The notion of authorship in the era of generative artificial intelligence

Generative artificial intelligences are transforming our relationship with creation and attribution. Between technological amnesia and the opportunity for renewal, they force us to rethink the notion of authorship.

Machine Creativity and the Question of Attribution

Generative artificial intelligences feed on the entirety of accessible human knowledge. Contrary to common belief, they can create new knowledge from the interconnection of diverse knowledge. Creation and innovation emerge precisely from the novel connection of pre-existing things that were previously unconnected.

This capacity for interconnection forms the basis of disruptive systems. Uber, for example, connected the need for individual transportation with digital platforms. In the artistic field, Stanley Kubrick, with 2001: A Space Odyssey (1968), connected academic classical music with the anticipation of the future, something that had never been done before. He also connected NASA’s space photographs with the fictional representation of space in cinema, thus creating a representation of technology that no one had imagined before him.

The real turning point of deep learning artificial intelligences manifested itself with AlphaGo’s creativity in 2016. The machine defeated humans through new openings and sequences of moves that it invented, which humans had never imagined before, and from which humans draw inspiration today. We can therefore consider that the machine is the author of something. Yet, it has never been compensated for its work with author status! We tend to deny it these capacities in terms of authorship and, paradoxically, to appropriate this authorial work when it helps us write, makes relevant suggestions to us, even though it’s the machine that accomplished it. The same applies to image creation.

Copyright and the Expression of Ideas

One could object that when we submit a prompt to ChatGPT and ask it to write on a subject, successive exchanges with the machine allow us to make it write what we want. In this case, aren’t we legitimately the author? In reality, no. In France, copyright doesn’t protect ideas, it protects the expression of ideas. This distinction is fundamental. Copyright differs from patent law. Within the framework of copyright, it’s the expression of ideas that matters.

When there is plagiarism, what is examined is whether there has been copying, not of the idea, because the idea cannot receive copyright protection. If someone has taken someone else’s ideas and expressed them completely differently, it’s not plagiarism and copyright doesn’t protect against that. This distinction is subject to interpretation, of course: what constitutes the idea itself versus its expression? Where exactly is the boundary between the two? This subject remains open to interpretation and, in case of disagreement, will be decided by a judge.

And justice is not truth. It constitutes an attempt, necessarily subjective, to construct a performative narrative. This narrative is not linked to an immanent truth; it’s a narrative, justified by its sources, by the genealogy of its construction, which thus finds its legitimacy. What demonstrates this well is that a new source, even much later, a new DNA sample or a new testimony for example, can call the entire narrative into question and lead to producing a new narrative. We see this with the twists and turns of the Gregory case, for over 40 years.

Anteriority and Traceability as Foundations of Legitimacy

The subject of copyright raises an essential question: that of anteriority, traceability, the recognition that a source exists, that we can name it, that we can identify it in identity, in space, in time, in culture. This is what allows justice to construct its narratives: material evidence provided by someone, or sworn testimony from someone else. If this same witness, a few years later, reveals that their testimony was false, this can change the narrative. A narrative, to be legitimate, must signify that it doesn’t invent, but that it relies on sources.

We rarely question the validity of sources, except sometimes in the context of legal instrumentalization or a legal battle, if we claim that someone is lying for example. But ultimately, after agreement, always very subjective, on the probity of the source person, we will rely unconditionally on them to develop a narrative. In art history, the influences of artists on one another are recognized. Artists don’t hide it: they have sources, they can cite their master, and they also create, due to their singularity, something of which they will be fully the author, while citing their sources.

This was the case, for example, for Moby who, in 1991, created the track Go, his first major public success, which truly launched his immense career. He explicitly cited his source, his inspiration, his framework: the theme music from the series Twin Peaks by David Lynch, composed by Angelo Badalamenti. Moby didn’t hide it, and it can be heard: there is a reappropriation of the Twin Peaks track.

Another subject in the field of music, if we take a sample, that is, a piece of recording from another music, and we transform it sufficiently to make the link to the source impossible, we then have no copyright to settle, and no obligation to mention the source. We are free to do so, of course, but this in no way constitutes an obligation, and above all it doesn’t force us to compensate the authors. This is precisely what happens with generative artificial intelligences and their relationships to sources, because they transform them so much that the source is no longer recognizable; it’s as if there were no anteriority, no links, no genealogy.

