At some point soon they're going to be turning AIs loose on the collected scientific archives of history, I'm very curious to see how much long-forgotten and undetected fraud is going to be dug up by them. Four retractions per ten thousand articles seems like an implausibly low average given that humans are involved in writing these things.