Trip Adler, co-founder of Scribd, explains how his platform solves the licensing problem that has stalled AI companies. He breaks down three license types for training AI on copyrighted books, compares the solution to Spotify replacing music piracy, and describes the multi-step clearance process that major publishers and authors need.
Chapters
Napster was also kind of legally gray area, right? And the answer to that was iTunes and Spotify, which is like building out the right legitimate model. So that's what we're trying to do. We're building out the iTunes and the Spotify of this space so that people don't even want to use the Napster anymore. They want to just do the legit version.
Trading a lawfully purchased book is fair use. It's only the pirated copies that were infringements. Patreon CEO called the fair use argument a bogus excuse, right?
First emotion is confusion about what's going on. I feel a little sad that there's so much confusion, so much conflict, because I feel like we should just build out solutions that work for everyone.
Earlier this month, you must know this, Encyclopedia Britannica and its Merriam-Webster subsidiary sued OpenAI for training on their copyrighted articles. And last year, Anthropic settled for $1.5 billion with authors, right? The largest copyright settlement in the AI era. But the court is ruling that training on lawfully purchased books is fair use. It's only the pirated copies that for infringements. But recently, Patreon CEO also called the fair use argument a bogus excuse, right? And you're in— you're sitting in the middle of a lot of all of these. When you see these headlines, I'm curious, like, what is the first emotion that comes to your mind?
Well, first emotion I think is, is it just, uh, I guess a little bit of confusion about what's going on. I also guess I feel a little, little sad that there's so much confusion, so much conflict, because I feel like rather than there being lawsuits and confusion, we should just build out solutions that work for everyone.
As a creator myself, the first thing that comes to my mind was It's just not fair. There has to be a solution at some point, right? If I purchase some books and I tear off the binder and I scan them and then just do training on top of that data, if I'm a creator, I don't think it's fair.
The way we've approached this is we've created the concept of what we call AI rights. We realized the world needed some new set of rights here to kind of make this possible. And we've created 3 categories of AI rights. So the first is training rights. That's the right to use books or other human-created works in model training. The second we call reference rights, and that's the ability to use books in RAG. And then the third is one that we call transformative rights, which we think is really exciting. So when I was, you know, just out of college, I launched the site Scribd to let people upload documents to the web. And I didn't at the time know anything about copyright. And of course what happened is people started uploading pirated books to the website. And then, you know, immediately within days we had all these ceases and desists and threats from the publishing industry. So this is why I really learned about copyright. Look back at like, you know, the Napster era, right? Napster was also kind of legally gray area, right? Um, and the answer to that was iTunes and Spotify, which is like doing the right— building out the right legitimate models, building out the iTunes and the Spotify of this space so that people don't— they don't even want to use the Napster anymore. They want to just do the legit version, and that's what we're building.
Let's say if I'm the indie hacker to take a stab at building an AI rights licensing platform for books, just like how you did it, and here's my high-level thinking I have, and maybe can you tell me what I missed?
Okay.
I'll start with building a database of books with the ISBNs, and then I'm going to cold call all the authors and let authors create 3 license types, like training rights, RAG, transformative. I'll set a price per book, but I admit there's a challenge here, which is I think, I think it's going to be hard for AI companies to—
Are you thinking about starting this company? Are we going to be competitors?
Only in this episode.
Okay. Only this episode. I think, I think what you said sounds right. I mean, yeah, I don't think every AI company is jumping to license these books and these rights, but we have enough traction, I think, to get this going. You know, you just need one deal, which leads the next deal, which leads the next deal, and then it kind of, the snowball starts to go downhill. We have a lot of big authors in our platform. We have really good brands. Also, I've just been working in Silicon Valley for, I think, almost 20 years here, so I have a lot of relationships. I think having just a good brand and having good relationships is kind of the first, kind of like the first kind of step. And then once you get the conversation going, it's just a, it's a matter of understanding what their needs are and how we can help. We're doing particular kinds of deals that are working. You know, we've been really focused on this whole multi-rights holder problem, which is something I think we've uniquely tackled. When it comes to trade books, which are kind of like the books you see at the bookstore, like those types of books, the rights get pretty complicated and there's multiple rights holders. So for a particular book, there's, you know, if you wanna license the AI rights, you typically need to get sign-off from the author, maybe the co-author, the publisher. Sometimes there's multiple publishers involved. There's often a literary agent involved. So there's a lot of rights that holders need to be cleared.
Multi-rights, meaning that all of those parties will get paid?
Yes, that's a large part of the complexity of what we're doing. One of the big reasons these large AI companies didn't license the books for training in the first place is that they just looked at how complicated the rights were. They're like, this would be impossible to clear the rights for, you know, because they all needed a lot of books. They needed like a million books. It was just impossible to clear the rights for the books. And that's really, you know, one of the key problems we've solved is clearing those rights across multiple rights holders.
How do you scale that solution?
We built a platform that allows an author to just sign up. Verify their identity, select the rights they want to license, and then with one click, sign a licensing agreement. That's, that's one, one component of how we solve it.
How many years in was Script that you felt like you figured it out?
With Script, kind of, um, it felt like we were failing and not going anywhere and no one wanted to talk to us. And then we launched the site, we got a lot of traffic. Suddenly, like, we had— my phone was ringing every day with investors trying to invest. But in reality, I think that that was In the grand scheme of things, that was just one small moment. And I think you wanna, you know, you don't wanna let any of these moments be too positive. What matters is just the long-term persistence and chipping away over a long period of time.
So, so if I can compare with learning how to bike, at the longest time when I'm practicing biking, I have no clue. And then once you figure it out, you feel like, oh, I figured it out. The hard part is when you're right before that figuring out moment, how did you, how did you keep yourself and your team motivated? Motivated?
I, I think it's just blind faith. You just gotta believe and go for it. Uh, you just believe in yourself and you just go for it, and eventually it works. There's always marching toward the vision, and sometimes you have days where you like accelerate towards the vision, sometimes you have setbacks, but you don't want any of those moments throw you off too much. You just want to keep, keep marching forward to the vision.