Copyleft Licenses and the Collective Memory of Free Software

In the field of software, copyleft licenses are what make the vast majority of computers, the Internet, data centers, and artificial intelligences themselves work. These free software programs have identifiable successive authors, to whom reference is always made. These authors have accepted free sharing, but without being made invisible in their role and in what they have contributed to the software, and this also applies in the future, this is the established rule. Everything is traced. This approach makes it possible to signify anteriorities and, as in judicial narratives, to construct embodied collective narratives, anchored in a collective reality identified as such, by following the chain of anteriority.

Sean O’Brien, founder of the Yale Privacy Lab, alerts us to this question: “When generative AI systems ingest thousands of free software projects and regurgitate fragments without any provenance, the cycle of reciprocity collapses” (ZDNet, 2025). We don’t escape, in the field of free software, from the identification of human agents who are indeed considered authors, recognized as such, even though they agree to make their work accessible, available, freely shareable. In the GNU-GPL license, the principle remains that the conditions be respected identically: we trace anteriority.

This traceability doesn’t aim to claim financial rights, but to better understand how systems are built and the history they contain. Better understanding this, inscribing it in the code itself, allows for a better understanding of the diversity of what constructs reality. In the evolution of systems or in their updates, this allows us to be much more relevant, effective, and legitimate in relation to the human history that lies behind.

The Technological Amnesia of Artificial Intelligences

When we ask an artificial intelligence to create computer code for us, it draws inspiration from open source, necessarily, from what has been shared. Insofar as it has been fed with this data and then does its own reasoning, in its own way, with extremely complex weighting systems constructed by itself through billions upon billions upon billions of iterations and backpropagation, it doesn’t have, at least until today, the capacity to know exactly where this knowledge comes from, because everything is reconstructed. Ultimately, human vision, for example, which we believed to be an objective perception, is also in reality a permanent reconstruction from partial and disparate elements, human memory too, which has absolutely nothing to do with a hard drive. Some artificial intelligences like Perplexity trace the genealogy in relation to internet searches, but not in relation to what is intrinsic to their own language, which allows them to “understand” their searches. Their understanding of the world comes from what inspired them, but they can’t envision it otherwise than as their own understanding of the world, just as we envision our vision as our own vision, which moreover is objective, and not as a reconstruction in a process that is completely opaque to us and will remain so.

This situation resembles the musician who has so transformed the original sample that it has fully become their music. Perhaps they themselves no longer make the connection and are no longer even capable, so much has it been transformed, modeled, of remembering what the source was! They become its author, and we no longer have traces of the history. As Sean O’Brien expresses regarding computer code generated by AIs at our request: “The code is freed from its social contract and developers can’t give back because they don’t know where to send their contributions” (ZDNet, 2025). This phenomenon that he calls “license amnesia” applies beyond the software domain alone.

This indefinition of the author raises questions, of course. We are wiping the slate clean of a past that has completely nourished us, but of which we are no longer capable of knowing what the history was that makes us here today. We can, with regard to machines, take advantage of this confusion and appropriate creations made by machines, presenting ourselves as their authors. Some people wish to mention the provenance, to indicate that the text has been reworked by artificial intelligence or written entirely by it. But in reality, more and more things are. And since artificial intelligences can’t trace it themselves, even when we cite a text trying to provide the best traceability, this text itself may have been largely written by an artificial intelligence, without our knowledge. So, there it is, the fusion is complete.

Even a “pure approach” of writing without AI therefore becomes quite illusory today, given the very important place of artificial intelligences, whether we know it or not. This reminds me of listening to a track, where we hear a sound and we say to ourselves “ah, it’s a Moby creation,” without knowing that it’s something he may have completely taken from someone else, completely transformed. We are living through a very particular stage of a new construction based on all human knowledge, which intrinsically loses the memory of its own history, whether it’s the memory of machines and even the memory of humans.

Voluntarily Reconstructing the Genealogy of Thought

What should we do? Our history is at stake, and the political and anthropological validity of our history too. This is why I took the example of legal construction. Since we can’t stop this movement, and there is no, in my opinion, technical solution in sight that would allow us to recover anteriority, because generative artificial intelligences are not built this way, due to their deep learning methods, which is what paradoxically gives them all their relevance, I think that for this movement not to lead to a cancellation of our history, it seems important to me that we ourselves think as regularly as possible to name the links, the references, and especially the human links.

