Milwaukee Ex-Pat Writer Makes the Case Against AI Theft

Expand

Photo © Savannah Lauren

Andrea Bartz

Andrea Bartz

Andrea Bartz’s byline has appeared on five best-selling and highly praised novels, a slew of magazine and newspaper articles, and even—way back when she was launching her writing career—pieces in the Shepherd Express.

But her latest credit isn’t for a new novel. Her name appears in the title of a class action lawsuit that has resulted in a landmark $1.5 billion settlement—that’s “billion” with a B—a case that could shape how big tech companies create artificial intelligence tools and AI-generated content that increasingly seeps into our daily lives.

“I couldn’t come up with a good enough reason not to do it,” Bartz told the Shepherd last week. “So I decided to put my name and face on it and to try to fight back.”

A Digital Library of Stolen Books

The case that bears her name, Bartz v Anthropic PBC, involves a claim that the tech company Anthropic used purchased books as well as pirated digital copies of more than 5 million books to train its AI model, Claude, in 2021 and 2022, without crediting or compensating the copyright holders and publishers. Anthropic digitized the purchased books and destroyed the original hard copies. It held onto all of the files—purchased and pirated—to create a library.

Among those millions of books that trained Claude were Bartz’s first novel, the 2019 thriller The Lost Night, and nonfiction books written by her co-plaintiffs, Charles Graeber and Kirk Wallace Johnson.

“I poured so much blood, sweat and tears into the book,” Bartz said. “And then to know it had been stolen for the express purpose of making a big tech company as rich as possible as quickly as possible was just insulting and infuriating.”

What’s Fair?

In June, Judge William Alsup of the U.S. District Court for the Northern District of California ruled on summary judgment that Anthropic’s use of stolen books to train Claude so that it could generate new text was fair use of the works, since the source materials were transformed into something new.

However, the judge found that holding onto the pirated copies in a virtual library likely violated the rights of the copyright holders. A trial date was set for December.

Marquette Law School Associate Professor Bruce Boyden, who specializes in copyright and Internet law, said that the Bartz case and others pending could provide more clarity about the rights of content creators, publishers and copyright holders when emerging technologies require a vast amount of material to “learn.”

“The fact that AI-generated content is something new that we never experienced before does tend to wreak havoc with all of the existing copyright law, which was not decided in an era where machines could basically generate things all on their own without some human supervising that,” Boyden said. “So it does create a lot of issues about exactly how do we interpret things like what is fair use.”

Bartz said the judge’s ruling kicked in settlement talks over how to compensate copyright holders and publishers of the pirated works. She and her co-plaintiffs wrestled with big questions of fairness for writers and publishers.

“We felt a huge responsibility to our class that we were representing, and to getting it right and to looking at it from every different perspective,” Bartz said.

As part of the $1.5 billion settlement approved in late September, Anthropic will pay an estimated $3,000 for each work in the class settlement. (Copyright holders should go to the Bartz v Anthropic Settlement Site to learn about potential claims.)

In a statement, Anthropic’s Deputy General Counsel Aparna Sridhar said, “We remain committed to developing safe AI systems that help people and organizations extend their capabilities, advance scientific discovery, and solve complex problems.”

Bartz called the settlement “imperfect,” but said it was the best result for this case. “We think it’s good, we think it’s fair, we think it’s an important first step in an industry that heretofore has nearly been unregulated and has sort of acted with impunity,” she said.

Who Benefits from AI?

Boyden said similar cases to watch involve movie studios’ attempts to claim rights to their copyrighted images that look substantially like AI-generated ones. But beyond those, Boyden said AI poses larger questions for our society.

“There does seem to be both a fundamental lack of fairness to using authors’ works to basically come up with something that’s going to replace them,” Boyden said.

Bartz acknowledges that AI can have positive impacts—in health care, for example. But she has serious concerns about users’ easy reliance on AI-generated content instead of using their own critical thinking skills and a healthy dose of skepticism.

“Who benefits if all of us sort of offload our thinking to generative AI?” Bartz said. “The big tech companies, not us. I think people should not use it if they don’t need to—and they should put a priority on human-made output.”