Wikimedia: Training should qualify as fair use, copying in answers should not

Wikimedia Foundation, the organization behind Wikipedia, filed an amicus brief on October 30 with the US District Court for the Southern District of New York. According to The Next Web, it is a motion for leave to file the brief, and the court has not yet decided whether it will be accepted. An amicus brief is a submission from an outside party that wants to give the court its view on a case.

The brief is part of In re: OpenAI, Inc. Copyright Infringement Litigation, which consolidates 12 lawsuits. The lead case is The New York Times Co. v. Microsoft and OpenAI. According to The Next Web, it is being heard by Judge Sidney H. Stein.

Two kinds of use, two assessments

Wikimedia sides with neither the plaintiffs nor the defendants. Instead, the organization proposes that two forms of use be judged differently under the US fair use doctrine, which in certain cases allows the use of protected material without permission.

The organization argues that using protected content to train large language models should be fair use when the models create new and different products from the source material. No license should be required in that case.

Wikimedia takes a different view of RAG and similar techniques, in which a chatbot looks up sources to answer a question. Here, Wikimedia does not believe the use is fair use. According to the organization, companies should respect license terms, including Wikipedia's CC BY-SA 4.0 license, which requires attribution.

According to The Next Web, the brief states: "While the pre-training of ChatGPT's general LLM was fair use, OpenAI's use of copies in ChatGPT's RAG feature is not fair."

Arguments and warnings

According to The Next Web, Wikimedia argues that RAG systems copy passages to answer users' questions and often reproduce substantial content from sources such as Wikipedia and the New York Times. According to the organization, unauthorized copying weakens contributors' motivation and strains servers. The newspaper also reports that Wikimedia relies on the Supreme Court's 2023 Warhol decision to separate the collection of data from its subsequent use.

Wikimedia also warns that an overly restrictive ruling could hamper innovation, which also benefits open source projects like Wikipedia. Conversely, a blanket fair use ruling could make the web more closed and weaken the quality of knowledge online, because there would be fewer secondary sources.

Wikimedia CEO Bernadette Meehan says: "Expanding access to knowledge through AI is a powerful goal, but it must respect the people, contributors, and communities that make reliable knowledge possible." General counsel Stephen LaPorte says: "A ruling that does not apply fair use principles carefully could threaten the sustainability of free knowledge."

These are Wikimedia's own views as an interested party, not a court ruling.