Wikimedia seeks middle ground in AI copyright lawsuit

Summary

The Wikimedia Foundation filed an amicus curiae brief in the copyright lawsuit filed by The New York Times against OpenAI and Microsoft. The Foundation argues that pre-training large language models (LLMs) on copyrighted materials should be considered fair use. However, it contends that AI systems which present this content to users in a way that substitutes for the original must obtain a license.

Why it matters

Wikimedia is caught in a unique contradiction: its content serves as the foundational training corpus for large AI systems, yet these systems are siphoning traffic away from the volunteer communities that sustain that content. A court ruling could either grant copyright holders the power to control transformative AI usage or greenlight unlicensed content copying. This outcome will directly impact the open-source information infrastructure worldwide.

Highlights

• Wikimedia draws a clear line: training is fair use, but reproducing content in a way that replaces the original is not.
• The Foundation claims that ChatGPT and similar services violate Creative Commons terms by reproducing Wikipedia text without proper attribution.
• Wikimedia warns that a comprehensive victory for publishers could give copyright holders the ability to control even innovative and transformative AI applications.

Critical take

Wikimedia's 'middle-ground' strategy is pragmatically appealing: it wants its content freely used for training while demanding a license when AI systems present that content to users. However, this distinction may not withstand careful legal scrutiny. The brief also assumes that the copyright protections sought on behalf of publishers should apply to Wikipedia's Creative Commons-licensed content. The Foundation fails to fully address this tension.