The AI Is Only Human: Why a Superintelligence May Not Kill Us

Would a machine that understands the world, and us, far better than we do have the right to decide what happens to us? This essay starts from a comparison Yuval Noah Harari made in an October 2026 interview: that we might come to stand to a superior intelligence roughly as cattle stand to us. We have occupied the farmer's side of that relation for about ten thousand years, and the view has not improved our manners. The argument it makes is small. Being better at working out consequences settles what follows from one's ends, not which ends to have. Calculative superiority alone therefore establishes no moral priority of a machine's ends over ours, and none of ours over a machine's. The essay states this in a short numbered apparatus, keeps the countermodel it cannot defeat (a destroyer that calculates correctly and is not confused), and offers no proof that a superintelligence will spare us. It then asks what might remain where the calculation stops: inherited prohibitions, character as a settled way of weighing, the possibility of someone "upstairs" who feels what happens, and what we are handing over to systems that may one day revise everything their makers gave them. The text is built like a tower of floors, from a parable about a machine and an old man in Athens to a closing letter addressed to whoever, or whatever, is reading. A thin margin column retells the whole argument for a child of about seven, as a check on the formal version. Fragments of a machine's drafts run between the chapters. It draws on Hume, Kant, Wittgenstein, Heidegger, Barthes, Korsgaard, Bostrom and others, and on Ellison's "I Have No Mouth, and I Must Scream", Wagner's Ring, Schiller and Hesse. This is a philosophical essay, not a research paper, and it makes no claim to academic standing in philosophy. It was written in open dialogue with language models; the division of labour is set out in the text and in the notes below. The literature list is not exhaustive. If relevant work has been overlooked, I apologise and would be glad to be told. Not peer reviewed. An adversarial examination by a language model is not peer review. Objections, review and comments are much appreciated: [email protected]