Wikipedia urges courts to find a middle ground on AI copyright

The middle ground approach

The Wikimedia Foundation is weighing in on a landmark legal battle between The New York Times and AI giants like OpenAI and Microsoft. As reported by The Verge, the nonprofit is taking a nuanced stance that avoids siding entirely with either the publishers or the tech companies. They are concerned that a simple, black-and-white ruling on fair use could have unintended consequences for the internet.

The foundation argues that the current legal arguments from both sides are too broad. They fear that a victory for publishers could allow copyright holders to gatekeep information, while a victory for AI companies could let them ignore licensing terms entirely. Wikimedia wants the court to recognize the difference between how a model learns and how it speaks to a user.

According to the filing, the goal is to protect the flow of knowledge between human authors and audiences. Wikimedia wants to ensure that AI remains a tool for innovation rather than a way to fragment the information ecosystem. They are advocating for a legal framework that respects the collaborative process of creating open knowledge.

Training vs. Regurgitation

At the heart of Wikimedia's argument is the distinction between pre-training a model and generating a response. They suggest that scraping websites to build training datasets should be considered fair use if the resulting model provides something new and different from the original sources. This allows the AI to learn patterns and facts without directly copying the work.

However, the foundation draws a line when a chatbot pulls specific material in response to a user query. If a system delivers a copy of a work while supplanting a reader's engagement with the original source, Wikimedia argues that it should require a license. In these instances, the AI output is simply substituting for the original work rather than transforming it into something new.

This distinction is critical for how the law treats different types of AI interactions. For example, using data to teach a model how to write a poem might be transformative, but providing a near-verbatim summary of a news article to a user might not be. Wikimedia believes the law must be able to tell the difference between these two actions.

Impact on the open web

The Wikimedia Foundation notes that its websites have become some of the most critical sources of data for machine learning. Because of this, the organization is uniquely positioned to see how AI affects the sustainability of free knowledge. They argue that a ruling that ignores fair use principles could stifle the development of open source technologies.

On the other hand, the foundation points out that these same AI systems can siphon away visitors from Wikipedia. This undermines the community of volunteers who maintain the site and provide the data that the AI relies on. They are trying to balance the benefits of AI tools with the need to protect the human contributors who make reliable knowledge possible.

The organization also highlights a specific issue regarding Creative Commons licenses. They argue that services like ChatGPT often reproduce text licensed freely but do not adhere to the specific terms of those licenses. Wikimedia wants to ensure that any remix or adaptation of their content remains under the same open standards that the original contributors intended.

Tools and sustainability

Despite their concerns about copyright, Wikimedia acknowledges that large language models provide significant utility to their editors. These systems can help the community fight vandalism, detect doxxing, and identify articles that need improvement. The foundation views these capabilities as valuable tools for maintaining a high-quality information repository.

The organization's stance is that the court needs to understand the collaborative nature of Wikipedia. They believe that the platform requires a strong foundation of accessible human knowledge to thrive. If the rules become too restrictive, it could impact access to information for everyone, regardless of whether they use AI or traditional search methods.

Ultimately, Wikimedia is seeking a balance that serves the entire information ecosystem. They want to expand access to knowledge through AI while respecting the people and communities that create it. Their goal is to ensure that the growth of technology does not come at the expense of the open web's underlying infrastructure.