The Bitter Lesson

Names and terms

The Bitter Lesson

A short 2019 essay by AI researcher Rich Sutton. It argues that over 70 years of AI research, general methods that use ever more computation have beaten approaches built on human knowledge, and that this is a hard lesson for researchers to accept.

Rich Sutton

Canadian computer scientist and a founder of reinforcement learning, co-winner of the 2024 Turing Award with Andrew Barto. The video credits its title to his essay of 13 March 2019.

Claude Shannon's 1950 chess paper

In 1950 the mathematician Claude Shannon published 'Programming a Computer for Playing Chess', an early blueprint for chess programs. The video uses it to mark the start of the 'human knowledge plus search' approach.

Deep Blue

IBM's chess computer, which beat world champion Garry Kasparov in a match in May 1997. It combined hand-built chess knowledge with massive search, which is the 'only way' the song says people believed in.

AlphaZero

A DeepMind program unveiled in December 2017 that learned chess, Go and shogi by playing against itself, given only the rules and no human games. It beat the strongest chess engines of the time.

Neural machine translation

In 2016 Google Translate switched from hand-written grammar rules and phrase tables to a neural network trained on large amounts of example sentence pairs. The song uses this as a field where learning beat rules.

Teaching them to see (2012)

In 2012 a deep neural network known as AlexNet won the ImageNet image-recognition contest by a wide margin, ending the dominance of hand-designed visual features. The video marks this as the moment learning beat hand-built vision rules.

AlphaFold

DeepMind's protein structure predictor. Its 2020 version solved much of the protein folding problem, replacing hand-tuned energy functions with a learned model. It appears in the video's list of fields won by learning.