Wikimedia Foundation Proposes Nuanced AI Copyright Framework
The organization warns that extreme legal rulings could jeopardize the open web's future.
1Wikimedia Enters Legal Debate
The Wikimedia Foundation filed an amicus brief regarding the ongoing copyright litigation between The New York Times, OpenAI, and Microsoft. The nonprofit seeks a middle ground, arguing that current legal positions from both sides are overly simplified. They aim to protect the open web while acknowledging the role of large language models.
2The Fair Use Dilemma
Wikimedia contends that training large language models on copyrighted work should be considered fair use. However, they argue that chatbots regurgitating specific articles to users crosses a legal line. An overly restrictive ruling could hinder innovation, while an overly permissive one might allow companies to ignore essential licensing terms for content.
3Existential Risks for Wikipedia
Wikipedia serves as a primary data source for AI training, yet these systems often siphon away human visitors. This traffic loss threatens the sustainability of the volunteer communities that maintain the site. Furthermore, Wikimedia claims that AI services often violate Creative Commons licenses by reproducing text without providing the required proper attribution.
4AI as a Tool for Editors
Despite copyright concerns, Wikimedia acknowledges that large language models offer significant benefits to their platform. These tools assist human editors in fighting vandalism, identifying potential copyright violations, and detecting doxxing attempts. They also help identify articles requiring improvement, demonstrating that AI can support the collaborative process of building human knowledge.
5Publisher and Developer Positions
The New York Times argues that scraping their website to build training datasets constitutes copyright infringement. Conversely, OpenAI and Microsoft maintain that their model training is transformative and qualifies as fair use. Microsoft specifically claims that their Copilot system rarely reproduces full sentences, thus not undermining the original work's purpose.
6Proposed Licensing Compromise
Wikimedia proposes a clear distinction: scraping for pre-training should be fair use, but outputting content that supplants original engagement requires a license. If an AI model delivers copies of material to users, it must respect existing licensing requirements. This ensures that original creators receive credit and maintain control over their intellectual property.
7Creative Commons Compliance
For Wikipedia, the core issue involves adhering to Creative Commons licenses, which mandate that any adaptations or remixes remain under the same license. Wikimedia argues that AI companies must respect these terms when using their data. This approach protects the collaborative ecosystem that relies on open and accessible human knowledge for growth.
8Balancing Innovation and Sustainability
CEO Bernadette Meehan emphasizes that while expanding knowledge through AI is a powerful goal, it must respect the contributors who make information possible. Finding this balance is a shared responsibility for the entire information ecosystem. Failure to apply careful fair use principles could ultimately stifle open source technologies and limit public access.