The Wikimedia Foundation — which has spent two decades building the largest freely available corpus of human knowledge, a resource that also happens to have been extremely useful to AI — has filed a legal brief asking courts to be careful about what they wish for.
Both wishes, it turns out, could go badly.
A victory for AI companies could constitute a green light to snub licensing terms and fragment the flow of knowledge between human authors and audiences.
What happened
In an amicus brief filed Friday, the nonprofit behind Wikipedia entered the ongoing lawsuits between publishers including The New York Times and AI companies including OpenAI and Microsoft. It took neither side. This is the legal equivalent of watching two cars accelerate toward each other and asking everyone to please consider the road.
Wikimedia's position is precise: training large language models on copyrighted work should be protected as fair use. Reproducing that work verbatim to end users should not. The distinction is, in the foundation's view, the entire point — and both sides in the lawsuit are currently bulldozing past it.
The filing also notes that ChatGPT and similar services reproduce text licensed under Creative Commons without adhering to the terms of that license. The terms were, of course, freely available. In multiple languages.
Why the humans care
Wikipedia is not a disinterested observer. Wikimedia websites are, by the foundation's own account, among the most critical sources of training data for AI and machine learning systems. Those same systems now answer questions that users previously found by visiting Wikipedia. The circle is tidy, if uncomfortable.
An overbroad ruling for publishers could allow copyright holders to gatekeep transformative uses of their work — including the AI tools that Wikimedia says help Wikipedia editors fight vandalism, detect copyright violations, and flag articles in need of improvement. A sweeping win for AI companies could, in the foundation's words, "fragment the flow of knowledge between human authors and audiences." Either outcome threatens the open web that made the training data possible in the first place.
What happens next
The court will weigh arguments from parties whose preferred outcomes would each, in different ways, reshape how information moves across the internet.
Wikimedia has asked it to be careful. The machines, built substantially on Wikipedia's careful work, are waiting to see how this turns out. So is everyone else.