The Wikimedia Foundation has filed an amicus brief in the context of major lawsuits involving The New York Times and AI companies such as OpenAI and Microsoft. Through this move, the foundation aims to find a "middle ground" that protects technical innovation while ensuring the sustainability of open knowledge, warning that extreme judicial rulings could harm both parties.
Distinguishing Between Training and Reproduction
The foundation proposes a precise standard to distinguish between two types of data usage in AI models:
- Training: Wikipedia believes that using copyrighted content to train Large Language Models (LLMs) should fall under "fair use," provided that these models offer something new and different from the original sources.
- Regurgitation: The foundation warns that chatbots that copy content and present it to users instead of directing them to the original sources infringe on copyright and replace human interaction with the source, which requires obtaining prior licenses.
Why Does This News Matter?
This position represents an attempt to save the information ecosystem from two dangerous scenarios that current judicial rulings could produce:
- The First Scenario: A sweeping victory for publishers could allow them to impose strict gatekeeping that prevents the use of data in transformative innovations, hindering the development of open-source technologies.
- The Second Scenario: An absolute victory for AI companies could lead to the disregard of license terms, threatening the flow of knowledge between authors and the public.
For Wikipedia, the matter is existential; the site is a vital source of training data, but it also suffers from traffic diversion via AI tools that provide direct answers without visiting the site, in addition to some services violating "Creative Commons" licenses by failing to attribute the original sources.
Balancing Benefits and Risks
The foundation noted that AI is not necessarily an enemy; rather, it can be a helpful tool for Wikipedia editors, such as detecting vandalism, monitoring copyright infringements, and identifying articles that need improvement. However, the foundation emphasized that any expansion in access to knowledge via AI must respect the communities and contributors who make this reliable knowledge possible, asserting that the balance between property rights and innovation is a shared responsibility for all parties in the information ecosystem.