Who owns the copyright to an AI creation?

That artificial intelligence (AI) is now capable of producing products that can compete with “works of literature, science or art 1” no longer needs to be questioned. ChatGPT and DALL-E have become household terms in recent months. Almost everyone has tried these computer programmes (because that’s just what they are) by now, only to be amazed by the results. To wit: easy-to-read Dutch texts and beautiful pictures, apart from the occasional anglicism or a sixth finger on a hand.

a right for the machine?

When people produce something like this, the product is easily subject to copyright. As soon as it can be said that a product contains its own original character and the personal stamp of the creator, it meets the so-called “work-test” and protection is a given. So does a machine now get this protection?

We can quickly answer to that: No, a machine cannot own a copyright. The Dutch Supreme Court closed that road in 2008, in the famous Endstra judgment, by explaining what the above-mentioned elements of the work-test mean. The problem lies mainly in the element “personal stamp of the creator”. According to the Supreme Court, this means that the work must contain “a form that is the result of creative human labour and thus of creative choices, and which is as such the product of the human mind.”

A photo creation by DALL-E and a text by ChatGPT have no such form. Sure: both computer programmes have calculated themselves silly to create a form, but that form is neither the result of creative human labour nor a product of the human mind. So: exit DALL-E and ChatGPT as copyright owners. (Not to mention, of course, the fact that the law does not recognise them as persons, as legal subjects, as a result of which they cannot own any rights to begin with).

a right for the makers of the machine?

Their creators then, perhaps? The writers of the code underlying ChatGPT and DALL-E? No, those programmers do not own the copyright to the AI-creation either. They do, however, won a copyright to the result of their programming: the computer programmes they wrote are undoubtedly copyrighted. But once these programmes are running and, as a result, certain concrete design has been produced in text or image, we no longer attribute this to the programmer. They may have created the possibility that these programmes are capable of producing copyrightable works. But such a work is only protected once ot is there: when there is actually a sensory perceptible concrete design. With this concrete design, the programmers have nothing to do – that was the work of the machine.

the instructor then?

OK – not the programmer(s) then either, but what about the instructor? The person that gave the prompt to the programme that produced the work? Here we are already getting a lot closer to a suitable candidate. Not yet close enough, probably, but we can fix that… More on this in the next instalment of this series.

(The image accompanying this blog was created by Dall-E from the instruction: “oil painting depicting a robot that is confused about copyright”)

  1. the description of a copyrightable work in the Dutch Copyright Act (Auteurswet) ↩︎
Who owns the copyright to an AI creation?