Do AI-generated images deserve copyright protection? A Canadian court challenge is putting that claim to the test

Open this photo in gallery:

Ankit and Kritika Sahni spent a lot of time on their terrace during the pandemic under lockdown in New Delhi, where they live. One evening in 2020, Mr. Sahni took out his iPhone and snapped a photo of the setting sun before it dipped below the roof of an adjacent building.

The photo is nothing special. But what he did next put him in the middle of a debate about copyright, authorship and artificial intelligence in three countries – which, in a way, is what he was hoping for.

After taking the photo, Mr. Sahni loaded it into an AI image generator, which rendered a new version of the picture in the style of Vincent van Gogh’s The Starry Night, albeit on the blurry side. He titled the AI image Suryast, the Hindi word for sunset, and applied for copyright in India, the U.S. and Canada.

Today, the Canadian copyright database shows a registration for Suryast as of Dec. 1, 2021, with two authors: Ankit Sahni and the RAGHAV Artificial Intelligence Painting App, the program used to create the image. It’s the first and likely only time in Canada that AI has been recognized as an author.

And that, according to some experts, is a problem. In July, the Samuelson-Glushko Canadian Internet Policy and Public Interest Clinic (CIPPIC) at the University of Ottawa filed an application with the Federal Court seeking to expunge the registration for Suryast. Failing that, the application asks for RAGHAV to be removed as an author. A non-human, according to CIPPIC, simply cannot be an author under law.

That is not how Mr. Sahni sees it, and he’s tapped a law firm in Toronto to mount his defence. “It’s definitely very important for us to see what the Federal Court has to find,” he said. “What happens in a significant jurisdiction heavily influences what happens in another.”

Lost gigs and lower pay: How AI is already affecting freelance artists and writers

The case could carry implications for the copyright status of AI-generated works in Canada. “We don’t have a lot of case law on computer-generated or -assisted work,” said Maya Medeiros, a partner at Norton Rose Fulbright who focuses on intellectual property. The federal government wrapped up a consultation about generative AI earlier this year with a view to updating the Copyright Act, but has taken no legislative action. Whatever the Federal Court decides with Suryast will undoubtedly influence future cases.

There could be serious consequences – such as enabling copyright trolls – if AI-generated media is given broad protection, according to some experts. Anyone could theoretically make scads of AI-generated images and launch copyright claims against others in court, resulting in a liability risk for creators and amounting to an obstacle for human authorship.

The rise of generative AI has brought with it many thorny questions. Some concern the input of these models: Should it be legal to scrape copyrighted works without permission or payment? Others concern the output: Does an AI-generated image deserve copyright protection? How much human involvement is necessary, anyway?

It’s precisely because of these unknowns that the Sahnis launched an experiment with Suryast. “We must nudge legal systems and governments to prepare answers to such questions,” Mr. Sahni said.

Open this photo in gallery:

The married couple are both IP lawyers at the same firm in New Delhi. Mr. Sahni, 35, has an irrepressible ebullience on the topic. On our first call, he talked for nearly 20 minutes uninterrupted. Growing up, he loved the Star Wars franchise and The Hitchhiker’s Guide to the Galaxy, but also had a more hands-on interest in technology. His LinkedIn profile boasts that at the age of 14, he became one of the youngest people in India bestowed with the title of Microsoft Certified Professional.

A few years ago, the topic of how copyright and IP law apply to AI was burbling up in their professional circles. Seeing as answers were unclear, the Sahnis went about getting them. “So many artistic works are being generated through AI where nobody knows who is to be given credit,” Ms. Sahni said.

They started researching neural style transfer, algorithms that can render media in different motifs, and came across Raghav Gupta. Originally from New Delhi, he was studying at the Mila AI institute in Montreal when he received a LinkedIn message from Mr. Sahni about building an AI image generator. “It was just something interesting,” Mr. Gupta said. “And it was something I could do quite easily.” He estimates it took him about 10 hours outside of his studies. Mr. Sahni anointed the application RAGHAV, in honour of its creator, and reverse-engineered the meaning: Robust Artificially Intelligent Graphics and Art Visualizer.

The app, never publicly released, works by recreating a source image in the style of another. The Sahnis chose the sunset photo because they thought it would make a good candidate for something in the vein of The Starry Night, which was used because the painting is in the public domain. With RAGHAV, Mr. Sahni selected the degree to which the dreamy swirls and other hallmarks of van Gogh’s painting would apply to the output, and Suryast was born. (Ms. Sahni reserved judgment on its artistic merits. “I’m not an artsy person,” she said. “I cannot even draw one straight line.”)

Mr. Sahni and his lawyer lobbed a variety of arguments at the U.S. Copyright Office when attempting to register the image: a work does not have to be entirely created by a human; Suryast sprang from Mr. Sahni’s artistic choices; he had input into both the source photo and the AI output; and Suryast is an original work.

Over three rejections, the copyright office disagreed: Suryast lacked sufficient human authorship; the human and AI contributions could not be untangled; it was derivative, unoriginal; and anything new in it came from RAGHAV, not Mr. Sahni. The office noted that Mr. Sahni was welcome to apply to register his original photograph, though.

