On the first day of the 2024 London Book Fair, panelists from the U.K., the U.S., and Canada convened to discuss the rapidly evolving situation with AI companies and copyright law.
Dan Conway, CEO of the U.K. Publishers Association; Maria Pallante, president and CEO of the Association of American Publishers; Nicola Solomon, CEO of the Society of Authors; and Glenn Rollans, former president of the Association of Canadian Publishers, made up the panel. Porter Anderson of Publishing Perspectives moderated the discussion.
Pallante gave an overview of the situation in the U.S., where there are "more than two dozen copyright and AI cases" currently pending. In most cases, Pallante said, the big tech companies behind AI have tried to frame their work as being in the public interest, and copyright as an "obstacle" holding back innovation. Typically, "if you're a downstream user" of copyrighted material, "you license what you do not own." Big tech could afford to do that, but tech companies "don't want to," though they rarely discuss it that way.
Broadly speaking, Pallante pointed out, these tech companies need three components for their large language models: technological talent, computing power, and authored works. Essentially, they want to have that third component "for free." Tech companies, Pallante explained, are saying, "we believe everything, input and output, is transformative." The AAP believes that it is "not at all." It's "not a brief snippet" that points to the original work, and it "goes to the very expressive quality of authorship." They reproduce human expression, over and over again, "in both inputs and outputs." That process "isn't transformative--it's leveraging the value of IP."
She said that on principle, "unjust enrichment" bothers lawmakers and regulators, and copyright holders are not asking for a "new legal regime," just transparency and the regulation of market substitutes. Copyright, Pallante noted, is already "on our side," and if publishers and authors can't be protected, "it isn't progress." She added that one positive thing is that lots of different licensing models are already emerging, including direct and collect licensing, and will continue to evolve, and she spoke favorably of the company fairlytrained.org, which certifies tech companies that do right by authors and publishers.
In the U.K., Conway recalled, some 18 months ago, the government was on the verge of creating a text and data-mining exception to copyright that would "effectively blow open" the current system. Publishing and other creative industries in the U.K. were able to "face that down." What ensued were "intense discussions" among copyright holders, tech companies, and the government; creative industries as well as tech companies are currently waiting for the government to announce what it has termed a "landing ground" on AI.
Conway noted that in those conversations, publishing has been heard "quite loud and clear," and while the industry does not benefit from some of the same tax credit systems that benefit the television or film industries, publishers "have actually been listened to." Part of what's been going on, Conway explained, is the difference in relative power between two departments within the U.K. government. One is the Department for Culture, Media, and Sports, in the purview of which the publishing industry would reside, and the other is the Department of Science, Innovation, and Technology. The "impasse" between the two has led to "policy paralysis." He emphasized that paralysis is the last thing anyone needs right now, as the large language models have "all the content already." In that sense, "their horse has bolted."
Regarding the "landing ground" that the government will establish, Conway said he couldn't know for sure what shape it will take. It might end up looking something like an "E.U.-style opt-out exception," where publishers would be able to effectively withhold their rights, but if they failed to do so, it would be free to use by any AI company. What the PA is hoping for are transparency measures and a "fundamental principle" from the U.K. government that copyright should be respected. He acknowledged that things may end up "in complicated legal cases."
Solomon agreed that the phenomenon of tech companies casting themselves as "being in the public interest" and the creative industries as "holding back progress" was problematic. Ultimately these large language models "are tools," and the question is "who benefits" from the use of those tools. She also called the language used to describe AI problematic. Large language models "are not ingesting, they are copying," she said, and she encouraged people to use "the language of machines" when talking about them.
She added that authors and other creatives aren't just worried about tech companies using their work--they're also deeply concerned about how consumers will be able to find human-created work in marketplaces flooded by machine-created content. Solomon recalled that when King Charles's cancer diagnosis was announced in February, more than 200 AI-generated books on the subject appeared for sale on Amazon the next day. Publishers and creatives have asked sellers like Amazon not only to label AI-generated products as such but also to allow consumers to filter out AI-generated content when searching.
Solomon said the Publishers Association and other have "done very well" in their discussions with the government, but noted that the industry is a "very small fish" in a "very big pond." She felt encouraged that around the world, people are realizing that some amount of regulation is necessary, and emphasized the need for these regulations to be established quickly. Authors and publishers can't wait for lengthy legal cases to be resolved: "We need answers now."
Rollans observed that AI has a "variety of different faces," and some of them could be extremely useful for publishers, particularly AI tools relating to data management and marketing. But the proliferation of large language models like ChatGPT has caused a "very aggressive change" in today's "information marketplace," one that the industry, and society as a whole, barely has a hand on. In his view, people perhaps overlooked "how quickly and aggressively" tech companies enter new marketplaces. Models are evolving at a "breakneck pace," tech companies are creating "defenses against transparency," and copyright "has huge issues."
He discussed the problem of publishers and copyright holders trying to compete with AI-generated products released by "less scrupulous" users, and emphasized that this problem, along with many others related to AI, is only magnified when going from the global north to the global south. In the global south, AI-generated content threatens to "overwhelm" legitimate products entirely.
Rollans also expressed incredulity that some tech companies are willing to "indemnify" users of their systems, essentially saying "if you are prosecuted for infringement, we will defend your interests there." He called it a "wild irony" that they have the capacity to do that, but "didn't have the capacity to license in the first place." --Alex Mutter