Global Legal Systems Struggle to Define AI-Generated Creative Work Ownership

Inventors, creators, and intellectual property lawyers in Canada and internationally are confronting a pressing question: how can existing statutory regimes, originally designed without artificial intelligence in mind, accommodate the surge of AI-generated creative works?

Amy Grenon, a partner at Aird & Berlis LLP, notes that this debate is unfolding globally. "If you are using AI, how do you protect the inventions that the AI will come up with?" she asks.

One of the most prominent figures raising this issue is Stephen Thaler, who has sought to register patents naming his AI system, Device for the Autonomous Bootstrapping of Unified Sentience (DABUS), as the inventor. Thaler’s efforts have been rejected by multiple jurisdictions, including the US Federal Court, UK Appeals Court, European Patent Office, Federal Court of Australia, and the German Federal Patent Court.

In the United States, Thaler named DABUS as the sole inventor on his patent application. The US Patent and Trademark Office (USPTO) rejected the application, citing the Patent Act’s definition of inventors as natural persons. Charles Morgan, partner at McCarthy Tétrault and former leader of its technology law group, explains that in May 2020, the US Federal Circuit Court upheld this interpretation in Thaler v. Vidal, ruling that AI tools cannot be recognized as patent inventors.

Similarly, the Canadian Intellectual Property Office (CIPO) issued Thaler a non-compliance notice for his DABUS patent application. CIPO stated that machines cannot hold legal rights in Canada nor transfer them to humans. However, CIPO suggested that Thaler could reapply by including a statement identifying himself as the "legal representative of the machine," according to Morgan.

Grenon observes that nearly all courts in Thaler’s cases have required a human inventor, though the exact reasoning varies by jurisdiction. She notes that Canada has yet to see litigation directly addressing this issue.

"The problem with these statutes is that they were developed with humans in mind, not AI," Grenon says. "So the question becomes: how do you try to get protection for AI-generated inventions?"

Beyond patents, AI raises complex questions in copyright and trademark law. These issues concern both the protection of AI-generated output and whether AI’s use of existing data infringes on other IP rights.

In 2021, Heritage Canada initiated a consultation to consider updating Canada’s copyright framework in light of AI and the Internet of Things. Currently, copyright protection generally requires a human author.

Despite this, some applicants have successfully registered AI-produced works by creatively attributing authorship. For example, Ankit Sahni from New Delhi registered an AI-generated painting titled SURYAST with CIPO, listing himself as author and the RAGHAV Artificial Intelligence Painting App as co-author. Morgan notes this is the first Canadian copyright registration naming an AI as co-author, though it is unclear if CIPO would have approved the registration without Sahni’s human authorship claim.

Similarly, Hugh Stephens of British Columbia registered a poem and image created using AI tools ChatGPT and DALL·E 2. Stephens identified himself as author but disclosed that the work was "created entirely by AI programs" with "minimal skill and judgement" from the human claimant.

In the US, the Copyright Office has taken a stricter stance. It recently denied two AI-generated work registrations: "A Recent Entrant to Paradise," produced by Thaler’s Creativity Machine, and the graphic novel "Zarya of the Dawn," which included images generated by the AI system Midjourney. The Copyright Review Board affirmed these denials, citing lack of human authorship.

In March, the US Copyright Office issued guidance confirming that human authorship is required for copyright protection. It stated that it will assess whether a work is "basically one of human authorship," with AI serving as an "assisting instrument," or if a machine created the "traditional elements of authorship." The guidance advises applicants not to list AI systems as authors or co-authors.

Generative AI systems like ChatGPT and Midjourney rely on crawling and copying massive datasets from the internet to train their models. This practice raises IP questions about whether the entities training these AI models have the rights to use such data.

Several class action lawsuits have been filed against AI toolmakers alleging copyright infringement through unauthorized use of copyrighted materials. One notable case, filed in November 2022 in US District Court in California, targets Microsoft, GitHub, and OpenAI. The plaintiffs, who remain anonymous, claim that Microsoft and its subsidiaries illegally reproduced copyrighted code hosted on GitHub and violated open-source licenses in training the AI coding tool Copilot.

Another case, Andersen v. Stability AI, MidJourney & DeviantArt, accuses the defendants of training their AI image-generation tools on billions of copyrighted artworks without artists’ consent, thereby infringing the rights of millions of creators.