CONCEPT
Prediction Without Comprehension
The discovery—made concrete by machine learning—that the capacity to forecast the world accurately can be entirely divorced from any understanding of why the world behaves as it does.
For three centuries after Newton, prediction and comprehension were taken to be the same act: you could reliably foretell a phenomenon only because you grasped the mechanism generating it, and successful prediction was the proof of genuine understanding.
Large language models have falsified this assumption in the most concrete possible way. A weather model built on deep learning outpredicts physics-based simulations without containing any physics; a protein-folding system predicts structures without modeling the forces that fold them; a language model predicts the next word in a legal argument without comprehending law. These systems achieve prediction through the statistical structure of data alone, with no detour through mechanism—and the prediction is real, reliable, and actionable. The concept names the fracture Newton never had to face: a world where we can foresee without understanding, where the correlation is rich enough to make the law unnecessary for many practical purposes, and where the absence of comprehension only surfaces when the world departs from the training distribution and the system fails,