Who is the author in artificial intelligence, if there is one?

By anthropomorphism, we naturally tend to speak of AI as a person in his own right, which it isn’t (or at least, not yet!).

However, by virtue of a universally recognized principle, only a natural person can be qualified as the author of a creation or work within the meaning of copyright law.

As a result, many questions arise in connection with “AI creations”. The US Intellectual Property Office (USPTO) was challenged last year by an applicant who declared an AI to be the author of a work. The application was rejected precisely on the grounds that only a human being could claim ownership of a copyright; however, this stance had the merit of setting a precedent and opening up a dialogue between lawyers and creators.

By way of analogy, in 2016, an American judge dismissed a case brought by the animal rights organization PETA against the exploitation of a selfie taken by a monkey, on the grounds that even if the monkey had taken the photos “independently and autonomously”, it was impossible to proceed with the lawsuit, given that animals do not have legal standing and therefore cannot sue for copyright infringement. The same applies to the paintings by “Pigcasso”.

Thus, it is essential to examine the role of the human in the creation of a work of art. If the artist has used an AI algorithm as a simple tool to create the work of art, and has made important creative decisions in the process (for example, by writing a very detailed prompt), he or she can be considered the resulting owner if the other conditions of protection required by copyright are met. On the other hand, if the AI algorithm has generated an image or a result autonomously, with little or no human intervention, it is not certain that a work within the meaning of copyright can be claimed.

For example, a piece of music created entirely by AI would not be protected by copyright, so anyone could freely distribute it. Controversy is still rife today, particularly if the creation of a new category of intellectual property rights should be necessary? Instead, the authors advocate rather to broadening existing concepts in positive law; indeed, law follows technology and must adapt to it, not the other way round.

What is certain is that AI will never be an author. Firstly, because an algorithm is not a human being as the law requires, but also because the works “ingested” by the AI for its training are not exploited in the conventional way as an artist would: as it learns, the AI deconstructs the content of these works in order to build a model based on their common specificities. They are used not for themselves as unique elements, but for their informational value. What the AI uses in its creative process is not so much the primary work as the set of characteristic features it has identified, taken separately (in other words, the AI doesn’t “see” the work as a human being does in an art gallery through its 5 senses, but reads a whole series of information in mathematical form that describe the product pictorially, somewhat like in the Matrix™ movie).

We could also push reasoning to the absurd and ask whether a total deconstruction of works by the algorithm, then a recombination of these in the form of statistical data, could create a new work ab nihilo with an original copyright or, on the contrary, added to all the pre-existing rights of its contributors?

Authorship can be manifold: very often, users of free versions of artificial intelligence tools “assign” the right so that their ” prompt” may be reused for algorithm training purposes, as well as the result of processing the data inserted by the algorithm. Thus, the author using AI tools for his creation will never be absolutely certain of being able to control the communication of his work to the public and its distribution, insofar as AI tool publishers guarantee neither the uniqueness nor the exclusivity of the result resulting from the use of the artificial intelligence tool.

Rather, it is the economic aspect that is currently being called into question, not least because of the investment and cost involved in developing and using these tools. The current behemoths of Artificial Intelligence feed these tools through their users’ interactions, and therefore monetize them. This point will be developed further in a future article.


In conclusion: due to existing legal uncertainties, it is not possible for an agency using artificial intelligence tools to guarantee exclusivity on the result of work to customers, nor the complete transfer of rights, insofar as it is not known who exactly is the author of the result (if there is even one).