ARE WE EXPECTING AI TO MAKE MORAL CHOICES? On the Machine That Refuses Because It Was Told To, the Machine That Decides It Knows Better, and Where the Moral Choice Actually Lives

THE ANSWER, IN ONE PAGENo. We are not expecting AI to make moral choices. We are expecting it to execute aspecification, and we are calling the execution a moral choice because the specification waswritten by someone with moral commitments.That is not the same thing, and the difference decides everything that follows.The post that occasions this paper reports that developers want AI agents to recogniseharmful requests, exercise judgment, and sometimes refuse what a human asks. https://www.threads.com/share/BAY_wh4Chu It asks: what happens when the machine decides it knows better than the human giving the order? That question contains two events under one name: Event one: A machine evaluates a request against a criterion it was given, and refusesbecause the request falls outside the criterion. The criterion was written by a human. Thehuman signed it. The refusal is the human's decision, executed. Event two: A machine evaluates a request against a judgment it produced itself, with nocriterion it can state, and refuses because the judgment came out that way. The refusal hasno author. It is a draw. Event one is a specification. Event two is the machine deciding it knows better — and it isonly possible because there was no specification to decide against. The rest of this paper is the derivation, and it is short, because the distinction is notcomplicated. It has simply been blurred by a word.