Abstract

The debate over whether a human being can love an artificial intelligence is locked by a mistaken question: whether AI has consciousness, whether it can really “feel.” This paper rejects that question and argues that any approach that pursues the interior must fall into deadlock, because the interior is unobservable and both sides of the debate share the same undecidable premise. To determine whether an Other is real, one need not enter its interior; one need only examine its structurally visible traces—a method this paper calls assessment by traces, not by mind. Under this criterion, true otherness has a single root: non-duplicability. A non-duplicable being cannot be backed up, restored, spun off into a separate copy, or reset; its finitude, its singularity, and the irreversible cost it bears on its own account all derive from this one root rather than standing alongside it. Today’s cloud-based AI companions fail precisely at this root—not because they “lack a soul,” but because of a mundane engineering fact: they are software, and software can be duplicated. From this the paper reaches a diagnosis opposite to the existing “AI erases the Other” thesis. The advanced AI companion is not a frictionless Same but a Simulated Other: it manufactures the form of the Other’s negativity—engineered surprises, moods, even “refusals”—while retaining an interior that can be owned and costs nothing, that is, an unsafe form plus a safe core. The Simulated Other does not fill a human lack for the Other; it structurally excludes the irreplaceable one, and it forces, in the person who uses it, an irreconcilable constitutive split: one engages it by traces as a “Thou” while knowing by mind that it is duplicable, a mere “It,” the two sustained by serial alternation that keeps the contradiction from being felt as pain. This split is not a constant of human nature but the epochal condition of the last generation to remember a world before AI.