Why AI Expert Richard Sutton Doubts Synthetic Data

Richard Sutton believes the real world is far too complex to capture in a simulation. He argues that synthetic data is just a tiny, limited copy of reality. Relying on it ignores the vast amount of information we have yet to discover.

He thinks human expertise is currently a bottleneck for progress. When we feed AI fake data, we limit its ability to learn beyond what we already know. This prevents models from reaching their true potential.

Instead, Sutton wants AI to act like real agents that learn from experience. He suggests that machines should interact with the world to grow their knowledge. This approach would allow them to evolve without relying on static, pre-recorded data.