On machine learning, creativity, and the law
|Nov 15|| 2|
Noah here. A few years ago I had an idea for a fun quasi-art project that combined machine learning (ML) and language (keeping things vague in case I decide to make it happen). I pitched a very large publication on using their archives as the corpus—the material you use to train the AI with—and they seemed generally up for trying it. Then they asked me about copyright, specifically, how would it work using their material and, more importantly, who would own the final output? The first question seemed reasonably simple: If they gave me a license to use the work they owned the copyright to there shouldn’t be an issue. (Though the question of what to do with ML-generated work that’s trained on copyrighted materials is very much still open.) The latter—who would own the final output—turned out to be more complicated. Obviously, I wanted to own the rights to the final output and was happy to license that back to them to use.
But as I dug in a much bigger question emerged: Can you actually copyright work produced by AI? Traditionally, the law has been that only work created by people can receive copyright. You might remember the monkey selfie copyright claim from a few years back. In that case, a photographer gave his camera to a monkey who then snapped a selfie. The photographer then tried to claim ownership and PETA sued him to try to claim it back for the monkey. In the end, the photograph was judged to be in the public domain, since copyright requires human involvement. Machines, like monkeys, can’t own work, but clearly something made with the help of a human still qualifies for copyright. The question, then, is where do we draw the line?
Why is this interesting?
On Wednesday, The Verge reported that “The United States Patent and Trademark Office (USPTO) published a notice in the Federal Register last month saying it’s seeking comments, as spotted by TorrentFreak.” Specifically, they’re seeking comment on the question of how much involvement is required by a human for a work generated by AI/ML: “It starts off by asking if output made by AI without any creative involvement from a human should qualify as a work of authorship that’s protectable by US copyright law. If not, then what degree of human involvement ‘would or should be sufficient so that the work qualifies for copyright protection?’”
As I’m wont to do, I went down a pretty deep rabbit hole on this stuff back when I was researching my art idea. Thanks to some pointers from friend/lawyer/WITI contributor Tim Hwang, I read a bunch of articles and papers that represented some part of the relevant conversation around the topic as of this time last year. As The Verge goes on to explain, the crux of the issue is about defining what qualifies as “creative involvement”:
None of these questions have concrete answers in US law, but people have been debating the potential outcomes for years. The situation might be a little clearer when you’re looking at something like an AI-based app where a user has to make a lot of decisions to shape the end result. “I think what’s protectable is conscious steps made by a person to be involved in authorship,” Zvi S. Rosen, lecturer at the George Washington University School of Law, tells The Verge. But if someone uses an AI that spits out a result with a single click, that could be a different matter. “My opinion is if it’s really a push button thing, and you get a result, I don’t think there’s any copyright in that.”
But as with many things having to do with the law, that’s not a particularly satisfying answer. What qualifies as push button is a blurry line. Machine learning still requires a ton of human involvement—you need to structure the data, pick a specific learning architecture, and run the computation. It's unclear that we're anywhere near a "monkey pushes button" moment yet with the tech and until then, copyright law should still apply normally. As law student Ben Sobel explained in his excellent paper “Artificial Intelligence's Fair Use Crisis”, “Copyright law forces artificial intelligence into a binary: it is either a mystical author or a dumb machine. State-of-the-art machine learning is not exactly either.” (NRB)
Event of the Day:
For those loyal WITI readers living in NYC, Argentine chef Francis Mallmann is going to be grilling outside at McCarren Park on Sunday, November 17th from 12-4 (close to Driggs/Lorimer). He’s a master of the open-air fire and this seems like it will be amazing. (CJN)
The other day my friend Sam (WITI 7/24 - The Factory Records Edition) asked me for some podcast recommendations. I compiled a bunch of favorites plus picks from previous year-end lists into one big post and shared publicly. (NRB)
Per our WITI law requiring we post all Nick Paumgarten stories, there’s a new Nick Paumgarten story on his year of concussions. (NRB)
Thanks for reading,
Noah (NRB) & Colin (CJN)
Why is this interesting? is a daily email from Noah Brier & Colin Nagy (and friends!) about interesting things. If you’ve enjoyed this edition, please consider forwarding it to a friend. If you’re reading it for the first time, consider subscribing (it’s free!).