I posted a version of this issue on LinkedIn. It struck a nerve. So I’m sharing a new version of it here with some additional thoughts for y’all. Specifically around the central question of… well, what do creators do?
* * *
Meta used all 3 of my books & millions of other books, ebooks, and research papers to train its AI.
All without consent, compensation, copyright concern, credit. You find out only by searching a database published a few days ago by The Atlantic.
It’s sketchy AF. I’ll tell you why in a sec. But first… should we care?
If you’re on the list, you have one of these two reactions:
🥳 Honored! Flattered! My work is worthy!
😳 Ummm WTH. What about consent, compensation, etc.?
And what about non-authors?
🤔 Should anyone care whether Meta used 72 books from David Sedaris? 200 from Margaret Atwood? 19 from Alexander Chee? The entire library of everyone everywhere?
What’s missing in this conversation is that Meta intentionally made ***A Choice***… as the kids say. They actively chose to steal the books, instead of going through the proper and legal channels, because the legal way was too slow.
TOO SLOW.
That’s problematic, isn’t it?
Meta years ago disavowed its famous Move Fast and Break Things motto. But that distancing was cosplay—they pretended to be what they are not under heavy makeup and a mask. Now they are right back at it.
I use AI. I see its value. I am not Anti-AI by any stretch.
But the scale of this is nuts. And the behavior is… gross.
But I’m getting ahead of the story…
* * *
If Meta’s Llama 3 was to compete with the likes of ChatGPT, it needed to be trained on a huge amount of high-quality writing—books, not Instagram captions or LinkedIn posts or what have you. Acquiring all of that text legally could take time.
“Yo ho ho! Should we pirate it instead?” they wondered.
“Abso-tooting-lootly,” they said back to themselves. “Let’s loot all the books!”
Meta felt it was “really important to get books ASAP,” since “books are actually more important than web data,” The Atlantic said.
Books are important. This last line is the only line in this entire fiasco I actually agree with, but in a broken-clocks-are-right-twice-
* * *
Enter LibGen, a pirate library with more than 7.5 million books and 81 million research papers. The Llama 3 team looted and rolled around in all the words like they were pirates rolling in gold doubloons. And they basically were.
So is it problematic? Yes. Here’s why:
🏴☠️ We’re normalizing the theft of IP to an absurd degree. A giant tech company should not feel entitled to scoop up all our words like they’re in a bulk bin at a Dollar Tree & run straight out of the door with them.
🏴☠️ Gen AI often presents like all-knowing oracles, severed from sources. This “decontextualizes knowledge” (The Atlantic‘s phrase). This uncouples the work from the author. It prevents true collaboration. It makes it harder for creators and researchers to build an acknowledged body of work.
🏴☠️ Writing is not colorless, odorless “data.” It’s words that make up sentences that make up paragraphs that make up pages that writers have created, crafted, coaxed into the world. Words with handprints, heart-prints, bitemarks, scratches from us trying to shape them into something that delights you.
Do I sound like I’m being precious about words? About writing? It’s because I am.
* * *
So what’s the alternative though?
Meta and other US tech companies say they need to stay competitive. From the perspective of an AI company, it is an arms race. Or a race for doubloons.
What about China! What about foreign competition! Are the bad choices justified?
Pavel has a point here, one I’ve heard echoed from others:
OK. But choosing between innovation and ethics is a false choice—like choosing between a wooden leg or a hook for a hand. (“Arr… I can’t decide if I want me hand replaced with a hook or me leg with a peg!“)
Let’s get out the whiteboard and workshop some ideas…
1) Legal alternatives to stealing exist! Meta and others could create licensing frameworks similar to how music streaming services pay royalties. Call me crazy… but maybe we just PAY THE WRITERS?!
2) The piracy approach is shaky & unsustainable, because of legal uncertainty and business risk… not to mention bad vibes. A more sustainable approach would involve fair partnerships with creators from the start.
3) If you must loot… loot the works already in the public domain. There’s plenty of great writing that is already in the public domain; more and more gets added every year.
Some amazing titles and authors are new just this year—including Faulkner’s The Sound and the Fury, Virginia Woolf’s A Room of One’s Own, Ernest Hemingway’s A Farewell to Arms, and Tarzan by Edgar Rice Burroughs. Winnie the Pooh is there, too.
4) Commission original works specifically for AI training. Maybe the scale needed makes this one unrealistic…. But we’re whiteboard-brainstorming, remember? No bad ideas at this stage.
5) Ask a marketer: Use weakness as a strength! I’d pay more for ethical AI. Ethical leadership could be a differentiator & competitive advantage. I can see the marketing campaigns already… can’t you?
* * *
Innovation without exploitation. Lunacy, right?
Maybe not. (Let’s put Innovation Without Exploitation on a billboard over the 101!)
Creative innovation doesn’t require taking shortcuts through copyright infringement, short-changing authors and creatives and pissing everyone off.
Tech companies have literally stacks of resources to develop proper licensing models that respect creators’ rights while advancing AI.
Which leads me to my biggest worry about all this: It sets a bad precedent.
The move fast and break things approach sails us into choppy waters. (Pirate flag flapping in the wind.)
If we accept this reasoning for books, what other creative works or personal data become fair game in the name of innovation?
* * *
Let’s remember why we’re building these models.
Is it just about technological advancement? If so, to what end? Is it just about doubling our doubloons over and over again…?
Or is it more than that? Is it around advancing society and culture?
If authors can’t make a living from their work because AI systems trained on their stolen content generate competing material, we risk undermining the very creative ecosystem that AI aims to enhance.
How do writers fight back?
Some authors whose work was stolen—including Sarah Silverman and Michael Chabon—had filed separate copyright lawsuits that were then rolled into one: Kadrey v. Meta is a class action suit in Northern California.
Last week, the Authors Guild assured writers that if our work was used by Meta to train their AI, you’re automatically included in the Kadrey v. Meta class action case. See more thoughts from the Authors Guild here.
* * *
Clarity, consent, compensation. Another billboard, maybe.
Is it too much to ask that AI companies be transparent about data sources and use, obtain permission from creators, and provide fair payment for using their works?
I can’t imagine why anyone would disagree.
Here’s the link to the pirated-books database.
And here’s a link to pirated movie and television writing.