There are three critical copyright issues that arise with the use of generative AI. In this video, we touch on the legal and ethical complexities around data ingestion for AI training, the potential for AI-generated outputs to infringe on existing works, and the challenge of securing copyright protection for AI-created content.
View Transcript
Jonathan Band
Copyright Attorney
Counsel to the Library Copyright Alliance
Cliff Lynch
Executive Director
The Coalition for Networked Information (CNI)
Gerry Bayne: Imagine a college or university using generative AI to create personalized learning materials for students or streamlining administrative tasks or conducting research. While the potential benefits are immense, these advancements bring significant copyright concerns that impact higher education. Here are three issues to consider with artificial intelligence and copyright.
Jonathan Band: The ingestion of data for purposes of training in LLM is controversial because huge amounts of data, copyrighted data are being ingested, and the people who own those copyrights, some of their attitude is as well, wait a minute, why is my stuff being ingested without my permission? The legal issue is, is that copying an infringement and in the United States, the relevant exception would be fair use. That is the issue in a lot of the ongoing litigation. Is the copying necessary to create these training databases? Is that a fair use or not?
Cliff Lynch: When you talk to people who make their living doing creative work, they are furious. One of the big mistakes that I think the academic legal community is making in general right now is they're doing this very sort of academic analysis about fair use, and they don't understand that there is a significant portion of the population that is absolutely furious.
Jonathan Band: The second issue has to do with, is the output similar to existing copyright works and is it therefore potentially infringing? Clearly, the user who's asking all these prompts would be liable. Kind of the more difficult question is what would be the liability of the service provider? AI companies are going to try to prevent that by filters. This is going to really be an issue with images, which I think will be the easiest one, so all the comic book characters, they have reference shorts so that you can't produce something or you shouldn't be able to produce something that looks like the historic Wonder Woman or Superman or whatever.
Cliff Lynch: You can talk about the potential for these large language models to produce infringing works, and I think there certainly is potential there, but when you really talk to copyright people, there's this kind of messy area about work in the style of, and I'm really very concerned that we're going to see a lot of activity in that, in the style of that is going to be very problematic as we go forward Here.
Jonathan Band: It is a very messy area of law. In general. If things were identical, it's easy. If they're very different, it's easy, but this whole intermediate area of trying to figure out, okay, well, is this an expression or idea? Very, very hard, and it's got to be hard here too. The third copyright issue is can you receive protection for a work that you've created using this generative ai? Imagine that you, with a bunch of prompts, you've created a very nice image, and now can you receive protection for that image? The general rule in the United States and in most countries, except for the United Kingdom, is that you need to have human creativity. All the human creativity can receive protection. How much of it is you and how much of it is the ai? The C office, at least for the time being, is taking the position that even if you ask lots of prompts, what comes out is not protectable, but then after you take that and then you kind of do something with it, then it could be protectable, but it's got to have a certain threshold which has not yet been determined.