The Bitter Lesson

Rich Sutton’s argument that AI progress ultimately comes from general methods that scale with computation, especially search and learning.

Links

• Read the essay in full: The Bitter Lesson

Overview

The Bitter Lesson observes a recurring pattern in AI: approaches built around human domain knowledge may help in the short term, but are eventually surpassed by simpler general methods that make effective use of growing computation.

Sutton draws this lesson from chess, Go, speech recognition, and computer vision. The practical implication is to favor methods that continue to improve through search, learning, and scale rather than encoding our own understanding of a domain into a system.