AI in science

The tradeoff between interpretability and predictive power is well known. Here, I argue that the sort of generative AI models currently gaining traction in the natural sciences are bound to make that tradeoff more severe.