With AI being used for everything from fanciful selfies to commercial book covers, the question of who owns AI generated art is becoming increasingly important. As with any emerging technology, lawyers and judges have been left to analyze the application of AI in the creation of artistic works using existing legal doctrines, creating uncertainty for both existing copyright holders and those seeking to assert copyright. This primer on Canadian copyright and AI generated art explores the legal uncertainty regarding rights, provides two examples of circumstances under which the use of AI could be considered an exception to copyright infringement rules, and points to lessons learned from litigation in the U.S.
Generative AI refers to machine learning algorithms that create new content, such as text, images, video, and audio, based on existing data. Text-to-image generative AI software, such as OpenAI’s DALL-E-2, creates images based on text prompts provided by the user. DALL-E-2 utilizes vast amounts of data scraped from the internet to train the algorithm, which may include copyrighted material.
Text-to-image generative AI is being used commercially in ever expanding ways. For example, businesses are using AI generated works to create illustrations for children’s books, art for newsletters, concept art for games and storyboards for movies. At issue is whether this AI generated art violates copyright held by others and is protected by copyright in the same way as other art.
Ownership of works created by generative AI
Canada’s Artificial Intelligence and Data Act (AIDA) is proposed legislation with significant business impacts that would regulate some aspects of AI that relate to international and interprovincial trade and commerce. While generative AI seems to meet AIDA’s definition of artificial intelligence, the proposed legislation does not address copyright concerning the commercial use of artistic works created using generative AI.
The Copyright Act provides copyright owners the exclusive, legal right to produce, reproduce, sell or license, publish or perform an original work or a substantial part of it. The key word here is “original.” An important Supreme Court of Canada case defines an “original work” as a product of the creator’s “skill and judgment” that “must not be so trivial that it could be characterized as a purely mechanical exercise.”
Some members of the Canadian legal community have argued that providing text prompts to a generative AI program is a “purely mechanical” process that does not result in an original work as defined by the Supreme Court. However, anyone who has spent time plugging text inputs into DALL-E 2 will quickly discover that there is a learned skill involved in the formulation of text prompts to create beautiful (or beautifully bizarre) AI generated art. The wrong combination of text prompts will lead to the program creating useless and often nightmarish images.
Generative AI programs may train their datasets using copyrighted material, so those who use generative AI for commercial purposes may unwittingly become entangled in copyright infringement claims. Under Canadian law, copyright infringement occurs when a person reproduces all or a substantial part of an original work without the copyright owner’s authorization or the availability of a statutory exception.
It is uncertain whether the Copyright Act provides protection to copyright owners regarding the use of copyrighted material in analytical processes such as machine learning. If a generative AI program uses copyrighted material when data mining, it could be argued that statutory exceptions to infringement may apply, such as temporary reproductions for technological processes and fair dealing. We explore both of these exceptions below.
Potential exception 1: Temporary reproductions for technological processes
The temporary reproductions exception aims to facilitate technological processes. The exception provides that temporary reproductions do not constitute copyright infringement if:
- The reproduction forms an essential part of a technological process.
- The reproduction’s only purpose is to facilitate a use that is not otherwise an infringement of copyright.
- The reproduction exists only for the duration of the technological process.
The second requirement above may prove to be problematic for generative AI users, as the data mining of copyrighted material may enable and result in an infringement in the AI generated work. For example, if a user prompts a generative AI program to create a work derived from existing copyrighted material, such as “Mickey Mouse, Homer Simpson and SpongeBob SquarePants in a presidential primary debate,” it could be argued that the exception does not apply, and the resulting work constitutes copyright infringement.
Potential exception 2: Fair dealing
The fair dealing exception is a general right of all potential users of a copyrighted work. Fair dealing permits the use of copyrighted material for specific purposes — research, private study, education, satire, parody, criticism, review or news reporting — so long as the use is “fair.” Six factors are used to determine if a use is fair: the purpose, character and amount of dealing; whether alternatives to the dealing exist; the nature of the work; and the effect of the dealing on the copied work.
The research use under the fair dealing exception may be relevant to works created using generative AI. Canadian courts have given the term wide interpretation, with the Supreme Court of Canada holding that “research is not limited to non-commercial or private contexts” and should be otherwise afforded liberal interpretation. One Supreme Court case, for example, found that listening to 30- to 90-second music previews to determine a user’s musical preferences constituted research for the purposes of the fair use exception. One could argue that the data scraping and machine learning processes of a generative AI program constitute research, but it remains to be seen if that argument will be advanced or accepted in Canadian law.
The Canadian fair dealing exception is notably narrower than similar exceptions in other jurisdictions. The fair use exception in the United States, for example, permits the limited use of copyrighted material so long as the user adds something different to a copyrighted work, such that the work is transformed.
Lessons from litigation in the U.S.
U.S. courts have considered various copyright issues involving relatively new digital technologies, including whether digitizing books and displaying excerpts constitutes fair use and whether using source code owned by another party qualifies as fair use. Nonetheless, these cases do not neatly address the distinct issues raised by the data mining and machine learning processes involved with generative AI programs, which could result in new works that mimic copyrighted material.
Jurisprudence emerging from the U.S. may soon provide insight, however.
In October 2022, the Supreme Court of the United States heard submissions on whether Andy Warhol’s illustration of Prince, based on photography by Lynn Goldsmith, constituted fair use. At the heart of the dispute is whether Warhol changed too little from the photo, such that his work was not “transformative” under the fair use doctrine. The Court’s decision remains pending as of the date of publication of this article.
In February 2023, stock image company Getty Images filed a copyright infringement lawsuit in the U.S. District Court in Delaware against Stability AI. This complaint followed Getty’s announcement in January 2023 that it “commenced legal proceedings” against the company in the High Court of Justice in London. In its January 2023 press release, Getty Images argued that Stable Diffusion (the generative AI program) “unlawfully copied and processed millions of images protected by copyright” for AI training purposes. Stability AI is also being sued, along with fellow generative AI company Midjourney, by three artists seeking a class action.
Owners of generative AI programs, their users and their legal counsel should closely monitor these cases, as they may provide some general guidance regarding the copyright issues raised in this article. However, the law in Canada and elsewhere will remain uncertain for the time being. This uncertainty, coupled with the ever-growing popularity of generative AI, will lead to an increasing number of lawsuits in Canada and other jurisdictions.
Copyright over AI generated works remains an open question in Canada.
Businesses that use generative AI-created works for commercial purposes should be aware that such works may risk violating copyright held by others, and that their AI generated works may not receive copyright protection in the same way as art created using legacy technology. They should pay close attention to the terms and conditions that apply to the generative AI software they use, since some software providers may seek to shift liability for copyright infringement to users of the program.
Prior to commercializing works created by generative AI, users should seek legal counsel to carefully review the applicable terms and conditions and determine whether their work risks infringing existing copyright or can receive copyright protection itself.
If you have questions about who owns the AI generated art used in your business, reach out to any of the key contacts below.