New Delhi:

A federal judge had issued a ruling in favor of Meta Platforms, the parent company of Facebook, dismissing a copyright infringement lawsuit brought by a group of authors who accused the company of using their works to train its artificial intelligence technology. The ruling, delivered by US District Judge Vince Chhabria, marked the second significant dismissal of copyright claims against the growing AI industry from San Francisco’s federal court within the same week. Judge Chhabria determined that the 13 authors involved in the lawsuit “made the wrong arguments” and subsequently dismissed the case. However, he clarified that the ruling was specific to the authors in this instance and did not imply that Meta’s use of copyrighted materials was necessarily lawful.

Attorneys representing the plaintiffs, which included notable figures like comedian Sarah Silverman and authors Jacqueline Woodson and Ta-Nehisi Coates, had not provided immediate comments following the judgment. Meta also refrained from making a statement.

Failure to present sufficient arguments

Chhabria expressed that the ruling should not be interpreted as indicating that Meta’s utilisation of copyrighted materials for training its language models was legal. Rather, he emphasised that the case highlighted the authors’ failure to present sufficient arguments and evidence.

Despite Meta winning the motion to dismiss, this victory might be short-lived. In his comprehensive 40-page ruling, Chhabria suggested that Meta and similar AI companies might be engaging in serial copyright infringement by utilising books and other human-created works for technology training, indicating that other authors might be encouraged to file cases that could proceed to trial.

Compensate copyright holders

The judge raised the question of whether companies had been involved in illegal activities by incorporating copyright-protected materials into AI training models without authorisation, suggesting that the answer would likely be affirmative in most cases. Chhabria inferred that in many situations, copying copyright-protected works to train generative AI models would be illegal without permission, implying that companies would need to compensate copyright holders to avoid copyright infringement.

He also dismissed arguments claiming that enforcing copyright laws established decades ago would hinder technological advancements at a critical juncture. Chhabria noted that while the technology was indeed groundbreaking, the notion that unfavorable copyright rulings could impede its development was absurd. He stated that if the companies believed that using copyrighted works was essential for training their models, they would ultimately find a way to appropriately compensate copyright holders.

Anthropic case

Earlier in the week, Judge William Alsup, from the same courthouse, ruled that AI company Anthropic had not violated the law by training its chatbot Claude on millions of copyrighted books. However, the company was still required to go to trial for allegedly acquiring those books from unauthorised websites instead of purchasing them legally. Alsup asserted that the process by which an AI system derived information from numerous written works to generate its own text was considered ‘fair use’ under US copyright law because it was deemed ‘quintessentially transformative’.

Chhabria critiqued Alsup’s reasoning regarding the Anthropic case, pointing out that Alsup concentrated heavily on the transformative aspects of generative AI while neglecting the potential harm it could pose to the market for the works used in training. Chhabria suggested that a case could indeed be made regarding such harm.

Lawsuit against Meta

In the lawsuit against Meta, the authors contended that the company was ‘liable for massive copyright infringement’ for sourcing their books from online platforms that housed pirated content and utilizing them in Meta’s flagship generative AI system, Llama. Their attorneys argued that extensive and uniquely composed passages found in books were crucial for effectively teaching generative AI chatbots the nuances of human language, asserting that ‘Meta could and should have paid’ for the rights to use and license those literary works.

U.S. copyright law

In recent court filings, Meta claimed that U.S. copyright law permits the unauthorised copying of a work if it transforms it into something new. The company asserted that the AI-generated content produced by its chatbots is fundamentally different from the books on which they were trained. Meta’s attorneys pointed out that, after nearly two years of litigation, there still had been no evidence to suggest that anyone used Llama as a substitute for reading the plaintiffs’ books or that they could do so. They argued that Llama does not output the actual works it has drawn from, even when prompted.