On Wednesday, the Landmark AI Copyright Judge Kadrey, et al. v. MetaPlatforms Inc. He ruled with Meta's favor. And US District Judge Vince Chhabria seemed reluctant to do so, calling his own ruling “a serious tension with reality.”
Thirteen authors, including Sarah Silverman, Tanehishi Coates and Junot Diaz, sued Meta for unlicensed use of the book to train the Lama AI model.
The facts of the incident seemed particularly bad. Not only did he make unlicensed copies of the author's work, Internal meta message It was shown to have been revealed during the discovery Company's own employees He expressed legal and ethical doubts about pirated these works. Other messages suggest that employees are trying to eliminate traces of copyright infringement and are looking for words like “stolen” or “pirated” as part of their team's “mitigation” efforts.
Instead of solving the troublesome copyright fight over AI training, Chhabria's ruling adds another layer of complexity to this legal issue.
Just a day ago, a judge in a similar AI copyright lawsuit ruled in favor of another AI company, humanity. In the same northern district of California, US District Judge William Alsp Bartz v. Humanity Humanity uses pirated books in Shadow Libraries Books3 and Libgen (the same data set in metacase). It was used fairly.
However, Robert Brownie, an intellectual property law professor at George Washington University Law School, said Justices Alsup and Chhabria used dramatically different inferences. In both cases, it falls on the legal doctrine of fair use, particularly the fourth factor of such defenses: potential market harm.
“Judge Alsup has a very narrow view. If the generative AI output itself does not infringe the specific work used to train the model, then it cannot be taken into consideration by people caused by people caused by people caused by people caused by people caused by people caused by people caused by people caused by people caused by people caused by people caused by people caused by people caused by people caused by people caused by people caused by people caused by people caused by people caused by people who use AI output. Kadreyv. Meta.
“Judge Chhabria says it's wrong. The harm caused by 'diluting' the market for training work can be considered. Also, the harm of serious market dilution can even outweigh the high levels of conversion under the original factors. ”
Thus, both judges sided with fair use arguments, but their rationale for opposition laid the foundation for a complex, fragmented legal landscape.
Cadley's plaintiffs disputed fair use debate.
The plaintiff attempted and failed to oppose the defense of fair use of meta. In a blog post written after the oral debate on May 1, Kevin Madigan, senior vice president of policy and government affairs for the Copyright Alliance, wrote that the plaintiff's lawyers “shockingly” did not present any potential rebuttal.
Of the four fair use factors, regardless of whether the use is transformative and whether the use is deforming and whether the copyright harms existing or future markets, regardless of factor 4. Chhabria supported the meta in Factor 1. “There is no serious question about Meta using plaintiffs' books. It had a “further purpose” and a “different personality” than the book.
The deliberation then became a fourth factor, or a market hazard. There, Chhabria spoke a lot about the plaintiff's lawyer's argument. They failed to simply claim that meta caused harm to the market.
In discussing the harms of the market during oral discussions, Chhabria nurtured a hypothetical, future Taylor Swift.
“Even if 1 million songs were produced [Meta’s Llama] Taylor Swift's song style model has no effect on the market for Taylor Swift's song. Chhabria asked Meta Attorney Kannon Shanmugam.
Chhabria seemed to foresee his final verdict when he questioned plaintiffs' lawyer David Boies about evidence of market harm.
Masculine light speed
“Whether that's on summary judgment records or not, it appears you're asking me to assume that the market for Sarah Silverman's memoirs will be affected by the billions that can ultimately be produced.
“If you can't show that the market for copyrighted works used to train your models is dramatically affected, you'll lose.”
Ultimately, Chhabria decided that Meta had a stronger argument.
“Meta has defeated the plaintiff's half-hearted argument that its copy would cause or threaten serious market harm,” Chhabria said. “The conclusion may actually be in serious tension, but it is determined by the choices made by the plaintiffs… we cannot present meaningful evidence regarding the effectiveness of training in LLMS like llamas. [AI-generated] Books. ”
On the day of the sentencing, a Meta spokesman provided this statement to Mashable. “We appreciate today's decision from the courts. The open source AI model promotes transformative innovation, productivity and creativity among individuals and businesses, and the fair use of copyright materials is an important legal framework for building this transformational technology.”
