The Wikimedia Foundation Intervenes in High-Profile Court Case Involving AI Copyright Law

The Wikimedia Foundation has intervened in a high-profile court case involving AI copyright law, arguing for a nuanced approach to protect language models trained on copyrighted material without allowing chatbots to repurpose that content for users.

The nonprofit organization's amicus brief, filed last Friday, comes as publishers such as The New York Times are locked in a dispute with OpenAI and Microsoft over the use of copyrighted work by AI systems.

Wikimedia is cautioning against an overly broad or simplistic application of fair use doctrine in these cases, which it believes could have far-reaching consequences for online communities.

The organization notes that its own websites serve as crucial sources of data for training artificial intelligence and machine learning models, but also acknowledge that the reliance on these systems has been detrimental to their visitor numbers.

The Wikimedia Foundation has raised concerns that copyright laws may be too restrictive for artificial intelligence (AI) systems, potentially threatening the open web. According to the organization, services like ChatGPT reproduce text licensed under Creative Commons but fail to adhere to its terms. This is not a new issue, as AI systems often rely on freely available data, including Wikipedia's vast repository of information.

However, Wikimedia acknowledges that these large language models (LLMs) can be beneficial for their operations. For instance, LLMs can assist editors in detecting copyright violations and doxxing attempts, while also helping to identify articles that require improvement. This highlights the complex relationship between AI systems and open-source knowledge platforms like Wikipedia.

In a statement, Stephen LaPorte, Wikimedia Foundation general council, emphasized the importance of applying fair use principles carefully when it comes to LLMs. He warned that an unjust ruling could undermine the sustainability of free knowledge and hinder the development of open-source technologies. This, in turn, may limit access to information for everyone.

Wikimedia has proposed a compromise on copyright infringement related to AI systems. The organization suggests that scraping websites should not be considered copyright infringement, as long as the extracted data is used to create something new and different from the original sources. This proposal aligns with OpenAI's stance that using LLMs constitutes fair use.

The use of Large Language Models (LLMs) by companies raises questions about copyright law and fair use. If these models replicate content from websites in response to user queries, they are essentially substituting for the original work, rather than transforming it.

In such cases, AI companies should respect licensing requirements, according to Wikimedia's stance. This would mean crediting the original source of information under Creative Commons licenses, which also require any adaptations or remixes to remain under that same license.

This approach aligns with the importance of respecting the people and communities that contribute to reliable knowledge, as emphasized by Bernadette Meehan, CEO of Wikimedia. By defending open human knowledge, Wikipedia aims to ensure that expanding access to information through AI does not compromise its core values.