Should artists whose work is used to train generative AIs like ChatGPT be compensated for their contributions? Peter Deng, vice president of consumer products at OpenAI, the developer of ChatGPT. When asked this afternoon on the main stage at SXSW, he didn't want to answer.
“That's a great question,” said SignalFire venture partner (and former TechCrunch writer) Josh Constine, who interviewed Deng Xiaoping at a wide-ranging fireside. Some onlookers shouted, “Yes,” and Deng Xiaoping agreed. “I hear from the audience that they do that. I hear from the audience that they do that.”
It is not surprising that Deng Xiaoping dodged the question. OpenAI is in a delicate legal position surrounding how it uses data to train generative AI systems like DALL-E 3, the art creation tool built into ChatGPT.
Systems like DALL-E 3 are typically trained on a large number of examples, such as artwork, illustrations, and photographs, sourced from public sites and data sets on the web. OpenAI and other generative AI vendors are aware that fair use, a legal doctrine that allows the use of copyrighted works to create derivative works as long as they are transformative, makes public data available without compensation or credit. They claim to be protecting their practice of scraping and using it for training. artists.
In fact, OpenAI recently claimed that it is impossible to create useful AI models without copyrighted material. “Training AI models using publicly available internet materials is fair use, as supported by long-standing and widely accepted precedent,” the company said in a January blog post. ” he wrote. “We believe these principles are fair to creators, necessary for innovators, and important to America’s competitiveness.”
Naturally, the creators do not agree with this.
A class action lawsuit filed by artists including Grzegorz Rutkowski, known for his work on Dungeons & Dragons and Magic: The Gathering, against several of OpenAI's rivals, Midjourney and DeviantArt. is being tried in court. Defendants claim that tools like DALL-E 3 and Midjourney reproduce an artist's style without the artist's express permission, and that users can create new works similar to the artist's original without the artist receiving compensation. It is claimed that it can be generated without any
OpenAI has licensing agreements with some content providers, such as Shutterstock, that allow webmasters to block their web crawlers from scraping their sites for training data. Additionally, like some rivals, OpenAI allows artists to “opt out” and have their work removed from the datasets the company uses to train its image generation models. (However, some artists have found the opt-out tool, which requires them to submit a separate copy of each image they want removed along with an explanation, to be cumbersome.)
Deng said he believes artists should have more control over the creation and use of generative AI tools like DALL-E, but he doesn't know exactly what form that will take. Ta.
“[A]Artists must participate [the] We need to protect the ecosystem as much as possible,” Deng Xiaoping said. “I believe that if we can find ways to make the flywheel of art production faster, we can help the industry a little bit more…In some sense, every artist has been inspired by those who came before them. I wonder how much this will accelerate it.”