The generalization test exists because scale changes what a choice means: a behavior can be beneficial once and dangerous when everyone adopts it. One family leaning on AI tutoring is probably fine; every family doing so while schools hollow out weakens the public educational ecology. One employer using AI screening is efficient; every employer using near-identical systems can render whole kinds of applicant invisible. The action did not become morally different — the system effect changed with scale — so no ecological judgment is complete until it asks what happens to the field if every similarly positioned participant does this?
It is a universalization question, but asked of the field rather than the will — a Kantian move made ecological rather than moral. Its value is diagnostic: if generalizing a strategy yields collapse, an arms race, a monoculture, or a depleted commons of field conditions, then a strategy that looks like a clear local win is in fact globally unstable, and its apparent success is borrowed against a boundary it does not count.