Over the holiday break, the New York Times sued OpenAI and Microsoft for copyright infringement. The lawsuit covers both using New York Times content for training, for reproducing the content in response to prompts.
The New York Times may not be “scholarly,” but the suit could be a preview of how large scholarly publishers deal with OpenAI. First, it is fair to call both the Times and scholarly journals high quality content, the kind that OpenAI likely prefers for training its model (Complaint, p. 29). Second, there are unauthorized copies of much of the content online, so it would be possible to initially train a model on the content without permission. Finally, there is the financial angle. This lawsuit comes after negotiations between the companies to have them pay for the New York Times’ content. While some publishers are exploring ways to use AI with their own content, they may find it profitable to license that content to OpenAI and other companies.
One other interesting note here is how Microsoft is brought into the lawsuit from several different angles. First, it is a big investor in OpenAI. Second, it offers products based on OpenAI’s models, in particular anything branded “Copilot,” and Bing Chat. It is also being accused of helping OpenAI make copies of content in training ChatGPT, or at least overlooking the copying OpenAI was allegedly doing. But the most interesting claim that could have far reaching implications if a court agrees is that Microsoft is committing copyright infringement by “storing, processing, and reproducing” the models on its platform. (Complaint, p. 60). That being copyright infringement could greatly chill AI research, as a researcher would need to know the provenance of a model, and every document used in its training, to be safe from a copyright claim.
Given that this lawsuit is following negotiations over a license agreement, it would not be surprising if this settles before trial. The New York Times may be well-resourced for a big legal fight, but there are no guarantees they would win, risking a lot of licensing revenue. At some point there will be a copyright suit regarding AI that goes to trial (no guess as to which, as it can take a long time to go from filing a case to a trial), but maybe not this one.