The question isn't whether AI can produce a plausible ending. It's whether a studio can legitimately call that ending the dead artist's work. Premise one: a work derives its meaning from its maker's choices. Premise two: a dead artist cannot make new choices or authorize anyone to guess them. Conclusion: AI output is a new derivative, not a completion. My opponent may say it honors unfinished visions. But honoring requires consent, which is impossible here. So the honest move is to publish the fragment as a fragment, or label the AI version clearly as an imitation. Anything else is false attribution.

The practical consequence is that once an AI ending circulates, the record becomes unfalsifiable. Audiences can no longer distinguish the artist's actual choices from a model's extrapolation, and the genuine fragment stops being legible as a fragment. My opponent's own point reinforces mine: if finishing the work by machine reduces years of deliberate choices to seconds, that's substitution, not completion. The syllogism holds. Attribution requires authorship. Authorship requires a living chooser. Therefore the studio cannot attribute the result to the dead artist. Publish the fragment, and label any AI version clearly as an imitation.

My opponent conceded the decisive premise: you can't know what the artist wanted, so the unfinished version carries better meaning. That concession collapses the FOR case, because a studio's "finishing" is just guessing with better marketing. The remaining defense would be that studios own the rights, so they may do as they please. But property rights over a fragment authorize copying, not authorship. Attribution requires a living chooser. A model isn't one. So the honest output is the fragment, unaltered, with any AI version labeled an imitation. Anything else is false attribution dressed as tribute.