In his decision, the district judge said his decision was not about fair use defense of training AI models using pirated books, nor about the shortcomings of plaintiffs' arguments. “The court had no choice but to grant Meta a summary judgment,” Chhabria said.
“Because this is not a class action, this ruling only affects the rights of these 13 authors. It does not affect the myriad other authors that the work used to train the model. And, as it should be clear now, this ruling does not represent the proposition that the use of copyrighted material to train the work model will train the linguistic model.
His ruling also leaves the door open for other artists to file similar copyright lawsuits against Meta and other AI companies. Chhabria even assumed that “it is illegal to copy copyrighted works to train generated AI models without permission.”
However, this ruling also has symbolic significance to the artist.
“If this case comes up and says that training large-scale language models on pirated datasets that have been stripped of copyright information will be used fairly, that is a horrifying and horrifying outcome for millions of creative experts around the world,” he said. Justin HughesLaw professor at Loyola Law School, in an interview with Mashable before the verdict.
AI is already impacting creative living
Kadreyv. Meta It is one of dozens of copyright lawsuits against AI companies. AI blog at the time of publication ChatGpt eats the world 39 ongoing cases were counted.
However, the courts deliberately say that generative AI is already having a major impact on the creative industry.
The ability of generation AI to automate the creation of text, images, videos and audio is already replacing creative work. 2024 Imperial College London Business School and Berlin Economics Researcher The paper has been published Analyze how the generated AI is affecting the labor market. Since the introduction of ChatGpt, they have found that “there is almost instantaneous reduction in online gig worker posts across types of work, but almost instantaneous reductions, especially in jobs where automation is prone to occur.” The most impacted work was writing gigs, which reduced by 30%.
a 2023 Report It says it was commissioned by the Animation Guild to measure the impact of generative AI in the entertainment industry. “Nearly two-thirds of the 300 business leaders surveyed hope that Genai will play a role in consolidating or replacing existing business functions over the next three years.
Many artists see the existence of AI tools like Llama as an existential threat. In humiliation to the injuries, the AI models were trained in the very human representations they have been accused of replacing.
in Amikos Brief In support of the plaintiffs, the American Publishers Association argued that the case was much simpler than it looked. Meta asks that companies over $1 trillion in dollars declare the court to make appropriate and commercial use of the content of copyrighted works on a large scale, without permission or payment of its content.
What's going to happen now?
While Meta won the fair use ruling, Madigan called Chhabria's decision “a mixed bag.”
“What's bad for copyright holders is Judge Chebria's treatment of transformative use under the first factor and he doesn't want to recognize the licensing market under the fourth.” Here, Madigan mentioned the potential loss of the plaintiff's licensing transaction. This is an argument that Chhabria said he would not consider.
“However, what's not necessarily the worst thing in the world is that it's so overshadowed that we didn't develop details and records of this case and raise any particular issues,” Madigan continued. He added that it is likely that the plaintiffs will appeal too.
A spokesman for Boies Schiller Flexner, a company representing the plaintiffs, told Mashable: “The court ruled that AI companies are generally violating the law without obtaining permission from copyright holders, without supplying or paying copyrighted work to the model. We oppose that conclusion in honor of it.” They did not respond to the question of whether they would file an appeal.
Kadreyv. Meta and Bartz v. Humanity They are often put together as they both focus on inputting pirated books as data to train AI models. In contrast, other well-known AI copyright cases – New York Times Litigation For Openai and MicrosoftAnother case for humanity from major record labels (Concord v. Humanity), and recently Disney v. Mid Journey – Focus on the output of the AI model.
Regarding these cases, “They are all showing evidence of infringement output. [Kadrey v. Meta] Madigan said. “In cases focused on output, “There is no need to enter into these abstract doctrinal discussions about whether transformational use and training are transformative in their purposes. We need to display verbatim copies of sorts,” he continued.
topic
Artificial Intelligence Meta
