The Wikimedia Foundation has filed an amicus brief in copyright litigation involving The New York Times, OpenAI and Microsoft, Lauren Feiner reports for The Verge. The nonprofit behind Wikipedia takes neither side. It wants courts to distinguish certain uses of copyrighted work to train AI from chatbot replies that copy and substitute for the original work.
Wikimedia argues that pretraining language models on copyrighted material without a license can be fair use when the models provide something new and different from their sources. But when a model supplies copies in response to queries and supplants readers’ engagement with the originals, Wikimedia says that use should require a license. Its argument does not treat every use of training data or every chatbot answer alike.
For Wikipedia material, Wikimedia says AI companies reproducing it should respect its Creative Commons license by crediting the original source and keeping remixes or adaptations under the same license, according to The Verge. The foundation also says language models can help Wikipedia editors fight vandalism, detect copyright violations and doxxing, and find articles that need improvement.
Wikimedia warns that an overly broad victory for publishers could let copyright holders gatekeep transformative uses, while an overly broad victory for AI companies could let them disregard licensing terms and disrupt the connection between authors and readers. Stephen LaPorte, Wikimedia’s general counsel, warned in a statement quoted by The Verge that a ruling that fails to apply fair use carefully could threaten free knowledge and stifle open-source technologies.
Comments