On August 26, 2022, Jason Allen won the Colorado State Fair Fine Arts Competition. At first glance, this might be nothing special – every competition will eventually lead to one piece being crowned the winner. This assessment changes once you consider that he submitted a piece of art that was not drawn by him, but created by AI. Jason Allen used the AI Midjourney to create his artwork, labeled “Jason Allen via Midjourney – Théâtre D'Opéra Spatial“. Convincing the jury, it awarded the first prize. This created a huge backlash among artists and also among lawyers.
Artists, on the one hand, see the threat of dying artistry. On the other hand, there is no legal foundation for this yet. It is undefined who owns AI-generated artworks and thus, is allowed to sell it or to submit it to competitions. Not least does AI-generated art carry a risk of disinformation and deep fakes that could be a cause for concern. For lawyers, these copyright issues arise, as does the question of whether AI-generated work is patentable.
As much as the frustration among other contest participants is understandable, because they submitted self-made artwork, is it justified when considering that Jason Allen put 80 hours of work into the AI-artwork? Is the time spent a determining factor in judging an artist’s performance? Is Jason Allen the one who is allowed to submit the AI-artwork to a competition and collect the $300 prize money? The question of who owns works generated by AI – and who can win prizes with them – is more relevant than ever, because the increasing use of AI cannot be proceeded in a legal vacuum. Pretending that the art competition took place in Canada and answering all those questions from a Canadian perspective, this vacuum could arise if closer analysis showed that the Canadian Copyright Act did not apply to art works created by AI.
Who are the stakeholders in this dispute and who is the owner of the work?
There are a few different possible owners of an artwork created by AI. In the Canadian Copyright Act, owners are referred to as authors. The author is considered to be the first owner of a copyright, pursuant to sec. 13(1) of the Canadian Copyright Act. First, the person who wrote the code, the AI’s developer, could be the author and thus could own the copyright. But the user of the AI, the person who has caused a particular project to be generated, might also be involved. Another possible author could be the AI itself. If none of those players deserves ownership of the work, the copyright can remain in the public domain – and thus no intellectual property applies to it.
Without the developer’s work of developing the code and training the AI, the outcome would not be possible. According Haochen Sun in 2022, the developer has the primary contribution to the outcome’s creation. However, in his paper awarded by the Canadian Bar Association in 2018, Pascal Visentin explained that it is the user who provides the data and causes the AI to create the work. Similarly, Mark Perry and Thomas Margoni from The University of Western Ontario argued in 2010, emphasizing that the developer did not initiate the peculiar work.
This initiation exists on part of the AI’s user. By selecting an appropriate prompt and starting the program, the user initiates the generation of the work in question. Selecting the right prompt for the desired work can take time and effort. However, compared to creating a work without the use of AI, Perry and Margoni do not see why the user should be rewarded for these actions.
They argue that AI-generated works should remain in the public domain, since “an AI” is not an object that can be rewarded by copyright. And, they propose a cautious approach to property rights, recognizing that AIs are capable of creating myriad works. They are concerned with a breakdown of property rights protection in the form of a market failure, if all these works were given property rights. In their answer to the process of Consultation on a Modern Copyright Framework for Artificial Intelligence and the Internet of Things that Innovation, Science and Economic Development Canada (ISED) published in 2021, the Canadian Federation of Library Associations and Canadian Association of Research Libraries agrees with this finding. They add that human authors might be disadvantaged because they are not able to produce this “mass output”.
This public domain approach seems to overlook the fact that copyright in AI-generated works will not be treated uniformly. Rather, it divides AI-generated works into two parts: those for which it makes sense to keep them in the public domain and those that require copyright for protecting the owner’s rights. For some pieces, it might be fair to keep AI-generated works in the public domain, especially if the AI is used primarily to create the work and not just to revise it. But imagining a work that is corrected by an AI and thus falls into public domain, although a human holds the biggest part in its creation. This does not correspond to reality and does not protect the input. To solve this problem, Sun proposes a distinction between AI works with human contributions, which should be rewarded sui generis IP rights, and those without such contributions. For these sui generis IP rights, Sun suggests, a shorter protection period of ten years could be considered – compared to the existing term of 70 years after the death of the author. Meanwhile, AI works without human contributions would still be treated as mentioned above, assigned to the public domain.
