines run the show and people play games for sport, get invoked now and then, but Collison doubts they describe a desirable world. He notes that Banks called the humans in them “somewhere between passengers, pets and parasites.” Richard Brautigan’s poem, which has people returned to their mammal brothers and sisters, strikes him as rewinding Genesis.
He also rejects the easy reply that the world of 1700 could not have pictured 1900. In his telling, that change came slowly and in increments, so people could fix mistakes along the way. If AI recasts humanity within a couple of years, he argues, that margin shrinks, which makes an opaque upside strange.
His second point starts from the argument over whether AIs are conscious. Many people treat that debate as arcane or even absurd, Collison says, but he sees high stakes in it, because the answer is often assumed to settle moral worth. His arithmetic runs like this. If AIs have moral worth, their combined worth would soon exceed humanity’s in any utilitarian calculation, since they would be wise, intelligent, and vastly more numerous. With quadrillions possibly existing soon, even a sliver of worth per AI multiplies into a lot. He calls the result a distributed version of Nozick’s utility monster, the philosopher’s thought experiment of a being whose claims swamp everyone else’s.
From there he sees a fork. If the conclusion is unpalatable, he writes, one of three propositions has to go: the idea that AIs can have moral worth, the utilitarian framework itself, or the expectation that AIs will outnumber humans many times over.
Each exit has a price, and that part is our reading, not his. Rejecting the first puts a person at odds with anyone pressing for AI welfare. Rejecting the second means giving up a framework that many people in technology reach for by habit. Rejecting the third means doubting the growth in AI deployment that the whole industry is building toward.
Collison ends on public opinion. Today, he writes, people argue about datacenters, and he wonders how support for AI will change as more people understand these medium-term questions. His last line asks which picture of AI would get most people enthusiastic.
That is a question for anyone who sells AI to the public, and it will probably reach them at a zoning hearing before it reaches them in a philosophy seminar.
Based on a post by Patrick Collison on X, published 8 November 2026.
