Some Studies in Machine Learning Using the Game of Checkers
Samuel described a program that adjusts its position evaluation from the results of its own games and plays better than its author.
Why it matters
The phrase machine learning appears in print here, alongside a demonstration that a program can outplay the person who wrote it.
The paper describes two mechanisms: memorising evaluated positions, and correcting the weights of the evaluation function from the difference between the value of the current and the next position. The second is what was later named temporal-difference learning. The program played itself to gather experience.