Wikipedia operator warns bad AI copyright ruling threatens web
The Wikimedia Foundation files an amicus brief in the NYT v. OpenAI lawsuit, supporting AI training as fair use while opposing article regurgitation.
Stock photo for illustration only, not from the actual event
- Wikimedia states AI training should be fair use while article regurgitation should not.
- Warns that overly black-and-white rulings could threaten the sustainability of free knowledge.
- Large language models are essential tools for Wikipedia editors to combat vandalism.
- Proposes a middle-ground compromise to protect both copyright holders and open-source tech.
The Wikimedia Foundation is staking out a nuanced middle ground in landmark artificial intelligence copyright litigation, urging courts to legally protect large language models (LLMs) training on copyrighted work while opposing chatbots that regurgitate that material directly to users. The organization cautions that an overly simplified legal ruling could either strip Wikipedians of useful AI tools or endanger the survival of the open web.
In an amicus brief filed on Friday, the nonprofit organization behind Wikipedia weighed in on a series of lawsuits where publishers including The New York Times are taking legal action against OpenAI and Microsoft, intentionally taking neither side. Wikimedia expressed concern that the litigants' legal positions push for overly broad and oversimplified applications of the fair use doctrine regarding artificial intelligence innovations.
Stock photo for illustration only, not from the actual event
At the same time, Wikimedia pointed out that its websites have evolved into foundational data sources utilized for training virtually all artificial intelligence and machine learning models. However, these same automated systems siphon away human visitors, undermining the collaborative communities that maintain the sites. Wikimedia argues that services like ChatGPT also violate copyright laws by reproducing text licensed freely under Creative Commons without adhering to the required terms of those licenses.
The Wikimedia Foundation's perspective highlights the profound economic and structural tensions reshaping the digital knowledge ecosystem. While open knowledge projects depend entirely on human collaboration, AI systems harvest these public contributions without ensuring reciprocal sustainability. Striking a legal balance between fostering open-source innovation and protecting source creators remains one of the defining regulatory challenges for the technology sector.
Despite these concerns, the organization acknowledged that large language models provide tangible benefits for maintaining Wikipedia, helping editors combat vandalism, detect copyright violations, and identify articles requiring improvement.
"A ruling that does not apply fair use principles carefully could threaten the sustainability of free knowledge, and stifle the development of open source technologies, ultimately impacting access to information for everyone in both cases."
Stephen LaPorte, Wikimedia Foundation General Counsel
Meanwhile, The New York Times continues to advocate for a more expansive definition of fair use as applied to large language models, alleging that OpenAI infringed on its copyright by scraping website data to construct training datasets. Conversely, both OpenAI and Microsoft maintain that utilizing LLMs constitutes fair use, with Microsoft noting in a previous filing that expert analysis showed Copilot rarely reproduces full sentences from grounding news content.
Wikimedia has proposed a practical compromise, arguing that website scraping should not be categorized as copyright infringement, and pre-training LLMs without a license should qualify as fair use when delivering something new and distinct from original sources. However, if those models extract copies of copyrighted material in response to user queries and supplant user engagement with original publishers, such actions should cross legal boundaries.
Source: The Verge
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment