The Illusion of Theory in Large Programs: LLMs and the Limits of Code Comprehension

Peter Naur's paper argues that the 'theory' of a large program isn't the code itself, but rather the deep understanding possessed by the programmers who built it. Using Ryle's analogy of a path, the code is the path, while understanding the program is like walking it. While LLMs can generate text that appears theoretically informed, this is merely surface-level; it lacks true comprehension. Naur emphasizes that understanding large programs requires continuous involvement in their development and maintenance—a knowledge that cannot be gained simply by reading code or documentation.