In January 2023 an unprecedented lawsuit was filed in a US Federal Court by three US-based visual artists. It is a ‘class-action’ comprising separate claims by three artists, who bound themselves into a single lawsuit based on the same legal issue (or classification). The lawsuit is against three companies, each of which allegedly used the claimants’ artworks to train an AI (Artificial Intelligence) visual art tool to power ‘text-based image creation’ – thereby violating each artist’s copyright.
Commercial use of AI in the contemporary art ecosystem could become a hot legal topic in 2023. The recent rapid growth and popular use of AI visual art tools has already prompted several significant legal controversies, now in train, which are unlikely to be satisfactorily resolved in 2023, and more may well emerge as this year unfolds.
The US class-action lawsuit’s first defendant is Stability AI, a London-based company that developed and recently launched its Stable Diffusion AI model: a digital tool that enables users to generate ‘professional-quality images with a simple text prompt’, which can also be applied to other tasks such as inpainting, outpainting, and generating image-to-image translations guided by a text prompt. The second defendant is Midjourney, a San Francisco-based research company that uses Stable Diffusion to power text-based image creation. The third defendant is the Los Angeles-based online community for artists, DeviantArt, which offers its own Stable Diffusion-powered generator called DreamUp.
Kelly McKernan is one of the claimants, a fine art practitioner who also creates watercolour and acrylic gouache illustrations for books, comics and games. McKernan entered their full name as a text prompt into the website Have I Been Trained, which revealed that some of their artwork had been used to train Stable Diffusion. Following initial ‘excitement’ at this discovery, it soon dawned on McKernan that any other users of Stable Diffusion could generate artwork in McKernan’s style by typing in that artist’s full name. Stable Diffusion had not sought McKernan’s prior permission to allow such uses of their original artwork, hence the copyright violation lawsuit.
Karla Ortiz is another claimant, a fine art practitioner who is also a leading film and entertainment industry concept illustrator. Ortiz discovered a free and open-source software tool, one of the initial versions of text-to-image AI generators: DiscoDiffusion. Ortiz realised this tool could generate images in her style and in styles of other artists she knew – including McKernan, with whom she joined in the class-action lawsuit. ‘It’s gross to me that AI-powered apps and services instantly spit out art based on a text prompt,’ Ortez says. ‘They trained these models with our work. They took away our right to decide whether we wanted to be a part of this or not.’
Sarah Anderson, who is a cartoonist and illustrator widely known for her webcomic Sarah’s Scribbles, is the third claimant. ‘Artists have a right to say what happens to their hard-earned works,’ Anderson argues. ‘It’s clear from the way AI generators rolled out that there was never any consideration given to artists, our wishes, or our rights, and this is our only option to be heard.’
Within a week of the US class-action’s filing, stock-image giant Getty Images filed a separate lawsuit, also against Stability AI, in London’s High Court. The claim also alleges that Stability AI ‘unlawfully copied and processed millions of [Getty’s] images protected by copyright and the associated metadata’ to train its AI model. Publicly responding to the filing of Getty’s lawsuit, Stability AI’s spokesperson said: ‘Please note that we take these matters seriously. Anyone that believes that this isn’t fair use does not understand the technology and misunderstands the law.’ A ‘fair use’ copyright violation defence by Stability AI may well feature in the US class-action artists’ lawsuit, but is unlikely to be available to defend Getty’s separate UK lawsuit.
Although UK and US copyright laws share the common aim of permitting certain uses of copyright-protected work without the copyright owner’s prior consent or licence, each legislature has enacted different ways of achieving this same objective. There are significant distinctions and differences between UK and US copyright law doctrines of permissibility, which could potentially produce different judicial results in the separate US and UK trials, albeit that both focus on use of the Stability Diffusion AI tool.
US copyright legislation enacts four broad criteria that US courts apply in cases of alleged copyright violation to decide whether a use is fair: the purpose and character of the use; the nature of the copyrighted work; the amount and substantiality of what has been taken from the copyrighted work; and the effect of the use upon the potential market for the copyrighted work. Application of these criteria affords US copyright courts significant flexibility to arrive at fair and just evaluation of the circumstances of each case. However, a fair use defence in US copyright law is like, but not the same as, UK’s permitted use defences.
UK copyright legislation adopts a more restrictive approach than the US by enacting specific limited circumstances in which any unauthorised use of a copyright-protected work is allowed. UK defendants are required to satisfy a copyright court that their use fits squarely within at least one of the many specified ‘permitted acts’. And some acts are only permitted if they are also ‘fair dealing’ for specific purposes, including: private study; criticism, review, quotation, and current news reporting; caricature, parody, or pastiche; educational instruction and examination and – potentially the most relevant for AI – for the purpose of non-commercial research. In other words, defendants are first required to satisfy a copyright court that their use or dealing fits squarely within at least one of the specific purposes; and, if so, that such purpose was fair dealing. Whether a purpose is fair requires a court to judge each case on its own facts and invariably involves, for example, assessing whether the dealing damages the copyright-protected work’s actual or potential economic market.
Accordingly, in Getty’s London lawsuit, Stability AI may possibly encounter difficulty in defending its Stable Diffusion AI tool on the ground of fair dealing for the purpose of non-commercial research. Stability AI appears to be a commercial business, and in any event it is questionable whether sharing and/or marketing its AI research would be judged as being fair. However, there is a further permitted act specified by UK copyright law that does not require a defendant to prove its activity to be fair dealing, namely text and data mining (TDM): an automatic process of analysing or scanning large amounts of text or data using custom-made scripts looking for patterns, discovering relationships or trends that are not usually visible through normal reading.
Introduced into UK copyright law in 2014, this new permitted act reflects the then rapidly increasing development of TDM and its use as an AI tool. But, as with the research fair dealing defence, TDM is permitted only ‘for the sole purpose of non-commercial research’, which is likely again to pose difficulties for Stability AI in Getty’s UK lawsuit. As for the US class-action lawsuit, US copyright legislation currently has no such specific TDM defence available, which means that Stability AI and its two co-defendants are very likely to rely on ‘fair use’ to defend themselves.
Beyond the US, some countries have recently considered amending their own national copyright laws to permit TDM research for commercial purposes – without a copyright owner’s prior consent. Allowing TDM for commercial uses without violating copyright is an understandably controversial issue. Some countries regard commercial TDM as a means of investing in national innovation and technology transfer. The EU, for example, is currently considering changing EU copyright law so that copyright owners could ‘opt out’ of commercial (but not scientific or cultural) TDM uses – against which change EU copyright owners argue that an ‘opt in’ to commercial use would be more just and fair.
Towards the end of 2022, the UK government proposed changing copyright law to permit TDM of digital formats of all creative works, including visual artworks, for commercial purposes without prior consent of authors/copyright owners of such works. In February 2023, this proposed change was considered by the House of Commons in a debate initiated by Sarah Olney MP, the Liberal Democrat Spokesperson for Business and Industrial Strategy, who introduced the debate with the following words: ‘The rapid rise of artificial intelligence seemingly knows no bounds. Each week, a new AI tool is launched that drives further change across business, science, the arts, and everyday life … It goes further and faster than humanly possible … it has been identified as one of the UK’s key growth industries, and it is vital that Government policy supports digital innovation to position the UK as a world leader in this field.’
Yet Olney also cited key concerns voiced by hundreds of UK-based creative workers in a recent survey she had commissioned to learn how AI was impacting their work: many said their work, for which they own copyright, had been used by AI commercial platforms without their consent or monetary recompense; and that bypassing their copyright undermined their skills and devalued the creative process, as well as having a detrimental impact on their income.
All debating MPs echoed Olney’s concerns and commented that, for example: a significant number of active legal battles regarding AI and intellectual property were currently going through courts; intellectual property rights were fundamental to the success of the UK’s world-leading creative industries; copyright laws not only protected the integrity of original work, but also provided a revenue stream to ensure that creators made a living from their work. At the heart of the debate were MPs’ concerns that the UK government was proposing an ‘all-out exception to copyright for text and data mining in order to promote AI, with no opt-out for rights holders’; and that AI platforms worldwide would be able to copy and use UK creative authors’ works to produce new material, which could then be merchandised without having to gain prior permission from the creator or to pay for a licence. This, it was pointed out, would see ‘a huge transfer of value from individual creators to AI companies’.
Such concerns were cited from key representatives of the music and publishing industries, actors and other performers. The Design and Artists Copyright Society (DACS) had also expressed concern that such a far-reaching exception to artists’ copyright law was ill-advised, and that there was no substantial evidence of the need for such a strong measure. Furthermore, MPs referred to a report by the House of Lords Communications and Digital Committee published in January 2023, ‘At Risk: Our Creative Future’, which had concluded that the government’s proposed TDM changes were ‘misguided’, and advised that they be dropped ‘immediately’.
Responding to such consensual concerns and urgings, the minister for science, research and innovation, George Freeman MP, said that the government had listened to the creative industries’ concerns, and would not now be proceeding with its TDM proposals, and would instead be seeking to reopen consultations with DACS and other organisations in the copyright community. Accordingly, Freeman had written to make it clear to other ministers that the government’s TDM proposals were ‘not correct’, had been met with a huge negative response from the creative industries, and that the government was looking to stop them. Freeman went further, reassuring the House ‘that we will not be proceeding with the proposals’.
There is evidently increasing controversy and ambiguity worldwide about the current legality of commercial data mining of copyright-protected creative works. And, as ever, legislators and courts will need time to catch up with rapidly developing AI innovations in order to provide fair and balanced legal certainty – applicable both nationally and internationally. Meanwhile, the outcomes of the two current lawsuits, by US artists and Getty Images, could prove to be landmark steps towards achieving clarity.
© Henry Lydiate 2023