Can Sarah Silverman’s AI Lawsuit Save Us From Robot Overlords?

When Sarah Silverman sued artificial intelligence titans OpenAI and Meta Platforms on July 7, her copyright lawsuit appeared to present a relatively straightforward claim: These companies did not secure Silverman’s and other authors’ permission before using their copyrighted works, including her autobiography from 2010 The bed ram, which is not OK, according to these suits. Silverman is joined by two other writers, novelists Christopher Golden and Richard Kadrey, in these suits; their civil complaints are seeking class-action status, which, if the court gives the green light, means many, many more writers could take action against these companies.

Indeed, OpenAI’s ChatGPT and Meta’s artificial intelligence projects rely on mass trawling of books to learn language and generate text, the suits say. Silverman’s suit alleges that these AI projects did not secure her and other authors’ permission to use their works before inhaling them, violating intellectual property rights. They also allege that these AI systems gained access to these books through fraudulent means, using libraries of pirated texts — or as the cases’ co-counsel Matthew Butterick puts it to Vulture, “Creators’ work has been vacuumed up by these companies without consent, without credit, without compensation, and it is not legal.”

Silverman claims that ChatGPT and Meta’s generation of text is the very receipt that proves they consumed them. If they can spit out summaries The bed ram and other copyrighted works, her suit argues, then those systems must have used stolen books to do it. The proposed class-action lawsuit asks for monetary damages as well as “permanent injunctions” to prevent these AI systems from devouring their work — and then using it to create text — without permission or payment.

While these are copyright cases, they may provide an opportunity to learn more about the shadowy world of AI as lawsuits unfold, and tech-wary catastrophists like this writer may wonder if their outcome might actually affect AI’s operation. If a judge or jury decides that ChatGPT cannot consume copyrighted material with abandon, will that ruling potentially limit what AI can do? Another way: Could a lawsuit be over The bed ram thwart a Skynet-like situation? Vulture spoke with Butterick, as well as two experts in law and artificial intelligence, to learn more about what this lawsuit can and can’t do. Neither OpenAI nor Meta immediately responded to requests for comment.

Butterick, who along with attorney Joseph Saveri is leading the cases, said the core problem is that AI doesn’t just come up with things that happen to sound like The bed ram or other books – it is entirely dependent on people’s creations. “These artificial intelligence systems, it’s kind of an ironic name because they’re built entirely and entirely on the work of human creators. All of these generative AI systems rely on consuming massive amounts of human creative work, whether it’s text for these language models, or whether it’s images for these AI image generators,” he says. “That’s how these neural networks work: They take in the training material, and what they do is they try to mimic it. When we talk about artificial intelligence, we must understand where it actually comes from: It is human intelligence, [but] it has just been separated from the creators.” The lawsuits also allege the books were assembled from sketchy online sources that didn’t have the green light to have them in the first place.

“I see this as an existential question for creators,” says Jacqueline Charlesworth, who represented publishers in a lawsuit against the Internet Archive’s library-like book lending system. A judge ruled in March that the Archive’s book sharing setup was in violation of copyright law. “What’s going on right now is that AI suddenly entered popular culture and tools were made available to everyone, basically, and it seemed like an overnight explosion. Although we know that many of these models were developed over time, [they] exploded.” There’s also the question of whether humans have the right not to be used by AI, whether they’re authors or humans in general. “People really should have the right to opt out of their works, their data, being used in these models, Charlesworth argues.

One of the biggest concerns about AI is the secrecy around how platforms like ChatGPT work these days. It is becoming more and more intricate in our daily lives, which means that a lack of detail about this system can prevent neutral parties from seeing problems or potential dangers, experts said. Fast company in March. For example, AI is known to reflect bad biases—such as racial prejudice—that we see among humans. Without knowing exactly what’s going on with how it learns or picks up potential biases, it’s hard to sort that out in AI, according to University of Michigan-Dearborn professor Samir Rawashdeh.

The discovery phase of Silverman’s lawsuits could potentially shed light on how these systems work. “A discourse that has been promoted and pushed by the AI ​​companies themselves is that these systems are essentially magical black boxes and they learn like a human, and there are all these metaphors thrown around in essentially keeping people from investigating how they work,” says Butterick. “And in doing so, they are trying to insulate themselves from any kind of legal investigation. That’s part of what these cases are about: Let’s open the black box. Let’s see what’s inside.”

Charlesworth expresses similar sentiments about AI procedures. “We’re going to learn a lot more about exactly how the models work and what the training data is,” she notes. “There’s not a lot of transparency there, and I think, especially if your model is already based on pirated books, that’s a big red flag. You’re copying books without permission. It’s offensive.”

It is impossible to predict the outcome of a lawsuit. But there is little doubt that the case will be a slam dunk for authors because of a landmark case involving Google Books nearly a decade ago. The US Supreme Court ruled in 2016 that Google Books’ practice of summarizing texts — and showing excerpts to users — did not violate copyright law, according to the Associated Press. Deven Desai, a professor of business law and ethics at the Georgia Institute of Technology, says the law currently allows the use of books to train software. Desai notes that the Google Books case resulted in “the ability to use books in transformative ways, including creating excerpts and training software in that sense, for machine learning,” so that machines can use books to learn under the law.

As for a copyright lawsuit averting an AI revolution? “It’s not really about GPT systems taking over [the world], but about whether they have to pay for their training data.” If OpenAI didn’t buy copies of the books, they probably should have. Maybe the pen won’t be mightier than the sword in a robot war after all.