Since, under sec. 5(1)(a) of the Canadian Copyright Act, copyright owners can only be human beings, the machine that created the work is eliminated as a possible owner. In contrast, this did not prevent the Canadian Intellectual Property Office (CIPO) from registering the AI RAGHAV as a co-author. Ryan Abbott, a lawyer who represented an inventor whose AI-generated work was denied by the United States Copyright Office, shares the CIPO’s opinion in an interview in 2022: AI should be acknowledged to be able to produce “functionally creative output”. Its protection “with copyright is vital to promoting the production of socially valuable content”.
By registering an AI as a co-author, the CIPO has taken a path that seems to strike a good balance. The work is not entirely unprotected, but neither is it attributed to anyone who does not deserve the credit. Together with Sun’s proposal to modify the protection term, the problem of the crowded market for copyrights appears to be headed towards a convincing solution. Yet, as long as the transfer of copyright to others than humans remains incompatible with the current Canadian Copyright Act, some authors contemplate the pros and cons of an eventual change in the law.
If the art competition had taken place in Canada, would the Canadian Copyright Act apply for AI-generated art?
According to the CIPO, copyright aims to protect copyright holders while “promoting creativity and the orderly exchange of ideas“. The World Intellectual Property Organization stated in 1997 that copyright protection is intended to “maintain a balance between the rights of authors and the larger public interest”. And the Supreme Court of Canada (SCC) views copyright as a “reward for the creator” in the decision CCH Canadian Ltd v Law Society of Upper Canada of 2004, which defines the essential requirements for copyright.
The Canadian Copyright Act protects works that are “original” (sec. 5(1)). According to the SCC in its 1957 decision Kilvington Bros. Ltd. v. Herbert Goldberg et al., a work covered by the Canadian Copyright Act must be an “original expression of thought of its originator, that originated from him and that he did not copy”. A little thought experiment can illustrate how the Supreme Court’s formula can be applied to art work created by AI: when applying this formula to a work of art created by an AI, the requirements seem to be met. The data that is merged by the AI does not create a simple copy of the input, but rather an entirely new work is generated.
In its 2004 decision, the SCC held that a work is covered by the Canadian Copyright Act if the creation of the work is an “exercise of skill and judgement”. The SCC further requires an “intellectual effort” to be considered “original”. The SCC cites as a negative example the alteration of the background which is not considered “original”. If this simple change, initiated manually and taking little time, does not produce an “original”, this indicates for me that a fortiori merging data into a new source, requiring only a few words entered as a prompt and clicking “start”, is not an expression of intellectual effort. Lastly, the process “must not be so trivial that it could be characterized as a purely mechanical exercise”, according to the SCC. What, if not AI-generated content, is the epitome of a mechanical exercise?
For the time being, AI-generated content does not fall under the Canadian Copyright Act. Much suggests that AI and AI-generated work should be covered by the Canadian Copyright Act in order to ensure transparency and legal certainty for developers and users of AI. When adapting the use of AI to the Canadian Copyright Act, the goal should be to strike a balance between protecting creativity and skills, on the one hand, and not restricting the use of works that do not deserve such protection, on the other, regardless of how the legislator decides when it comes to who is entitled to copyright.
To avoid legal uncertainty, the question of who is considered the owner of the AI-generated work should be answered by the legislator. Practicably, it falls in the public domain if there is no human contribution to the work. And for works with contribution a shorter protection period might be examined – be it for the developer, the user or the AI itself.
By giving this legal framework, AI-generated art participating in competitions can be assessed on a legal basis in the future, which might reduce discomfort among artists and lawyers.