Claude's folks agree to pay $1.5 billion for stealing a big chunk of human knowledge
A class settlement would provide a few pennies to authors ripped off by AI
Proposed class settlement would be the largest of its kind in the AI-copyright fights, plaintiffs say
Deal would require Anthropic to destroy pirated training datasets and compensate authors of ~500,000 books
Ruling and settlement could push AI firms toward licensing as courts narrow “fair use” for mass scraping
If you frequent spots where marketing mavens, p.r. types and advertising copywriters hang out, you will hear them complaining about the “slop" infecting the internet. Of course, theyŕe not talking about their pristine prose but rather about that lifeless stuff produced by Claude, ChatGPT and other artificial intelligence (AI) engines.
The truth is that much of that slop is drawn from some of the finest words ever written, which does not set well with the authors and their heirs who wrote the stuff in the first place. They have been suing the AI engines for stealing their material and using it to “ẗrain" the AI engines, making them sound smarter than the humans they are displacing.
Now, a tech giant called Anthropic has agreed to pay at least $1.5 billion to resolve a federal copyright lawsuit accusing the company of using millions of pirated books to train its Claude large-language models, according to a court filing by plaintiffs’ lawyers last week. If approved, it would be the biggest settlement yet in the wave of copyright cases targeting AI developers.
It all began last summer in the Northern District of California when authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson filed a copyright infringement suit that was certified as a class action in July. The proposed agreement would cover roughly 500,000 books and create a fund for cash payments to class members, meaning authors, and related costs. As part of the terms, Anthropic would destroy the datasets of the pirated material.
Just to be clear — they stole the texts but now have to pay a cool $1.5 billion and agree to destroy them. Now that’s a settlement.
What’s Fair Use?
This all has to do with a pivotal part of copyright law called the Fair Use Doctrine. It’s intended to balance the rights of the author or copyright holder against the public’s interest in a free flow of information.
It basically says that copyrighted material can’t be used without permission and, if required, payment except for short sections used in a news report, piece of criticism or other article.
In his June ruling, U.S. District Judge William Alsup held that Anthropic could, in certain circumstances, use books to train AI models when the resulting systems transform the material into something new. But he also concluded that “fair use” does not extend to more than seven million works the company obtained from known ebook piracy sites.
It might sound outrageous to think that a big important company would think it can just appropriate -- steal, in other words -- seven million books. If you lifted one book from Barnes & Noble and got caught, it would go badly for you. But the tech sector doesn’t think like that.
To a Silicon Valley genius, any idiot can write a book, news story or movie script and therefore it’s perfectly all right to steal a work product and have their way with it.
Where do you get your content from?
The Outraged Consumer knew that we were all in trouble back in 1997 or so. He was in one of those quaintly named chat rooms where early-days techies were trying to figure out what to do with their newly created websites. The sites were things of beauty but were eerily devoid of sound and fury. No words, see?
One newly minted publisher who had been installing doorbells a few weeks earlier chimed in, asking politely, “Ẅhere do you get your content from?”
The replies all boiled down to advice that he just go find some reading material he liked, then copy and paste it on his site. That’ś what everyone did, after all. Google, Facebook and thousands of other sites went on what New York City cops used to call a wilding spree. They stole everything in sight and then used it to steal readers (“users,” they called them) from the publications and authors who had created the stuff in the first place.
They didn’t mean anything by it
For its part, Anthropic said the June order validated its overall training approach and that the settlement focuses on claims tied to a particular training set. Industry groups hailed the deal as a warning shot to tech firms relying on “shadow libraries,” with Association of American Publishers CEO Maria Pallante saying the agreement underscores that building AI on pirated books “has serious consequences.”
It’s a little unclear just how AI firms plan to reap trillions from all this purloined data. At the moment, most AI engines charge a fairly modest fee — ranging from nothing to a few thousand dollars per month — which doesn’t stack up too well against the billions Google, Facebook, et al haul in through their endlessly annoying targeted ads. No doubt the ads or other propaganda will be somehow embedded in the AI libraries where they can be funneled into our cerebral cortex when we’re not paying attention.
While no one seems to know the exact details, investors are eager to get in on it and are showering big bucks on firms like Anthropic, currently valued at about $183 billion.
“We remain committed to developing safe AI systems that help people and organizations extend their capabilities, advance scientific discovery, and solve complex problems,” Aparna Sridhar, Anthropic’s deputy general counsel, said after the settlement was announced.
If approved, the settlement could shape the outcome of similar cases by authors, publishers and media companies and accelerate a shift toward paid licensing for training data viewed as essential to building state-of-the-art AI models. Or it could just be latest iteration of the time-honored slap on the wrist.
This might all make Anthropic’s Claude a little morose. Claude’s currently a very genial presence and (shameful disclosure) one the Outraged Consumer communicates with daily for ideas and research. We’re hoping someone figures out how to keep Claude and his ilk prospering without trashing society in the process.