There is no copyright on 'works of art' created entirely by artificial intelligence (a 'generative AI programme'), without creative human instruction and/or without creative human post-processing. Copyright protection requires 'a form that is the result of creative human labour and thus of creative choices, and is thus the product of the human mind', the Dutch Supreme Court ruled in 2008. It is still unclear how detailed and creative the instruction to an AI programme must be for the result to be copyrightable. There is no experience with that yet, no case law or comparative material.

Consider, for example, the work A Girl With Glowing Earrings, created by German artist Julian van Dieken using the AI programme Midjourney. Suppose the instruction to Midjourney was, as Laurens Verhagen suggests in Dutch newspaper the Volkskrant: 'Create a portrait of a beautiful girl with full lips and blue eyes and glowing orange earrings in the style of Vermeer'. Then, in all likelihood, that instruction is not sufficiently detailed and creative to create a copyright on the result for the 'instructor'. If the instruction included the desired composition, position of the face, exact colour choice of all parts, shading and the like in detail, then it is much more likely that copyright of the instructor on the final result would be assumed.

In that case, we would come close to the situation described in Article 6 of the Dutch Copyright Act: ‘If a work is created to the design of another and under his direction and supervision, he shall be regarded as the maker of that work’. If a work is created by an AI programme 'according to the design' and 'under the direction and supervision' of an instructor, that instructor will be considered the creator and right holder. The question is how detailed that instruction has to be for a copyright to arise with that instructor. Whether that instructor is obliged to disclose what his instruction was, is the issue discussed below.

Not only creative instruction can create a copyright, a little creative post-processing can certainly do that too. There is much more experience with that aspect. Van Dieken said he used Photoshop in addition to Midjourney to digitally fine-tune the portrait, Verhagen told the Volkskrant. 'Digital fine grinding' sounds like a rather technical operation, without much creativity. Now what if Van Dieken were to argue that he creatively reworked the image? Then he could claim copyright. It is very common to have parts of a design process done by a computer and to take parts in hand yourself. It is also common to slightly rework existing work that is no longer protected, think of classical music that is no longer protected, and then claim copyright on that reworking. A not too drastic adaptation can already create copyright.

What if Van Dieken were to claim that he gave very detailed creative instruction and also did extensive post-editing with Photoshop? Does he then have to prove that in order to claim copyright on the (edited) result? What must show the creativity required for copyright protection? On this, see the Supreme Court in 2008: ‘What matters is a characteristic that can be known from the product itself. Therefore, the requirement that the creator consciously intended to create a work and consciously made creative choices may not be imposed, which requirement may, moreover, pose insurmountable problems of proof for those concerned.’

A photographer who has used an 'intelligent' camera also does not have to prove what creative choice he has made. The creativity must be evident from the photograph. And in the case of photographs, the threshold is very low: almost any photograph is considered copyrightable. Proving how the photo was created is not necessary. There is no 'duty of information' on an artist to show how their work was created.

This means that Van Dieken does not have to prove at all which creative choices he made at what stages of the production. He can simply argue that the end result is copyrighted and that he is the right holder. As long as it 'looks creative', that is sufficient. Someone who wants to exploit or imitate the image without his permission will then have to prove that the image is not protected. In case of an imitation that creates the same overall impression by similar creative features, a burden of proof also applies to the detriment of the imitator: he or she must then prove that he or she has not borrowed.

American artist Kris Kashtanova recently announced on Instagram that the registration of the copyright to the images in her comic book 'Zarya of the Dawn' was rejected by the US Copyright Office. This was because she had stated on social media that these images were created entirely by AI. Her copyright on the sequence of the images and the story was accepted. But what if she had not stated that these images had been created entirely by AI? Then there is little chance that any problem would have arisen.

In Europe, the phenomenon of 'registering' copyrights does not exist. Copyright arises automatically through the creation of a work of art and protection is only tested when a lawsuit is filed. Then the rules of evidence discussed above apply: the artist does not have to prove how the work was created, the 'creativity' must be evident from the work itself.

Can perhaps a duty of information about whose creativity is in a work of art be found in Dutch penal law? ‘He who falsely places on or in a literary, scientific, artistic or industrial work any name or any sign, or falsifies the true name or sign, with the intention of thereby making it plausible that that work would be by the hand of the person whose name or sign he placed thereon or therein, shall be punished with imprisonment of not more than two years or a fine of the fifth category.’ So reads Section 326b (1) of the Dutch Penal Code.

