A full year before generative artificial intelligence services swept the public’s imagination and spooked creative workers everywhere, the Writers’ Union of Canada asked Parliament to protect Canadian authors from the damage the tools could cause.
On behalf of its 2,300 members, the professional organization implored policy-makers in a fall 2021 letter to back “a respectful market for licensing” in copyright law as AI companies profit “relentlessly” from the hard work of authors. It’s a position it has taken more seriously with each passing month, especially since public-facing tools such as ChatGPT have brought AI-powered text generation to people around the world.
“It feels like imperialism,” said Danny Ramadan, the Syrian-Canadian novelist who won the Lambda Literary Award for Gay Fiction for The Foghorn Echoes and chairs the Writers’ Union. “If someone is going to use my work for any kind of commercial purpose, it is my right to be compensated.”
The power struggle over books in the AI era came to a head late last month when the U.S. Authors Guild published an open letter to generative AI services demanding that they ask permission to use writers’ work and compensate them accordingly. More than 8,000 writers, including Margaret Atwood, have since signed it. At the same time, Canada’s Mona Awad and American author Paul Tremblay began making a case for a class-action lawsuit against ChatGPT creator OpenAI for infringing on the copyright of books such as Awad’s Bunny. Soon, comedian Sarah Silverman and two other authors launched a nearly identical campaign to start a class action against OpenAI and Facebook owner Meta.
The long, arduous process of class actions likely won’t yield results any time soon for authors hoping to keep their books from being digested and summarized by ChatGPT or similar services, such as Google’s Bard, without a licence.
But they illustrate the precariousness that all creatives must now live under, as even visual artists now worry about AI aping their styles and film unions warn that legions of background actors and their support staff could lose their jobs to AI replicas.
“I think there needs to be a pause, and that legislation and regulation have to catch up with AI development‚” said Glenn Rollans, the publisher of Edmonton’s Brush Education imprint and chair of the Association of Canadian Publishers copyright committee. Until then, “it’s like grabbing hold of a moving train.”
The Writers’ Union has been warning about AI for more than a decade, ever since it supported the Authors Guild in its unsuccessful attempt to take Google to task over its massive Google Books repository. U.S. courts ruled the tech giant’s efforts amounted to fair use of copyrighted books, and the publishing industry started to realize that the California company’s scanned snippets were the beginning, not the end, of the internet’s copyright challenges.
“We figured these large tech firms were preparing larger sets of data they could use,” said John Degen, the Canadian organization’s executive director. “We were right.”
When OpenAI launched ChatGPT to the public late last year, a flood of similar services hit the market. As authors played with these tools, some were surprised by just how sharply they could spit out accurate summaries of their books.
Awad’s lawyers alleged in a court filing last month in San Francisco that OpenAI had “made copies of [her] books during the training process” of its language models. (Through a publicist, Awad declined to comment on the ongoing legal matter. OpenAI did not respond to a request to comment, but CEO Sam Altman told U.S. senators this year that content owners should “get significant upside benefit” from the technology.)
The legal complaint alleges that “ChatGPT generates summaries of Plaintiffs’ copyrighted works – something only possible if ChatGPT was trained on Plaintiffs’ copyrighted works,” adding that the service profits “richly” as a result. Moreover, the service may have digested repositories of hundreds of thousands of books, some of which the filing calls “flagrantly illegal shadow libraries” that may have violated authors’ copyrights.
The allegations in Awad and Tremblay’s lawsuit have not been tested in court. But the case may still reveal more about the inner workings of ChatGPT than is currently known – including, possibly, the source material behind OpenAI’s book summaries.
“All of this is based on claims of copyright infringement, but right now the basis of that claim is ambiguous and people are not fully certain of exactly how their works are being used,” said Michael Duboff, an entertainment lawyer with Edwards Creative Law in Toronto.
There are allowances in Canadian copyright law for “fair dealing,” for instance, that allow Canadians to use copyrighted works for research, education, news reporting and other reasons. But Duboff points out that such endeavours are non-commercial and that many generative AI services require payments or subscriptions.
Such lawsuits may help set precedents in courts – which, though slow, still tend to move faster than federal policy-making. Canadian authors could get additional protections in the long term from Bill C-27, a long-overdue attempt to modernize privacy laws and build protections around AI systems.
Some of the proposed legislation’s language around mitigating the risk for Canadians of vaguely defined “high-impact systems” could be used to target AI systems that may harm creative work such as books, says Valentine Goddard, a lawyer and inter-arts curator who is consulting with Ottawa on the new AI policy.
“Whatever impact Bill C-27 has, there will be a lot of blind spots, so we need sectoral consultation,” said Goddard, who is also the founder of the Art Impact AI program, which seeks to bring artists’ voices to the fore in generative AI debates.
But for future generations of authors, it may be beneficial to give AI regulations some breathing room, says Sean Michaels, the Giller Prize winner for his 2014 book Us Conductors. He used generative AI in putting together his latest novel, Do You Remember Being Born?, which examines the relationship between art and AI and will be published in September.
He likens the current conversation to the one around music sampling in the early nineties – which ended with strict copyright rules and has for years impeded the art of sampling, particularly in hip hop.
“I always think it’s a good idea for independent artists to demand as much compensation as possible from big corporations,” Michaels said in an e-mail. But “the challenge is how to navigate the need to defend our livelihoods alongside the debt we owe our kids – who will dream up masterpieces, travesties and entire new genres using all the tools we don’t like, don’t get, and would sooner see abolished.”