≡ Menu

Reginald Godwin on Copyright and AI Training: “Copyright, AI, and the Great Illusion”

Related:

See Reginald Godwin, “Copyright, AI, and the Great Illusion,” Libertarian Alliance (June 28, 2025).

I was asked by an old friend for my thoughts on this. I reproduce them below.

***

Well, the author has some good intuitions and seems to be on the side of the angles, but is basically confused and floundering. His points are not systematic or rigorous and somewhat confused. He is right that this is absurd and ridiculous, he is right that copyright is harming AI. I have written on this. See Whereupon Grok admits it (and AI) is severely gimped by copyright law and Libertarian and IP Answer Man: Artificial Intelligence and IP.

But the problem is copyright law per se, not that it has gotten “out of hand.” The author has misstatements and confusions such as:

  • “What the California court ruled was this: training AI on books may well constitute “transformative” use under American copyright law, and thus may not, in itself, be illegal.” Well, in the first place, copyright infringement is infringement, not “stealing”. Stop calling patent and copyright “property”; stop calling copying “theft”. So it is not that copying something is “illegal” under copyright law, the question is whether it is “infringement”. Not stealing, not a crime. (though there are criminal copyright offenses, but that is another matter and I doubt the author had that in mind)
  • “The case has been hailed as a copyright milestone. It possibly is that, but is also a symptom of a deeper decay in legal reasoning and our understanding of what machines, and laws, are for.” This is blather that says nothing. It sounds impressive to a layman I guess but it does not get to the root of the issue, which is IP, and one cannot understand this problem without a sound theory of property rights, not just intuitions about “fairness,” which is all this author seems to have.
  • “Let us first deal with the word “stealing.”” As noted above copyright infringement has nothing to do with stealing or theft. It does not even necessarily have to be copying–it can be a literal copy, but other acts can also be infringement of the rights of the copyright holder: such as (see Grok):
    • Unauthorized Reproduction — “non-literal copying” or “substantial similarity infringement,” which happens when the new work reproduces the protected expression (not just ideas, facts, or concepts) of the original in a way that is too close, even if the wording or details differ. This includes paraphrasing text, mimicking the structure of a song, or recreating visual elements of an artwork with minor changes. Writing a novel with a plot, characters, and setting that closely resemble those of a copyrighted book, even if the wording differs. Etc.
    • Creating Derivative Works Without Permission (my guess is this is the main charge publishers would use against AI companies, for the training part; plus the copying needed to acquire and store the source material–which is probably why the judge in the mentioned case “refused to dismiss the case outright, allowing a trial to proceed on the question of how the books were obtained and stored“). This would be modifying or adapting a copyrighted work to create something new, such as a sequel, translation, film adaptation, or remix, without the copyright owner’s consent. Like a movie based on a novel. Or a translation of a book. Or maybe a trained AI neural network!
    • Others infringement actions include: Unauthorized Distribution– Sharing or distributing copies of a copyrighted work without permission, whether for profit or not; Public Performance Without Authorization; Unauthorized Public Display; Importation of Infringing Copies; and Contributory or Vicarious Infringement (facilitating or benefiting from someone else’s infringement without directly copying the work); and Moral Rights Violations.
  • “Anthropic did not reprint the authors’ books. It did not sell them under its own name. It did not post them online for free. It fed them into an algorithm that adjusted statistical weights across billions of nodes to predict language. This is not copying. It is not publishing. It is not infringement in any meaningful sense.” This is a legal conclusion based on the author’s intuition that this is absurd; but as he later admits, copyright law can have absurd results! (“It is the digital equivalent of reading—something no one has yet managed to outlaw. … The idea that reading a book, even by machine, constitutes infringement is as preposterous as accusing a violinist of stealing Brahms after practising the concerto. … “). It may well be copyright infringement–creation of a “derivative work”–and the question is, if so, whether “fair use” (or the UK “fair dealing” idea) provides the accused infringer with an affirmative defense.
  • “Such an argument only holds up if one accepts that ideas and language can be owned in perpetuity, and that exposure itself is a form of reproduction. But that is not copyright. That is a doctrine of pre-emptive censorship.” Sure. But so what? Copyright IS CENSORSHIP. It is already absurd. Copyright is Unconstitutional
  • “Copyright was not conceived as a permanent entitlement.” So what? It’s long enough to do harm, and does more harm when it’s 100+ years at it now is; but it was harmful even when it was 14 or 28 years! “It was a limited monopoly granted to encourage creation.” Not really. In any case that is not the role of law and such a law is not just.
  • “If copyright cannot distinguish between reading and copying, it is no longer law. It is sabotage. The correct response is not to extend the licensing regime but to abolish it.” He’s right but if by “it is no longer law” he means “it is not just law” then he is right, but then it was never “law” since it was always unjust!
  • “Reduce copyright terms to five years.” This would be an improvement but he gives no reason for this arbitrary term, or for why it should not be zero! Optimal Patent and Copyright Term Length; Tabarrok and Murphy: Why Are US Drug Prices So High? ; Tabarrok: Patent Policy on the Back of a Napkin
  • “In England, our copyright law remains, for the moment, a little more cautious in its absurdities. The Copyright, Designs and Patents Act 1988 does not recognise the American fantasy of “transformative use,” though our courts have sometimes flirted with the concept under the guise of “fair dealing.” There are limited exceptions for quotation, criticism, parody, and research—but none that comfortably encompasses the training of commercial AI models.” See this quick Grok analysis. The author seems confused in his apparent criticism of the US doctrine of “transformative use”. He seems to agree that training an AI on existing books should be permitted. Thus he should want the US doctrine of transformative use because it helps make it more likely a use has a defense from copyright infringement, by qualifying for fair use treatment. He should want this for UK “fair dealing” standards too. As Grok explains, “U.S. fair use is more flexible, allowing AI training to be evaluated holistically, even for commercial purposes, if transformative and non-reproductive. UK fair dealing is rigid, requiring AI training to fit s.29A’s non-commercial research purpose, making commercial uses (e.g., commercial chatbots) ineligible without licensing.” So instead of criticizing the “transformative” use idea he should want it broadened and applied also to UK law.
  • “If we had a real Parliament, it would repeal the copyright statutes and begin again.” Better yet–abolish copyright. “If we had judges with imagination, they would rule that language cannot be owned. But we have neither.” It is not the judges fault at all here. They are stuck; instead of doing justice (common law), they are tasked with interpreting unjust and arbitrary legislation. ; Another Problem with Legislation: James Carter v. the Field Codes
Share
{ 0 comments… add one }