Washington DC In a report published on January 29, the American Copyright Office (US Copyright Office, or “USCO”) became interested in copyright protection of the results generated using artificial intelligence systems (Ia) generative. He thus examines the content generated downstream by a generative AI system and not the harvest data upstream to cause AI models, the study of which is announced for the next months.
The USCO first considers that existing American legislation is sufficient to treat the creations generated by AI, so that it is not necessary to modify the law. He then examines the degree of “Human contributions” necessary for the works generated by AI to be considered as works protected under copyright. Four main recommendations are stated, according to progressive reasoning. First, content entirely generated by an AI system cannot be protected by copyright. Then, the use of AI to assist human creativity does not affect eligibility for protection by copyright, provided, however, that the results are “Enough” controlled by humans. Third point, the question of whether the human contribution to the results is “Sufficient” fall under casuistics and must therefore be analyzed on a case -by -case basis.
Finally, the report is pronounced on the protection of content generated using prompt (” instructions “). Although not final because given “In the state of the technique”, Its fourth recommendation is nonetheless clearly stated. The American office observes, on the basis of the current functioning of the generative AI, which prompts alone do not provide sufficient control for us to dissociate what the artist has created from what has been generated automatically. Two reasons are put forward: the first is due to the fact that technically it would be impossible to understand the influence of a prompt or a series of prompts on the results generated by the generative AIs which are based on complex systems and of which The models have billions of parameters. To this complexity is added the unpredictability of the results as soon as, in many AI systems, the content generated can vary from one request to another, with a prompt however identical. The office thus concludes that, taking into account the current state of generative AI technology, prompts alone do not provide sufficient human control to make users the authors of the results. And to add that the lack of protection by COPYRIGHT extends to very detailed prompts and containing the expressive elements desired by the user. Nothing nevertheless prevents the benefit of protection to the authors who have selected, coordinated and arranged the material generated by the AI in a creative way, the creative part intervening then upstream and/or downstream of the prompt.
And in France?
The solutions of the American office are transposable in France. No judicial decision has yet been rendered. The debate is therefore open. One certainty, however: with regard to French law, the author of a work of the mind within the meaning of copyright can only be a natural person, the only one capable of conceiving of an intellectual creation that is specific to him, Reflecting his personality, by manifestation of free and creative choices. Content, whatever it is (text, image, video, music, etc.), generated by an artificial intelligence cannot therefore be protected itself under copyright.
What should however be discussed is the question of whether, by a series of prompts expressing free and creative choices, an author can, within the framework of an active iterative dialogue with the machine, design a singular work bearing the imprint of his personality. In doing so, the author would limit the probabilistic results proposed by the AI to reduce it to the status of simple material performer of its directives and thus benefit from the protection of copyright. In our view, a positive response is essential.
