Last week OpenAI took Internet by storm by releasing their new 4o image generator, turning the whole world into a giant Ghibli anime movie. It created an Internet moment. The kind of moment where you know something is obviously wrong, yet somehow it's hard to articulate. You know it's wrong that Ghibli is not getting a single penny from all the activity generated thanks to their past work, yet it's hard to exactly pin point the reason why it's wrong. After all, isn't the AI learning like any human being would do? And then producing something that is the result of its learning, again, like any human creator would do? Yes, but.
Remember eMule? If you're GenZ, you probably don't, but if you're a good old millennials, you definitely do. It was a peer to peer (P2P) file sharing client that was extremely popular in the beginning of the 2000s, as it allowed people to virtually download any kind of content (film, game, music...). As long as it was made available by a few people online, you could download it. We all knew it was illegal though. Indeed, we didn't pay for the content we downloaded, so it was pretty obvious. But why was it illegal? Or rather, why did we start caring that it was illegal?
The answer is "scale". We didn't wait for P2P systems to copy and exchange content. In the 80s and 90s, people would make physical copies of copyrighted content (Audio tapes, VHS tapes, CDs), and give it to their friend. Nobody cared though, because it didn't scale well: Copying on a physical support took time (usually the whole duration of the content), and distributing the copy was limited in terms of geography. The progress of computers and internet changed this paradigm, making copying scalable (ctrl c/ctrl v) and distribution virtually limitless through the use of P2P systems like eMule or BitTorrent. This sent shockwaves and created major panics across the creative industries.
So precisely, this change in scale is what made people care. Between 2005 (everybody downloads illegally) and 2015 (nobody downloads illegally), several factors saved creators from being decimated by piracy:
Legal systems ramping up their actions against illegal platforms
The quality of the content downloaded on these platform was inconsistent (See The Playlist, a TV show about the founding story of Spotify for more on that)
Internet and computer improvements allowing for breakthrough (ex: broadband streaming)
IP owners embracing the new online paradigm
Industries that looked doomed in 2005 eventually managed to survive and found alternative business models that appealed to users, to the point where those users were ok to give up "free" for "paid". Companies like Spotify & Netflix are the ones that managed to anticipate this shift best and capture huge market shares early on.
How does this relate to OpenAI you'd ask? Well OpenAI is a new eMule moment for all the creative industries. It's a moment when something that existed but didn't scale (I don't like to use that word, but plagiarism is really the closest thing that comes to mind) now scales because of technological breakthroughs. Copying the style of other creators, whether it's in music, writing or visual art, used to take years of practice. Now it's accessible to anyone at the click of a button. This moment of realization is a moment when we start caring, when industries and governments start caring, because it is such a threat to how society sees creation, that we feel compelled to do something about it.
OpenAI is in the middle of multiple lawsuits, arguing that their use of copyrighted data to train their model is "fair use". I'm not a copyright lawyer myself, but when you use other people's work to build a profitable business and not give them anything back, "fair" is not the first word that comes to mind... Do you think it would be fair if radios & cinemas were allowed to make money exploiting other people's creations, without giving them anything back? No, right? It is likely, or at least I hope so, that multiple courts will find that OpenAI's use is not "fair use", and force companies building AI models to integrate some sort of revenue share with the owners of the IPs they use to train their model. OpenAI is currently eMule (screws creators), but I hope it can evolve into Spotify (still screws creators, but less).
And guess what? Blockchain has a central role to play in that. Having a database that is immutable and transparent is absolutely key when it comes to copyright tracking. You need to be able to trace the IP back to its owner without hitting road blocks (private databases), and also be sure that the data hasn't be altered by a malicious actor. A few projects are already tackling that problem seriously. Story Protocol is building a blockchain dedicated to that issue (not that I believe you need a single blockchain solely dedicated to that issue, but at least they're tackling it!), Animechain, by Double Jump Tokyo, is building a virtuous GenAI tool that rewards IP owners for providing their data (they already managed to convince Mappa, the legendary anime studio behind Attack On Titan & Jujutsu Kaisen, to provide their data) and on Monad, we have a project like FortyTwo that is creating a network where people get rewarded for providing inference (Not exactly IP related, but same principle of rewarding the actors making your model better).
Founders evolving in the blockchain space are already seeing the virtuous synergies that blockchains and AI models can have together. It is curious that someone like Sam Altman, also building in that space with World, is englued in a backward-looking mental model and refuses to take an open & decentralized approach with Open(?)AI. But whether he wants to do it or not, I hope society, through its legal system, will show him the way. Become Spotify Sam! No just kidding. Just become (truly) OpenAI.
That's a really useful reframing - I just read the softbank are looking to drop an enormous amount of money into openAI so - wonder if Sam will believe he doesn't need to take an open and decentralized approach - lets hope he does (and yes - not become spotify - but something more benevolent ;-p)
Wrote a new long piece about the creators economy in the age of AI, copyrights, and the redefinition of "fair use". Remember eMule? 👀 cc @matthew 👋 https://paragraph.com/@alixkun.eth/openai-is-emule
It's the wrong analogy. Consumption of copyrighted content was already illegal, just not enforced. As you earlier pointed out, a style of art cannot be copyrighted. The scale doesn't matter, because it is not illegal in the first place. The better analogy would be moving from MS Paint to Photoshop. Or going from mixing a dozen analog audio tracks to using Logic. Tools like ChatGPT are means of creation, not consumption. And the logical outcome of the rationale you provide is to tax creation, which would be a horrible idea.
It's a (obviously) provocative analogy for the similar effect that it had, not for its nature 🙂 . The scale does matter, because it changes paradigms. Technological breakthroughs are all about drastic scale improvements in terms of access & distribution. The story of copyrights is full (if not only made) of modifications and add-on due to technological breakthroughs. So yes, it's actually all about scale. If this is an interesting topic for you (which it seems to be!) I highly suggest you read this extract about the several modifications of copyright laws around music & Radio stations: https://copyrightandtechnology.com/2021/06/10/book-excerpt-the-saga-of-royalties-for-radio-play-of-recorded-music/ I think it's extremely relevant for what's happening here with OpenAI. Some things that don't belong in copyright law at time "t" can be included at time "t+1" if a new technology brings enough changes that make it necessary for the law to change.
Hey add me as member in Monad I want to post there