Artificial intelligence: “Copyright protects a precise creation, but not a way of creating”

Use of protected works, status of images created by generators… Law professor Alexandra Bensamoun takes stock of the legal problems created by the development of AI.

Professor of law at Paris-Saclay University, Alexandra Bensamoun is the author of a mission on the challenges of AI applied to the cultural sector. This study, commissioned by the Ministry of Culture in 2020, is part of a general awareness that all sectors, even visual creation, are concerned by the advances in these technologies. For the time being, while the balance of power is unequal between the actors, the legal ground is still fallow and the European orientation is moving in the direction of total data sharing, a liberal Far West which does not sufficiently protect artists and rights holders.

What concrete legal problems do image generators and artificial intelligence pose to visual creation?

AI has begun to invest in the creative and cultural sectors, both at the stage of creation and at that of the dissemination of protected content. The use of AI technologies in artistic creation raises two types of questions, which also follow the creative process. First, upstream, we can wonder about the fate of the works that will feed the AI. Indeed, generative AIs often use “machine learning” technology. This is a technology that mimics neural networks; the reasoning there is inductive, based on the experience acquired through the ingestion of content which may be works protected by copyright or related rights. Then, downstream, there is obviously the question of the status of theoutput, “outgoing” artistic creation. Is it a work of the mind, as such protected by copyright?

Can copyrighted works be used to “feed” AI?

European law has imposed in its latest directive on the subject – 2019 /790, known as “Digital Single Market”, of April 17, 2019) – two exceptions of text and data mining [fouille de textes et de données, soit le processus suivi par les IA pour extraire des connaissances ndlr], where the application of the law is suspended: the first is exclusively for the benefit of academic research, the second is very broad, for all uses (including commercial); it corresponds precisely to the described hypothesis of learning [par les intelligences artificielles]. But its very broad scope has been counterbalanced by a possibility for rights holders to ‘opt-out’ [avoir le choix de retirer ses œuvres]. The latter can, via a machine-readable process, signify that they do not wish their content to be mined. This may explain why some platforms solicit creators to ask them if they accept or withdraw. We can therefore freely use protected content to “feed” an AI, in terms of learning, only if we fall within the framework of one of the exceptions.

Can we market images “like” Picasso or Monet or Annie Leibovitz?

We see flourishing in different artistic sectors creations “in the manner of”, as if borrowing the style of an author, for example The Next Rembrandt, in the manner of the Dutch painter, or Daddy’s Car, songs like the Beatles. An author’s style cannot be reserved. Copyright does not protect ideas but formal achievements. We protect a specific creation, but not a way of creating. In general, intellectual property does not allow the reservation of ideas, methods, theories, but only their translation into the sensitive universe of forms. Also, it is quite possible to create creations “in the style of”, which are content to borrow the style without retaining the form of a work. This is exactly what the AI ​​does: as part of the search, it deconstructs the content to extract trends. The reconstruction does not reproduce the form elements of the excavated work, protected by copyright.

However, all of this must be qualified: firstly, it is possible, if the work is still protected by law, that the author objected to its content being searched by having exercised his opt-out; then, if copyright is not intended to apply in this case, other legal mechanisms could be invoked (but with less efficiency), such as parasitism, for example.

What status do images created from generators have? Are these original creations?

First of all, a distinction must be made between the creations generated by an AI and the creations assisted by an AI. Only the former are of interest to us here. Indeed, when AI is only a tool in the hands of the creator, classic copyright law is intended to apply. The use of an instrument, however sophisticated, is irrelevant to access to protection.

To answer the question, it is necessary to make some reminders. Mainly, the object of copyright protection is the work of the mind, that is to say a creation of original form, which bears the imprint of the personality of the author. The conditions of access to this protection refer to the need for there to be an author, a natural person. Indeed, the creator can never be a legal person or an animal – we remember the litigation in the United States, about the monkey who had taken a selfie by stealing a camera… However, if the creation is generated by a IA, who would be the creator-author-natural person? The user ? But if he was content to press a button or provide a few ideas, he could not mark the work with the imprint of his personality. Is it rather the designer of the AI, who finally drew the field of possibilities?

Finally, is current copyright still adapted to the arrival of creative AI?

The European Parliament adopted a resolution on 20 October 2020 on intellectual property rights for the development of technologies linked to artificial intelligence, inviting the Commission to ask itself the relevant questions in this area. Obviously, this is a political choice, which will no doubt have to be made on a European scale: it can be decided that copyright will receive these new kinds of creations, perhaps by forcing a little about the conditions of access to protection or by adapting them; one can still consider that it is relevant to create a special law; we can finally support to exclude these realizations from the reservation.

For the time being, the European Union is first seeking to secure uses and imbue its values, with the proposed regulation, AI Act, of April 2021, currently under discussion. The focus will then be on the law of liability, to manage the risks created by the use of an AI and the compensation for damages. Perhaps the next step will target intellectual property… To be continued, then.