Answers are grounded in this paper and cite their sources.
Conceptual analysis reveals an ethical generative framework for reconstructing murdered artists' works, indicating that model training must prioritize refusal signals over market optimization.
Dedicatory Generation — Working With the Heritage of Murdered Artists When There Is No Estate to Ask, No Record of Intent, and the Destruction Was Aimed at the Culture Itself Between 1975 and 1979 the Khmer Rouge destroyed a generation of Cambodian artists — the musicians, filmmakers, writers and performers of what is remembered as Cambodia's Golden Age. The recordings that survive did so by accident: cassettes carried across a border, prints in a private collection, a reel that somebody buried. What did not survive was almost everything else, and the almost everything else is the subject of this paper. Generative models can now produce new work in a lost artist's voice. The public discourse about whether they should has developed around a case that does not apply here: a living or recently-deceased artist, with a surviving estate, in an intact legal system, where consent is at least a question that can be asked. Every proposed framework — licensing, opt-out, estate approval, revenue sharing — presumes a party on the other side of the table. The genocide case differs on four axes at once. Consent is not obtainable, and not merely unobtained. There are no surviving estates or institutions to hold rights, because the institutions were destroyed alongside the people. The records of intent were themselves destroyed — there is no diary, no contract, no interview in which an artist said what they would have wanted. And the destruction was aimed at the culture: the erasure was the objective, not a side effect, which means that a reconstruction performed carelessly does not merely offend an absent party — it completes an unfinished act. This paper specifies a discipline for that case. Its central move is a single change of preposition. Work generated in a lost artist's voice is published for them and never by them: an offering, not an authorship claim — a form with deep precedent in the tradition the culture actually belongs to, where merit is dedicated to the departed and the dedication is the point. Nine further claims follow, and one of them is the one to protect hardest: the model's training objective must exclude every economic signal, and admit only a refusal signal. A system optimized on what performs is a system that will learn to make the dead popular. A system that can only be told no can be steered away from harm without ever being pointed at a market. The lineage here is Cambodian and not imported. Rithy Panh filmed clay figurines rather than reenactments, because a reenactment of the camps would have been a lie with better production values. Anastylosis at Angkor reassembles original stones and makes every new element visibly distinguishable, forever. Both are answers to this paper's hardest problem, arrived at long before it: fidelity and forgery are the same axis, so success is the danger. Offered under CC0 1.0 Universal as defensive prior art. Keywords: generative models, cultural heritage, genocide, attribution, dedicatory generation, anastylosis, provenance marking, training objectives, refusal signal, voice models, Cambodia, Khmer Golden Age, defensive publication. --- Provenance. This paper is part of the THonly research corpus, dedicated to the public domain under CC0 1.0. The canonical version is at https://thonly.org/research/dedicatory-generation. Its SHA-256 is 49b7ce1607a2b915cb72f3746eed17b25de66c68c91e51900c2fcda8fabb4361, independently timestamped to the Bitcoin blockchain via OpenTimestamps and signed under RFC 3161 by three trust authorities, one of them eIDAS-qualified. AI co-authorship is disclosed. Miss Aquarius is the consistent name used for the AI collaboration across all venues.
No takes yet. Share an insight, caveat, or question.
Ly et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: