OpenAI Attorneys Say Licensing All Training Data Is Impossible

Obtaining consent for all copyrighted content used to train AI models would be impossible given the enormous volume of data they consume, OpenAI Inc.’s top intellectual property attorneys said Thursday.

“If you were to do private negotiations for every piece of content that you need to train one of these models,” they wouldn’t exist, said Sy Damle, who represents OpenAI in copyright litigation, at the Vanderbilt University Music Law Summit in Nashville, Tenn. “We just have to accept that as a fact in the world, and then we can start having a conversation about what are the right policies that ...