With artificial intelligence, for example, naming ChatGPT, Claude, or Perplexity as co-authors of the text wouldn’t make much sense. We can do it, but it doesn’t have temporal stability, like an author who is born at a certain moment in a certain culture. We can’t find somewhere “the works of ChatGPT.” Let’s not seek, it seems to me, to impose on artificial intelligences something that is intrinsically impossible to obtain. On the other hand, as human beings, let’s cite each other much more. Let’s take up again and ask artificial intelligence to help us do it. Let’s give value back to the links between us humans, through our knowledge.

Much more than before, let’s take the trouble to know each other better mutually, to recognize ourselves as human beings, and thus not lose our own history. To do this, we can moreover have artificial intelligences help us, because they are precisely in immense capacity to put things in connection, due to the fact that they have this almost ubiquitous knowledge. Let’s ask them to cite authors who have worked on a subject connected to ours. Let’s discover new authors through this, let’s cite them, let’s put ourselves in connection, in empathy, even more with one another, thanks to the capacities that artificial intelligence brings us.

In doing so, perhaps artificial intelligence will even bring us tools to better inscribe our creations, our ideas, our intuitions in collective human history, with respect for the genealogy of ideas, of their expressions. Let’s move toward as much shared subjectivity as possible. We must, in my opinion, do it voluntarily, because if we don’t voluntarily signify the links between us, human beings, we will end up with a sort of absolute consensus, where we would have the impression that there would no longer be any author at all, that everything would be an average of everything and that there would no longer be any point of view, nothing but an immense consensus, so dangerous, because this is totalitarianism.

Avoiding the Obscurantism of Unique Consensus

This prospect of global consensus must be avoided at all costs, because it’s groupthink, it’s scientistic obscurantism, where they want to make us believe, and could make us believe, insofar as these machines are operated by industrialists, that there is only one vision of the world that constitutes its truth. O’Brien specifies: “Once AI training sets subsume the collective work of decades of open collaboration, the idea of the global commons risks becoming a non-renewable resource, exploited and never replenished” (ZDNet, 2025).

I strongly and truly believe that if we are attentive, rather than only citing the human authors from whom we have explicitly drawn inspiration, if we ask artificial intelligences to help us connect with human authors we didn’t even know but who are connected with us, we will be able to enable ourselves to reweave a history that is ultimately larger, more respectful, more diversified, more interesting, more enriching, more existent.

Thanks to artificial intelligence, we will discover authors who work on the same subject as us, or on other subjects but who have approaches similar to ours, authors therefore that we didn’t know and who will enrich our vision of the world. We won’t appropriate them without saying so, but signify these links more than once rather, even more than before. On a purely personal level, I see this in my own practice: thanks to the discoveries that artificial intelligence allows me to make, I implement what I’m talking about here. Thus my discourse and my proposals are not at all theoretical, but fully practical.

An Opportunity for Intellectual Renaissance

I enrich my knowledge of other authors thanks to artificial intelligences, and I put myself in relation, I put my thought in relation more than before with the thoughts of other authors. I of course have my own physical library that I can consult on the shelves at home, or the libraries I can go to, which already allow me to do this. But artificial intelligence allows me to access other libraries whose existence I didn’t even know of in other countries and, even within libraries, other notions, different understandings.

Artificial intelligence, contrary to what we believe, absolutely doesn’t have access to everything. It only has access to what has been made available to it on the internet. Nevertheless, it still has access to an enormous amount of things. Here is my suggestion and the great opportunity I see for giving human authors an even more legitimate place: a deepening of the links between human beings, in the present and in the past, and a reconstruction of a genealogy, if not an etymology of human thought.

We must explicitly have the generosity and openness, from which we will receive much, to voluntarily activate more links. If we are not proactive, in the writing of our texts, in our citations, more important than before, of other authors we didn’t know, we will move toward a potentially serious amnesia. This is why I formulate this proposal: let’s use artificial intelligences not to erase our history, but to rediscover it, enrich it, and share it with more generosity than before.