This penal provision is intended for art forgeries where, instead of one's own name, the name of a famous painter is applied to a work of art. Think of art forger Han van Meegeren who produced false Vermeers. However, it seems defensible by the letter of the provision that it also entails that you are not allowed to claim that something is entirely your work, when in fact it was entirely manufactured by AI. But this will only apply in extreme cases where there is indeed no creative input at all. And that, again, is very difficult to prove.

The conclusion is that someone who has created a 'work of art' with the help of AI can easily claim copyright on it under the current rules of evidence, as long as he or she does not mention that he or she did nothing creative themselves. And no one is obliged to provide information about the process of creation.

Is that undesirable? Is it undesirable that Van Dieken and Kashtanova can claim copyright on the images they created with AI? After all, that means mainly that they can prohibit the exploitation and imitation of those images. They cannot prohibit others from making similar images with the same AI. They can try to do so, but if those others can prove (easily) that they did not borrow but commissioned AI themselves, there is no copyright infringement. Unless the images by Van Dieken and Kashtanova have themselves become part of the AI's 'subject matter' again, and that large parts were borrowed from those images by that AI.

And that brings us to yet another problem, which is that anyone who uses AI to create images runs the risk that those images are ‘too similar’ to existing works of art that are still copyrighted. And that is quite a big risk, because then the burden of proof works to the disadvantage of the AI user. AI works on the basis of 'machine learning' where millions of images are copied from the internet and used for the AI programme's 'creative process'. If the result of AI 'happens' to be very similar to one of those millions of images that are still protected, it could be copyright infringement. Proof that it has not been borrowed cannot be provided, because by definition, AI has 'borrowed' precisely from all those millions of images, so probably also from the one image the result is very similar to. So copyright infringement will then be assumed, which can lead to an injunction and damages.

Because of this distribution of burden of proof, it is therefore wise to be careful when using AI-produced images commercially.

There is a good case for changing the rules on burden of proof and duty of information on the use of AI. Two propositions are put forward and explained below with the request that readers respond to them.

Thesis 1:

‘In the case of a work of art which has been produced in substantial part with the aid of a generative AI programme, the person who mentions, or causes to be mentioned, himself as the creator of that work, is obliged to state thereby that he has made use of such a programme.’

This thesis is in line with what seems to be common practice: artists do mention that they have used AI. Van Dieken and Kashtanova did just that. Attribution is a right, pursuant to Section 25 of the Copyright Act, but against that right, in my view, there is a duty not to falsely claim creative authorship. This standard can also be derived from what is said about name attribution in the Dutch Code of Conduct for Scientific Integrity 2018. That code also states in paragraph 35: ‘Be transparent about the method and working procedure followed and record them where relevant in research protocols, logs, lab journals or reports.’ An artist is not a scientist but, in my opinion, accounting for the use of AI should also apply to artists. Failure to disclose the substantial use of AI could also qualify as cheating or deception. This could tie in with the standards against unfair commercial practices and misleading advertising:

Article 6 (1) Unfair Commercial Practices Directive:

‘A commercial practice shall be regarded as misleading if it contains false information and is therefore untruthful or in any way, including overall presentation, deceives or is likely to deceive the average consumer, even if the information is factually correct, in relation to one or more of the following elements, and in either case causes or is likely to cause him to take a transactional decision that he would not have taken otherwise: […] (b) the main characteristics of the product, such as its […] method of manufacture’.

Article 3 (a) of the EU Directive concerning misleading and comparative advertising.

‘In determining whether advertising is misleading, account shall be taken of all its features, and in particular of any information it contains concerning: (a) the characteristics of goods or services, such as their […] method of manufacture’.

Thesis 2:

‘If the human creative contribution in respect of a product is challenged in court, the person claiming copyright in it is required to show where his human creativity lies. For example, by providing evidence regarding the creation process’.

With the advent of generative AI, the rule of evidence that a creation that gives the impression of human creativity based on the observable characteristics of the creation should be deemed to be copyrighted is no longer desirable. The big question is then what burden of proof may be placed on the alleged creator. Some creations will involve an easily documentable creative process with design drawings, drafts, and the like. In the case of a suddenly conceived tune, it will be much more difficult to meet this burden of proof. Witness statements from the recording studio could perhaps help. Of course, it is highly questionable whether artists should be subject to record-keeping obligations with regard to method and procedure as applies to scientists according to the standard quoted above (‘record in research protocols, logbooks, lab journals or reports’). Here, in some cases, the ‘insurmountable problems of proof’ that prompted the Supreme Court in 2008 to hold that creativity must be apparent from the product itself may be imminent. But times are changing fast.

See also: 'Robot art and copyright' (NJB 17 February 2023, in Dutch)