Turing Award Winner Rich Sutton Warns: Relying on Synthetic Data Is a Big Mistake

Rich Sutton, a Canadian computer scientist and Turing Award winner, said on a Sequoia Capital podcast that the AI industry's growing reliance on synthetic training data might be leading it in the wrong direction. He believes that artificially created data can't fully replace real data and called it "a big mistake."

Synthetic data is generated by algorithms or AI models, not collected from the real world. While it's used in areas like self-driving cars and fraud detection, Sutton stressed that real experiential data—like AI agents learning by interacting with their environment—is more important. He gave an example: you can't predict how a drone will behave in the real world using synthetic data, because the real world is full of complex variables.

Tech giants are also hunting for more real data. OpenAI has been looking for large, proprietary datasets, and Google recently paid $10 million to buy internal data from bankrupt Spirit Airlines, showing the growing value of real data. Last month, Sutton co-founded Oak Lab with his former student to develop AI agents that learn from their own experiences. The startup hasn't announced funding details yet.