Out-Law / Your Daily Need-To-Know

Out-Law Analysis 5 min. read

Midjourney decision intensifies questions over definition of ‘human authorship’


A recent decision by the US Copyright Office to cancel part of a copyright registration for artwork that was partially generated by an AI programme could have important implications for copyright protection.

On 21 February the US Copyright Office cancelled artist Kristina Kashtanova’s copyright registration, granted last September, for her comic book Zarya of the Dawn due to “inaccurate and incomplete information”. The comic book contained art generated by Midjourney, an AI programme that generates images from text prompts, but Kashtanova did not disclose this fact.

Kashtanova argued that she had authored every aspect of the comic book and that the AI had merely “assisted” her, but the Office concluded that Midjourney ultimately used a different process to generate an image, which was “not the same as that of a human artist, writer or photographer”. Kashtanova may have “guided” the structure and content of each image by making text prompts but it was ultimately Midjourney which originated the “traditional elements of authorship” in the images. This meant the images generated by Midjourney were not “the produce of human authorship”. The Office replaced her original registration with a new, limited registration covering the text of Kashtanova’s comic book, the selection and arrangement of the images and the text – but not the images themselves.

Kashtanova’s lawyer expressed surprise at the ruling, stating that the Office erred in focusing on the “output of the tool, rather than the input of the human”. That comment strikes at the heart of the issue with which this case is concerned: where does input of the human end, and output of the tool begin? The Office’s ruling has certainly clarified the US’s position on whether AI-generated work attracts copyright protection, but what is less clear is what actually counts as an ‘AI-generated work’, and at which point copyright protection ceases to apply.

‘Computer-generated works’ in the UK

The UK has long recognised ‘computer-generated works’ as attracting some form of copyright protection. Unlike other jurisdictions, the UK protects computer-generated works which do not have a human author under the 1998 Copyright, Designs and Patents Act (CDPA). Instead, the author of such works is the person “by whom the arrangements necessary for the creation of the work are undertaken”. Protection lasts 50 years from the date the work is made, which is shorter than the protection that is afforded to literary, dramatic, musical, and artistic works; sound and music recordings; and films with a human author. Protection for these works lasts for 70 years after the death of their creator.

When first proposed in 1987, Lord Young of Graffham said that the CDPA was “the first copyright legislation anywhere in the world which attempts to deal specifically with the advent of artificial intelligence”. However, generative AI is challenging the traditional definition and concept of computer-generated work, particularly in relation to delineating when human authorship ends and AI authorship begins. Computer-generated work is defined in the CDPA as where “the work is generated by a computer in circumstances such there is no human author of the work”. However, the definition is unclear on what “no human author” actually means, beyond the vague “making arrangements” definition mentioned above.

In the case of generative AI, part of its popularity has been users wanting to input weird and wonderful prompts to see what images the AI produces. It is unclear whether this would constitute human authorship such that the work is taken beyond the scope of the CDPA definition, potentially muddying the waters when deciding which parts of the work are eligible for regular copyright protection as a work created by a human.

Some may argue that because the human has no real control over the eventual output and cannot predict what it will look like, their intervention is minimal, and the work is almost completely computer-generated. Others might assert that the AI system is merely a tool for the human to create an original work, in the same way that a paintbrush is a tool for an artist. This is what Kashtanova originally tried to argue: that Midjourney was a tool she used to create her comic book images, but the Office was not convinced.

Some commentators believe that it is impossible for a copyrighted work to be generated without any human authorship. For example, humans will be involved in training AI systems, and those AI systems are trained on materials which were originally produced by human creators. It is still thought that we are a long way off an AI system making the creative decisions or “thinking” for itself. Therefore, clarity on what constitutes adequate human authorship or intervention where AI is partially involved is still sorely needed.

What constitutes human authorship?

The UK government has recognised that change is necessary, and recently published results from a consultation calling for views on copyright and related rights in relation to AI. The government’s summary states that: “[t]o the extent that a work is made with assistance from AI but involving human creativity, it will be protected like any other work. Copyright protection will protect the work to the extent that it is the human creator’s ‘own intellectual creation’, and the first owner of the work will be that creator. The AI in these cases may be considered to simply act as a tool which allows an artist to express their creativity”.

The Office’s ruling on Kashtanova appears to reflect this rationale – that only the human-created parts of a work are protected by copyright, which makes sense given that AI is not recognised as a legal person who can own rights. However, the real issue is identifying which parts of the work were created by the human, and which parts were created by the AI, so that copyright protection can be applied to the correct parts. To its credit, the Office did try to differentiate between human input and AI input, by recognising that where Kashtanova had substantially edited an image generated by Midjourney, such changes would constitute human authorship and attract copyright protection.

It is possible to infer from this that direct human control and intervention over an AI output will constitute human authorship for the purposes of works with both human and AI input, likely because the human has direct control over, or has “designed”, the output in that scenario. Contrast this with a situation where the AI programme’s output is almost exclusively generated with limited prompts or decisions from the human – for example, just a couple of words – and the human does not edit or alter the resulting output in any way. Is the human authorship too limited in this instance for the work to attract copyright protection? The Office appeared to think so in its Midjourney ruling.

Issues of originality

It remains unclear whether the Office’s objection to Kashtanova’s copyright registration arose from the fact that a human author was necessary to attract copyright protection, or whether it deemed the work to lack originality due to the AI programme’s input. Both issues are closely related as AI is not currently recognised as being able to create original works when it comes to being afforded copyright protection.

Some commentators believe that the current legal requirement in the UK for a computer-generated work to be “original” in order to attract copyright protection under the CDPA is paradoxical, as originality has always been conceptualised in relation to human characteristics, such as skill, labour, and judgement.If the law refuses to recognise AI as a legal person capable of enjoying intellectual property rights, then it does seem illogical to consider a computer-generated work as original for the purposes of copyright law, without acknowledging some form of human intervention in that work.

IP rights were invented to protect and encourage creativity and innovation because this benefitst society. Lawmakers will have to grapple with whether awarding this protection to AI-generated work is in line with this aims, or whether it actually discourages human endeavour because work is more easily generated by computer.

Co-written by Concetta Scrimshaw of Pinsent Masons.

We are processing your request. \n Thank you for your patience. An error occurred. This could be due to inactivity on the page - please try again.