Should we pay artists for their contributions if we used their work to train ChatGPT or other generative AI systems? When questioned on the main stage of SXSW this afternoon, Peter Deng, vice president of consumer products at OpenAI. The company that makes ChatGPT, was reluctant to respond.

When Josh Constine, a former writer and venture partner at SignalFire, posed the issue to Deng at a comprehensive fireside chat, he said. “That’s a great question.” Deng recognized some of the “yes” shouts from the throng of observers. “The audience is telling me that they do. I’m getting word from the people that watch them.

That Deng sidestepped the inquiry is hardly shocking. When it comes to how it uses data to train generative AI systems, such as the ChatGPT-integrated DALL-E 3 art-creation tool. OpenAI is in a precarious legal position. Large datasets from public sources, such as images and artwork, are used to train DALL-E 3 and related algorithms. OpenAI claims that using copyrighted information for training is justified by fair use, claiming transformation. They are able to use public data and scrape it without giving acknowledgment or payment to the original artists because of their legal position.

OpenAI contends it’s essential to utilize copyrighted content for AI development. They assert training models with publicly accessible internet content constitutes fair use, citing legal precedents. The company emphasizes this approach is fair to creators, vital for innovation, and crucial for U.S. competitiveness.

Naturally, creators don’t agree

Artists, including Grzegorz Rutkowski of Dungeons & Dragons fame, filed a lawsuit against OpenAI. Midjourney, and DeviantArt, alleging replication of their styles without consent. OpenAI offers licensing agreements with some providers and allows artists to opt out of dataset use, though some find the process burdensome.

Though he is unsure of the precise form that this would take, Deng stated that he thinks artists should have more agency in the development and application of generative AI tools such as DALL-E.

“Artists must integrate themselves as much as possible into the ecosystem,” Deng stated. “I think that we can really help the business out a bit more if we can figure out a way to make the flywheel of making art faster. Every artist has, in a way, been influenced by the artists who have gone before them; I wonder how much of that this will hasten.