Who owns the copyright in AI-generated art?

‘AI’ artwork generators like Stable Diffusion and DALL-E are taking the internet by storm. They offer users the ability to rapidly create incredibly detailed and powerful images based on prompts — which can be anything at all. Want to see an astronaut cat riding a surfboard across a fantasy landscape in the style of Picasso? No problem. However, when your artwork has been generated, who actually owns it — yourself, the ‘AI’ or an existing copyright holder that your artwork is based on? Here, our attorneys Arthur Roberts and Alain Godement break down what exactly ‘AI’ artists are and what the law states regarding copyright ownership.

What is an ‘AI’ artist?

Stable Diffusion and DALL-E are machine learning models (and supporting software) capable of generating digital images from image and text prompts. The output images that these tools can generate (which include the header image of this article, generated using Stable Diffusion) are surprisingly real-looking — in that it’s believable that the work could have been generated by a real person.

A machine learning model is a collection of data that has been trained on training data. The training data in this case is hundreds of millions of tagged images, scraped from the internet. Most of those images are likely protected by copyright. In Stable Diffusion’s case, the machine learning model itself is surprisingly small — 4GB of numbers. The machine learning models themselves do not comprise any ‘images’ in any sense that could be understood by humans.

While the output of these ‘AI’ artists can appear very human-like — and perhaps even show understanding of the prompts they receive — they are in no way human. Such models are trained for the single purpose of receiving prompts and generating images — so this is not the beginning of any kind of Artificial General Intelligence (AGI) — and you can sleep soundly knowing that robots will not be taking your job any time soon.

What is copyright and how does it arise?

Copyright — at its most fundamental level — is an intangible property right (or intellectual property right) that may subsist in certain works, depending on the subject matter. Just like any other property right, its owner(s) can assign its ownership to third parties as they see fit. Contrary to popular belief, there is no requirement to register a work for it to be protected by copyright — it simply arises automatically once the work is completed.

Generally, copyright subsists in works if they are ‘original’ — meaning that the author has exercised a modicum of skill, labour and judgement in its creation. In the UK, this is a fairly low threshold, with copyright able to subsist in something as mundane as a timetable.

It’s important to highlight that only the manner in which an idea is expressed — rather than ideas in and of themselves — can be protected by copyright.

For example, you may have the idea of a story relating to a galactic rebellion fighting off an evil empire, assisted by futuristic light-sabre wielding knights — but until someone has written the story (or, indeed, filmed the movie) — no copyright protection can subsist. In other words, it isn’t the idea that you have in your mind of the ‘STAR WARS®’ stories that is protected — only how these stories are expressed in things like movies, books, video games and soundtracks.

Interfaces between copyright and ‘AI’ generated art

In the UK, the Copyright, Designs and Patent Act of 1988 (CDPA 1988) — in a rather forward-thinking manner — already included provisions related to computer generated art:

“computer-generated”, in relation to a work, means that the work is generated by computer in circumstances such that there is no human author of the work” — Section 178, CDPA 1998.

Therefore, the answer to whether copyright can subsist in computer generated works of art is an easy ‘yes’, since the UK test for originality has a low threshold. So long as the work being produced hasn’t been copied — and results from a modicum of skill, labour and judgement have been applied — copyright can subsist. CDPA 1998 (Section 12(7)) provides for a reduced copyright protection term of 50 years for computer generated literary, dramatic, musical or artistic works — down from 70 years for ‘human’ works.

Of the many difficulties that arise when AI and a human are involved in the creation of a work of art, the question who owns the work itself is of central importance.

‘In the case of a literary, dramatic, musical or artistic work which is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken.’ — Section 9(3) CDPA 1988.

On the face of it, this provision appears to resolve the issue of ownership. However, who can say, with certainty, who made the necessary arrangements?

Could it be:

1. The person(s) or organisation that created the software?
2. The person(s) or organisation that taught the software (if the software is AI-enabled)?
3. The user, through their free and creative inputs into the software?

The infamous Naruto — macaque photographer extraordinaire

The answer to that question is reminiscent of the now infamous case of the macaque photographer (named ‘Naruto’) who stole a camera and proceeded to take a selfie.

The owner of the stolen camera claimed that he, in fact, owned the copyright attached to the image — given that he had made creative contributions in the setup of the camera (and allegedly having befriended the family of monkeys into accepting him as one of their own).

The matter was escalated to a Californian US District Court (Naruto v. David Slater et al., No. 16-15469, 9th Cir. 2018) when PETA claimed that the photographer had made an unauthorised publication of the picture in a book, because Naruto owned the copyright.

The US District Court dismissed the case — and of the judges stated (quite bluntly) that: “if Congress and the president intended to take the extraordinary step of authorizing animals as well as people and legal entities to sue, they could, and should, have said so plainly“.

In France, for example, Article 514-14 of the Civil Code provides that “Animals are living beings imbued with sensibility” — which, quite subtly, is just shy of legally stating that animals have a personality. Originality in French law requires the ‘imprint of the author’s personality”.

It’s important to note here that animals and ‘AI’ enabled software are distinct from one another — within the current state of the art, an ‘AI’ is unable to exhibit curiosity, sensibility or indeed personality.

Assessing copyright disputes in ‘AI’ generated art

As far as AI is concerned, the question of authorship/ownership of AI generated work will hopefully be resolved in the next few years in the UK, which last year published its National AI Strategy.

In the meantime, disputes arising from computer generated works in the UK from software/‘AI’ — such as Stable Diffusion or DALL-E — will need to be assessed on a case-by-case basis and will likely depend on the level of contribution of each of the parties in the creation of the work. Stable Diffusion’s software allows for quite a range of user inputs and its creators have explained its functioning in detail.

While joint authorship is possible under CDPA 1988, it seems unlikely that a court would rule in favour of ascribing ownership to a software which — it bears highlighting — is not a legal entity yet (even though it appears to have ‘human like’ characteristics).

The patent world is currently discussing whether an AI can be listed as an inventor on a patent application. While not directly comparable (because the standards for “authorship” and “inventorship” are different), we have previously outlined how different patent offices around the world are dealing with AI inventorship.

We are still some way from sentient Artificial General Intelligence — but as laws evolve with society, it isn’t beyond the realm of possibility to imagine AGI making its way to some form of legally recognised ‘entity’ status, in the same way that animals have done in certain countries over time.

Our advice for using AI-generated artwork

The risks associated with using these tools to generate art are largely unknown at this stage, as they have not been substantially tested in any courts. However — depending on how risk averse you are — we would suggest the following:

1. Consider not using any artist’s name or work in the generation of artwork — especially where their work is still under copyright protection. Without knowing the specifics of the tools and prompts used, it’s difficult to give a definitive answer on whether using the style of an artist would be considered an infringement of their rights — but it may nevertheless annoy said artist and result in unwanted bad press and/or infringement proceedings.
2. While it might be easy to use these tools, their usage does come with terms of service, licence agreements and the like. Such terms can impose limitations on what can be done with the generated art, who owns the final output, the content you’re allowed to generate, and more. Being aware of what you can and cannot do when using a tool is very important, as your ability to use it may be rescinded or considered infringing of the tool author’s rights if you violate their terms.

If you’re innovating in AI or machine learning, or have questions regarding copyright and ownership, get in touch with us for a free initial chat about your situation.