OTTAWA – The battle between AI companies and copyright holders notched an early win for publishers in the U.S. in mid-February when a court ruled that a legal research firm didn’t have the right to use a rival’s content.
But even as the number of legal cases grows, a definite answer to the question of whether artificial intelligence companies can use copyrighted content to train their AI products is still a long way off.
“We’ve been having this conversation for quite some time already,” said Carys Craig, a professor at York University’s law school who specializes in intellectual property. “But it’s still early days.”
“There’s a lot of things going on at the same time, and it’s not clear at all where all these balls in the air are actually going to land.”
Generative AI can create text, images, videos and computer code based on a simple prompt, but the systems must first study vast amounts of existing content.
A coalition of Canadian news publishers, including The Canadian Press, is suing OpenAI in an Ontario court for using news content to train its ChatGPT generative artificial intelligence system. There haven’t been any developments in the case since it was launched in late November.
In mid-February, a group of major U.S. media companies and the owner of the Toronto Star filed a copyright infringement lawsuit against Canadian artificial intelligence company Cohere in a New York court.
It followed a series of similar lawsuits launched in the United States, including some involving news publishers. The New York Times is suing OpenAI and Microsoft, while the owner of the Wall Street Journal and New York Post has targeted Perplexity, an AI-powered, conversational search engine. Some of these lawsuits date back to 2023.
In mid-February, a U.S. court ruled that Ross Intelligence, a now-defunct legal research firm, was not permitted under U.S. copyright law to use content from Thomson Reuters’ own legal platform, Westlaw, to build a competing platform.
Jane Ginsburg, a professor at Columbia University’s law school who studies intellectual property and technology, said “there has been only one case of 20 or 30 that has gotten to the stage of being decided on the merits of the infringement claim and the fair use defence.”
“All of the other cases, to my knowledge, are either just recently filed or are in preliminary stages, and are being addressed largely on procedural issues rather than the substantive copyright issues,” she added.
Craig said the U.S. decisions have no bearing on what happens in Canada and “are certainly not authoritative in any legal way.”
But courts dealing with such a novel issue may still look to previous cases for guidance.
“I suspect that the American cases will prove important but not determinative of the direction that Canada takes,” Craig said.
She noted that because the various cases involve different platforms with different technical qualities, “it’s not always clear exactly how far-reaching their particular reasons or rulings will be.”
While the courts are interpreting existing law, Ottawa has been consulting on how it could update Canada’s copyright legislation to confront the emergence of generative AI.
Canadian creators and publishers want the government to do something to rein in companies using their content to train generative AI. Artificial intelligence companies, meanwhile, maintain that using the material for training doesn’t violate copyright and limiting its use would restrict the development of AI in Canada.
The federal government recently released a “what we heard” report on those consultations. It said the government “continues to consider how Canadian concerns posed by generative AI, including those raised by cultural and technology industries, might be addressed.”
In the U.K., the government is consulting on whether to let tech firms use copyrighted material to help train AI models if the creators do not explicitly opt out.
That led 1,000 musicians to sign their names to a silent album in protest. Elton John and Paul McCartney have spoken out against the plan and some British newspapers have run wraparounds over their front pages criticizing the government consultation.
Craig said that in Canada, the “lack of consensus coming out of the consultations means that ultimately a policy decision is going to have to be made.”
“And that policy decision is going to depend, I think, on … politics and to some significant degree on what other jurisdictions are doing, in particular, developments in the U.S. and developments in Europe,” she said.
“So that’s a lot of stuff that still has to shake out.”
In Canada, any changes to the law almost certainly will have to wait until after a federal election, which could be called within weeks. The long period of uncertainty may lead the parties involved to reach licensing agreements before these questions are settled.
Craig said the hope among cultural industries and publishers is that they can establish that using materials for AI training can result in copyright liability.
“Then you have the baseline from which to begin to negotiate, not just settlements in particular cases, but to negotiate transactional licenses or to look to policy makers for collective licensing solutions,” she said.
Ginsburg said more and more publishers are licensing their content to AI companies. The Associated Press, for instance, has signed deals with both OpenAI and Google’s Gemini.
She said the quality of the data AI companies can get by scraping the internet or using pirated books isn’t very good, and as AI outputs get put back on the internet and are re-scraped, the data is continually degraded.
“I suspect that the need to have good-quality source data will ultimately push the copyright owners and the AI companies closer to the negotiating table,” Ginsburg said.
— With files from Tara Deschamps and The Associated Press
This report by The Canadian Press was first published March 2, 2024.
