Wikimedia proposes a compromise in the AI ​​copyright case: training can be considered fair use, and reproduction of the original text should be authorized

Abstract:

The Wikimedia Foundation, the operator of Wikipedia, submitted an amicus brief in an AI copyright lawsuit involving companies such as The New York Times, OpenAI, and Microsoft, advocating that the court should not confuse artificial intelligence training with chatbots for reproducing content to users. The foundation did not support either party's overall position, but warned that an overly simplistic ruling could harm both open knowledge and the Internet content ecosystem.

Wikimedia believes that website scraping itself should not be considered copyright infringement; copyrighted works are used for pre-training of large-scale language models without permission. If the model output is "different" from the original text, the fair use principle can be applied. However, if the chatbot directly outputs article content and replaces readers’ access to the original work, the AI ​​company should comply with licensing requirements.

This claim is related to the Creative Commons Attribution-ShareAlike license adopted by Wikipedia. Wikimedia said the model reproduces Wikipedia text released under an open license, but may not comply with conditions such as attribution and share-alike. The foundation is worried that if the court broadly determines that AI training is unrestricted, it may give a green light to ignore permissions and weaken the flow of content between authors and readers; conversely, if copyright owners obtain too broad control rights, it may also hinder innovative and transformative uses of works.

Wikimedia also pointed out that AI tools not only bring risks, but also help Wikipedia editors identify destructive modifications, copyright violations and human flesh information, and identify items that need improvement; however, chatbots directly providing information may also prevent users from visiting Wikipedia and other original websites, weakening open knowledge projects that rely on community maintenance.

The New York Times sued OpenAI and Microsoft in 2023, accusing the two companies of grabbing a large number of news articles without permission and using them to train AI systems; the two companies claimed that the relevant training was fair use. Wikimedia warned that if the court fails to carefully weigh the differences between training and generated content, the ruling may affect future AI development, open networks, and the way users obtain information.