🎲

Arthur Samuel

Original nameArthur Samuel

American Computer Scientist, Coiner of the Term "Machine Learning"

Those Who Conceived That "a Machine Could Think" · Pioneers

Coining "Machine Learning"The Checkers ProgramEarly Practice of the Self-Play Training Method

Who they are

Arthur Samuel (1901–1990) was an American computer scientist who in 1959, while working at IBM, coined the term "Machine Learning," defining it as "a field of study that gives computers the ability to learn without being explicitly programmed." The checkers program he wrote for the IBM 701 computer was one of the earliest programs able to improve its play through repeated games against itself rather than wholly relying on preset human rules, and this "self-play" training idea reappeared in more powerful form nearly sixty years later in modern reinforcement-learning systems such as AlphaGo Zero. Samuel’s checkers program in a 1962 public match beat one of the then top American checkers players, Robert Nealey, a victory that drew wide media attention at the time, seen as the earliest prototype of the repeatedly reenacted story of "a machine can beat a human at an intellectual game."

Primary sourcesSamuel, "Some Studies in Machine Learning Using the Game of Checkers" (1959)

Key stories

Letting a Machine Play Itself — an Idea Older Than It Seems

To keep his checkers program improving, Samuel designed a training mechanism letting the program play against itself repeatedly and adjust the weights of its position-evaluation function by the game results, and this core idea of "self-play" is of one lineage in principle with the training method by which DeepMind’s AlphaGo Zero, nearly sixty years later, learned Go from scratch by playing itself. Samuel is thereby often seen as one of the earliest practitioners of the research direction of reinforcement learning, though "reinforcement learning" as a formal disciplinary name was not systematically established until decades later.

Relationships

Disciple

🎯Richard Sutton　Samuel’s self-play checkers program is widely seen as the earliest practical prototype of the research direction of reinforcement learning

Ally

🏷️John McCarthy

Echoes today

Below are how modern works borrow or reinterpret this name or story — not the original material. The two differ, so keep them apart.

The modern reprise of the "self-play" training ideaThe training idea of "letting a machine play against itself" that Samuel first practiced reappeared near sixty years later in a far more powerful form in DeepMind’s AlphaGo Zero system, becoming one of the core methods of reinforcement learning.

Appears in

"Can Machines Think?" — The Birth of AIThe Naming of "Machine Learning" and the Prelude of Cybernetics

← Back to Figures