The U.S. government has thrown its support behind OpenAI as it continues to battlegrounds with The New York Times over the copyright issues of training AI with copyrighted content. The technology behind ChatGPT could in general be considered fair use under U.S. copyright law, the administration has maintained.
The position was outlined in a legal brief Tuesday in Manhattan federal court. The government’s involvement does not mean the case would be decided, but could bolster the stance of technology companies in other copyright battles. The controversy has expanded to a broader battle between AI creators and developers over the unauthorized and uncompensated usage of their work.
The federal government’s case is a sign of the increasing influence of artificial intelligence on the U.S. economy, tech industry and national security efforts. The officials have indicated that imposing limits on AI companies’ ability to train their AI systems would impact the nation’s competitiveness in the global market, especially given the intense competition between governments and companies to create more sophisticated AI technologies.
The short memorandum reads, ‘The United States is deeply interested that this court reject the notion that it is infringing on copyright law by training these LLMs on copyrighted texts,’ and highlights their concerns about scientific progress and national security. The argument is not just one between technology companies and copyright holders, but rather about the need for AI development in the national interest.
That’s a stance that has been vehemently rejected by the New York Times. In the newspaper’s criticism, a spokesperson condemned the administration for prioritizing interests of big technology firms and neglecting the interests of writers and journalists, among others, the newspaper said, who have been used to create AI systems.
The spokesperson said, “There’s a way AI and creators can coexist – AI companies just need to pay them fairly for the content that makes their AI possible, and copyright law does. The claim raises one of the key issues that has also been on the surface in many ongoing copyright battles with AI these days: how can companies use so much of existing creative work to train AI without permission or having to pay licensing fees?
The controversy is especially notable since today’s AI systems require vast amounts of data to train. Large language models are trained on enormous amounts of text and patterns in the text. This enables them to elicit reactions, summarise, answer questions and carry out many other activities. If books, articles, photographs, music and other creative works are copyrighted, however, the question of the legality of such use arises when training data includes copyrighted material.
This debate is growing more complex, as copyright law was established decades before generative AI had been invented. The courts are being called upon to apply mature concepts of copyright to technology capable of processing and reproducing information in volumes which were not envisioned by many of the copyright rules in existence.
American leadership in the field of artificial intelligence has been a top priority for the U.S. administration as well. U.S. Associate Attorney General Stanley Woodward stated in a press release that AI dominance is essential in support of national security, prosperity and economic mobility for all Americans. This Administration will never allow our Nation to be at a disadvantage due to a misunderstanding of copyright law.
The government’s involvement is also timed by calls from American authorities for other governments to do the same, even as they seek to fulfill the concerns of artists and others. U.S. Commerce Secretary Howard Lutnick also called on G20 officials to advocate for fair use and give AI companies permission to train their systems on creators’ work while also looking for ways to safeguard artists.
The government’s stance comes at a critical moment for OpenAI. The company has been sued numerous times by authors, publishers, media outlets and other copyright holders for using their content to train AI models without their consent. Other tech firms such as Anthropic and Meta Platforms have been accused of similar offenses with their AI systems.
The lawsuit by the New York Times was filed in 2023 and includes OpenAI and Microsoft, one of its biggest financial supporters. The newspaper states that their millions of articles have been used without consent to train OpenAI’s chatbot AI technology. The company has refuted the claims and has stated its support for the wider use of publicly available content for AI training.
The case of The New York Times is thus not just a battle between a newspaper and a tech firm. The ultimate result might have implications for the way AI builders acquire and leverage training information throughout the tech industry. A decision might impact publishers, authors, artists, musicians, and other creators whose work increasingly is involved in the discussion on AI.
The issue that lies at the heart of the controversy is that of “fair use.” There are some instances of copyright material that can be used without permission under the U.S. copyright law. The courts consider a number of factors in deciding whether a specific use is qualified. An important question is whether or not the new use alters the material used and whether or not it has a different function. The commercial use and the market impact of the use of the original work may also assume significance.
Training a model is very different from merely republishing copyrighted content, argue AI companies. They argue that models learn statistical and linguistic patterns from very large databases of data, instead of being just databases of the original texts. But, copyright owners say that the training process still involves copying and processing their content, and that the AI models can sometimes produce results similar to the original content.
