Should artists whose work was used to coach generative AI like ChatGPT be compensated for his or her contributions? Peter Deng, vp of consumer products at OpenAI – the maker of ChatGPT – was reluctant to offer a solution when asked on the important stage at SXSW this afternoon.
“That’s an important query,” he said when asked by SignalFire enterprise partner (and former TechCrunch author) Josh Constine, who interviewed Deng in an in depth Fireside interview. Some in the group of onlookers shouted “yes” in response, which Deng took note of. “I hear from the audience that they do. I hear from the audience they do.”
That Deng dodged the query just isn’t surprising. OpenAI finds itself in a tough legal position relating to the best way it uses data to coach generative AI systems just like the art creation tool DALL-E 3, which is integrated with ChatGPT.
Systems like DALL-E 3 are trained on an infinite variety of examples—artworks, illustrations, photos, etc.—typically sourced from public web sites and datasets on the Internet. OpenAI and other generative AI providers argue that fair use, the legal doctrine that permits using copyrighted works to create a secondary creation so long as it’s transformative, violates their practice of scraping public data and using it for training without compensation or Even using recognition shields artists.
In fact, OpenAI recently argued that without copyrighted material it will be not possible to create useful AI models. “Training AI models using publicly available Internet materials is fair use, as confirmed by long-standing and widely accepted precedents,” the corporate wrote in a January statement blog entry. “We view this principle as fair to creators, crucial for innovators, and important to U.S. competitiveness.”
Unsurprisingly, the creators disagree.
A category motion lawsuit by artists including Grzegorz Rutkowski, known for his work on Dungeons & Dragons and Magic: The Gathering, against OpenAI and several other of its competitors (Midjourney and DeviantArt) is heading to court. The defendants argue that tools like DALL-E 3 and Midjourney reproduce artists' styles without their express permission, allowing users to create recent works just like the artists' originals for which the artists receive no payment.
OpenAI has licensing agreements with some content providers reminiscent of Shutterstock, allowing webmasters to dam its web crawler from crawling their site for training data. Additionally, OpenAI, like a few of its competitors, allows artists to remove their work from the datasets the corporate uses to coach its image-generating models. (Some artists have described However, the opt-out tool, which requires submitting a single copy of every image to be removed together with an outline, is cumbersome.)
Deng said he believes artists would have more freedom to develop and use generative AI tools like DALL-E, but wasn't sure exactly what which may seem like.
“Artists must be a part of the ecosystem as much as possible,” Deng said. “I feel if we will discover a method to speed up the flywheel of art creation, we'll really help the industry a bit bit more… In some ways, every artist has been inspired by artists who got here before them, and I ponder how much that’s thereby accelerated.”