The Sahnis had better luck in India, where Suryast is registered with RAGHAV as the co-author. The Indian copyright authority sent a notice revoking the registration at one point, but later withdrew it, according to Mr. Sahni.

That leaves Canada. The registration here isn’t that meaningful on its own. The copyright process is administered by the Canadian Intellectual Property Office (CIPO), which does not verify or examine claims. All it takes to register a work is an online form and a fee of $63. A work is technically protected the moment it’s created, and the real debate begins if there is a court challenge.

Open this photo in gallery:

That explains how Victoria resident Hugh Stephens registered Sunset Serenity last year, a picture of an Ontario lake generated by OpenAI’s Dall-E 2 and a corresponding poem written by ChatGPT. Mr. Stephens, who writes a blog about IP and spent much of his career in international trade, simply wondered if it would be possible. “I’m going to pursue it to the ridiculous extreme,” he recalled thinking. “I’m going to create a piece of so-called art, and a piece of absolute crappy poetry.”

The full title that he wrote for CIPO even includes the line “demonstrating minimal skill and judgment on the part of the human claiming copyright.” He wanted to register AI as the author, but CIPO requires biographical details, such as an address. And, well, where does an AI model “live” anyway?

Mr. Sahni was not deterred by such formalities. RAGHAV has an address in CIPO’s database – it just happens to be the same as his. “We did sense that somebody at that time just wanted to be a bit, let’s say, naughty, and they just let this one go,” he said.

CIPPIC, the institute challenging Mr. Sahni’s registration, might agree on that point. Director and general counsel David Fewer wrote in the court application that CIPPIC has raised its concerns about the registration with CIPO, but the office has refused to remove RAGHAV as a co-author. “You can list an author,” he said in an interview. “You can’t list a washing machine or a box of rocks.” The application charges that because of these “oversight failures,” CIPPIC had no choice but to go to court to correct the record.

He didn’t actually expect Mr. Sahni to hire a lawyer and fight the case, though. “But then he did,” Mr. Fewer said. He’s now preparing additional material to file to build his case. “A little more certainty would suit everybody,” he said. “Our goal in doing this is to start to take those steps, to give Canadians some guidance.”

First, he wants to clearly establish that only humans can be authors. As for the image itself, CIPPIC’s arguments mirror those made by the U.S. Copyright Office. Suryast was created through a “purely mechanical exercise of data entry and algorithmic luck,” according to the application, which argues there was no human skill or judgment involved.

Gary Daniel, a Toronto lawyer with Deeth Williams Wall, which is representing Mr. Sahni, declined specific questions about the case, but said that he, too, would like some answers. “Good, bad or indifferent, you want clarity,” he said.

Some aspects of the case could be tough to defend. The Copyright Act does not explicitly state that only humans can be authors, but there are many indications that is what the law contemplates. With joint authorship, for example, copyright resides with the last to die, plus 70 years. “AI can’t die because it’s not alive,” said Carys Craig, a law professor at York University who specializes in IP. “If it doesn’t die, how long does the copyright last?”

She was disappointed when she first learned of Suryast’s registration since the government had already launched an AI-related consultation in 2021. “To the extent that CIPO is recognized as an arm of government, you would think to register the work seems to put a thumb on the scale in favour of the copyrightability of AI-generated works, which I thought was both premature and misguided,” she said.

Open this photo in gallery:

The question of whether AI-generated media should be eligible for copyright protection is also the wrong one to ask, Prof. Craig said. Copyright is a tool to encourage and reward human creativity. As such, the more important question is this: “How do we encourage human authorship and human creative collaboration in the midst of a cultural environment that will be absolutely replete with AI-generated images?”

Because it is so cheap and easy to pump out AI-generated content, maybe copyright protection isn’t necessary. “To my mind, we don’t need to be encouraging and rewarding people for using AI,” she said.

Mr. Sahni argues the inverse. “I most definitely believe that an AI-generated work, be it an image or whatever, should be given protection,” he said. “Not doing so is going to discourage the adoption of technology by humanity.” He demurred on making a case for AI authorship for now, saying he first wants clarity on copyright protection for AI works.

The U.S. has so far tackled that issue by protecting the human contributions to an AI-generated work while disregarding the machine-created elements, which is no easy task. In 2023, the copyright office recognized limited protection of a graphic novel consisting of AI-generated images. The human-authored text and the arrangement of the images were granted protection – but not the images themselves.

In Canada, there will likely be more complicated cases than Suryast ahead. Say you write an elaborate prompt for an AI image generator, and refine it based on the outputs until you finally arrive at the picture you envisioned in your head. Does that count as sufficient human involvement? Maybe, Mr. Fewer said. “What a judge is going to have to do is make an assessment as to whether what’s in front of her embodies human choices that reflect skill and judgment, or whether everything is ultimately just products of a machine,” he said. “That’s going to be tricky.”

I was left with one other tricky issue. Mr. Sahni provided a copy of Suryast to run with the article. How, I asked, should the image be credited for publication? Ankit Sahni? Or RAGHAV, too?

“Haha,” he typed over LinkedIn. “That precisely is the million-dollar question.”