Wikimedia Foundation warns a careless AI copyright ruling could hurt the open web

The Wikimedia Foundation, the nonprofit behind Wikipedia, has staked out a middle ground in a series of landmark cases on artificial intelligence and copyright. On Friday it filed an amicus brief in lawsuits brought by publishers, including The New York Times, against OpenAI and Microsoft, siding with neither party.

The danger of extremes

In the foundation's view, both sides are pushing for "overbroad and simplified" applications of the fair use doctrine to AI. An unqualified win for publishers, Wikimedia fears, would let rights holders "gatekeep" even innovative and transformative uses of their work. An unqualified win for AI companies, by contrast, would be "a green light to snub licensing terms" and fragment the flow of knowledge between authors and audiences.

For Wikipedia, the outcome could be existential. The filing notes that Wikimedia sites have become one of the most critical sources of data for AI and machine learning training. Yet the same systems draw visitors away, undermining the communities that maintain those sites. The foundation also argues that services like ChatGPT reproduce text licensed under Creative Commons without honoring the terms of that license.

At the same time, large language models are useful to Wikipedia itself: they help editors fight vandalism, detect copyright violations and doxxing, and find articles in need of improvement.

The proposed compromise

The foundation argues that scraping websites should not be treated as copyright infringement, and that pre-training models on copyrighted works without a license should count as fair use when the result offers "something new and different from the original sources." But if a model retrieves copies of material in response to a user query and delivers them in a way that supplants engagement with the original, that should require a license. In that case, Wikimedia reasons, the AI output merely substitutes for the original work rather than transforming it.

For Wikipedia, honoring its license would mean crediting the original source under Creative Commons terms, which also require any remixes or adaptations to remain under the same license.

Where the parties stand

The New York Times alleges, among other things, that OpenAI infringed its copyright by scraping its website to build training datasets. OpenAI and Microsoft maintain that using language models constitutes fair use. OpenAI has called regurgitation of work "a rare bug." In a September filing, Microsoft cited expert analysis finding that Copilot rarely reproduces full sentences from news content, which it said "hardly undermines the transformative purpose" of model training.

Wikimedia general counsel Stephen LaPorte warned that careless application of fair use principles could threaten the sustainability of free knowledge and stifle open source development. CEO Bernadette Meehan added that expanding access to knowledge through AI is a worthy goal, but it must respect the people and communities who make reliable knowledge possible in the first place.

Source: The Verge