Computer-generated art reached a tipping point in 2022. Experiments with AI-generated images go back decades (such as the computer programs of Harold Cohen starting in the 1970s). In 2021, OpenAI’s Dall·E was able to create images in response to text prompts, including those combining unrelated concepts (such as “a snail made of harp”), but the results were merely competent in artistic terms. In 2022, a variety of “generative AIs” were launched, including Dall·E 2, Midjourney, Stability AI and tools from Google (Imagen) and Meta (Make-A-Scene). Some or all of these are now capable of producing images of remarkable quality (see, for example the Midjourney community showcase).
AI-generated works have won awards: The Crow, an “AI-made” film won the Jury Award at the Cannes Short Film Festival and the story of an AI artwork winning the Colorado State Fair’s annual art competition was reported in The New York Times. AI-generated art was used for magazine covers, including Cosmopolitan and The Economist. Some stock imagery libraries, including Adobe and Dreamstime, started to accept AI-generated images. Computer-generated art has been used for one-off images for NFTs and has been predicted to be essential to create varied, detailed, immersive 3D spaces and characters needed for the metaverse.
Generative AI also showed its power in creating clear, sustained text in a variety of formats and styles with the launch of ChatGPT in November, generating daily interest in the news, debates over bans in schools, universities, scientific conferences and websites and speculation that the technology could be baked into Microsoft Office. 2022 was also the year for advances in AI-generated computer code, 3D models, image descriptions, video, audio and algorithms. (And, of course, the debate on computer-generated inventions rumbled on across the world.)
There is uncertainty over how much selection and refinement is needed to produce the best work (see examples of selection, refinement and reworking here and here) and getting the right prompt is challenging enough to have created a market for effective prompts. There will be debates over whether generative AI is, for the purposes of copyright, a tool used by a human author or is an autonomous creator of works (see previously on this blog e.g. here, here, here and here).
Very few jurisdictions expressly provide for copyright in computer-generated works. These include (and may be limited to) Hong Kong (section 11(3), Copyright Ordinance), India (section 2(d)(vi), Copyright Act, 1957), Ireland (Art. 21, Copyright and Related Rights Act 2000), New Zealand (section 5(2)(a), Copyright Act 1994), South Africa (section 2(h), Copyright Act 1978) and the UK (Copyright, Designs and Patents Act 1988, section 9(3)).
In 2022 the UK Government responded to a consultation over whether to alter the UK provisions for computer-generated works (CGWs) and determined to leave them alone, concluding “There is no evidence at present that protection for CGWs is harmful, and the use of AI is still in its early stages. As such, a proper evaluation of the options is not possible, and any changes could have unintended consequences. We will keep the law under review and could amend, replace or remove protection in future if the evidence supports it.” (For a comment, see here)
The DABUS team (the Artificial Inventor Project) started testing copyright law back in 2018, seeking to register with the US Copyright Office (USCO) A Recent Entrance to Paradise, an image created by DABUS. Registration was refused in August 2019, in line with previous US case law and guidance. In February 2022 the Review Board of the United States Copyright Office again refused registrationon the basis that human authorship is required. (The team has appealed the decision, so watch this space.) In September 2022, the USCO registered Zarya of the Dawn, a graphic novel generated using MidJourney. However, the USCO subsequently gave notice of potential cancellation and requested details of the human involvement in the creation of the work (see here) (as of 25 January, it appears the registration has been cancelled but only, at this stage, because of system error). (For comparison, the Canadian Copyright Register has included since 2021 a registration naming an AI, RAGHAV, as a joint author with a human.)
Litigation against generative AI started in 2022 with a class action launched against Copilot, an AI trained to generate computer code using examples from Github. It ramped up further in the first weeks of 2023 with a UK action by Getty Images against Stability AI and a class action against Stability AI and others in the US. (The first case alleging infringement by generative AI may have been the 2018 Canadian litigation regarding Adam Basanta’s mixed-media installation, All We’d Ever Need is One Another, which generated works “via a process involving scanners which capture light hitting glass, computer-generated mouse movements, and analysis by a series of deep-learning algorithms which determine whether or not the final product is ‘art’“).
Leaving the law aside, generative AI triggered some backlash in 2022. While some stock image libraries allowed AI-generated art, others moved to ban it (e.g. Getty Images, Newgrounds, and PurplePort). The Los Angeles Times reported in August that an AI rapper, FN Meka, with over 10 million followers on TikTok was signed by Capital Music Group and almost immediately dropped following complaints of cultural insensitivity. A journalist used Midjourney to illustrate an article in The Atlantic and was attacked on social mediafor not hiring an illustrator. And, in December, Tor, a “Publisher of Science Fiction, Fantasy, Horror, Mystery, Thriller and Suspense, and Other Speculative Fiction“, apologised for using an AI-generated image for a cover design saying “we licensed an image from a reputable stock house. We were not aware that the image may have been created by AI”.
2. Text and data mining
There is wide disparity in the scope of exceptions in national copyright laws permitting copying for the use of training AI. Notable examples include US case law on fair use, the UK’s non-commercial text and data analysis exception, the EU’s text and data mining exception (covering commercial activities subject to opt out) and Japan’s exception (article 30-4, which requires that the work “is exploited in a way that does not involve what is expressed in the work being perceived by the human senses“. A global review of text and data mining exceptions was published in Science in December.
In June, the UK Government announced a “plan to introduce a new copyright and database exception which allows TDM for any purpose” (commented on this blog), but by the end of the year it was clear the Government wished to rethink this (see here, here and here), perhaps because of the position set out in August by the Publishers Association. The UKIPO has since been speaking with specific stakeholders.
3. Parody, pastiche and caricature
The EU’s harmonised copyright exceptions for parody, pastiche and caricature (introduced into UK law in 2014) was tested in the UK, France and Italy. The focus in these cases was on parody. Those interested in computer-generated AI will be on the lookout for cases on the limits of the “pastiche” exception, since many users seek to generate works in the style of specific artists: in September, MIT Technology Review reported that Greg Rutkowski, a fantasy illustrator, had discovered his name had been used 93,000 times in prompts to generate art. (The issue of works in the style of artists may arise in the US class action against Stability AI and others, but under the more general US exception of fair use.)
4. Deep fakes and voice cloning
In November, we learned of a new UK comedy, Deep Fake Neighbour Wars, promising “the very latest in AI technology to turn the UK’s best new impressionists into the world’s most famous celebrities – only here they are ordinary people who happen to be embroiled in petty silly neighbour disputes“. Just into the new year (on 5 January), a new speech synthesiser, VALL-E, was announced with the promise to “to synthesize high-quality personalized speech with only a 3-second enrolled recording of an unseen speaker as an acoustic prompt“. These developments will encourage lawyers to review copyright, passing off, unfair competition, human rights and constitutional laws for ways to enforce likeness and voice.
5. Copyright infringement and enforcement using AI
Finally, for those ready for a 172-page report, the European Intellectual Property Office published its Study on the impact of artificial intelligence on the infringement and enforcement of copyright and designs. The report looks at risks from AI (including infringements by generative AI) and AI tools for enforcement (such as automatic detection of attempts to re-register previously registered designs).
To make sure you do not miss out on regular updates from the Kluwer Copyright Blog, please subscribe here.
Kluwer IP Law
The 2022 Future Ready Lawyer survey showed that 79% of lawyers think that the importance of legal technology will increase for next year. With Kluwer IP Law you can navigate the increasingly global practice of IP law with specialized, local and cross-border information and tools from every preferred location. Are you, as an IP professional, ready for the future?
Learn how Kluwer IP Law can support you.