"A computer can be programmed so that it will learn to play a better game of checkers than can be played by the person who wrote the program." — Arthur Samuel, 1959
Arthur Samuel (1901–1990) was an IBM researcher who coined the term "machine learning" in his landmark 1959 paper. His checkers program, running on the IBM 701, was the world's first self-learning AI — it improved by playing against itself, using two internal agents he called Alpha and Beta. If Alpha beat Beta, Alpha's strategy became the new standard.
This is the direct ancestor of AlphaGo's self-play training.
This implementation uses Samuel's actual techniques:
- Alpha-Beta pruning (minimax with cutoffs — Samuel invented this for checkers)
- Evaluation function with his original parameters: material, king value, advancement, center control, back-row guard, mobility
- Move ordering: captures first (his "forward pruning")
- Self-play: Watch Alpha vs Beta mode shows Samuel's original training approach
- Red = You, Black = Alpha
- Standard checkers: mandatory captures, kings on back rank
- Try 7 ply for a real challenge
- Watch mode shows the AI playing itself, as Samuel did in 1959
| Mode | Description |
|---|---|
| vs AI (Alpha) | Full Samuel minimax, 3–7 ply |
| vs AI (Easy) | Random move selection |
| 2-Player | Pass & play |
| Watch | Alpha vs Beta self-play |