Arthur Samuel

Overview

Arthur Samuel was an American electrical engineer whose career ran from telephone hardware at Bell Telephone Laboratories to computer programs that improved at a board game. He is remembered mainly for two things: a checkers program written for an early IBM computer, and a 1959 paper whose title gave wide currency to the phrase "machine learning."^[2]

Education and early career

Samuel was born in 1901 in Emporia, Kansas. He earned a Master of Science degree from MIT in 1926 and stayed on there as an instructor in electrical engineering from 1926 to 1928.^[1]

In 1928 he moved to Bell Telephone Laboratories, where he worked until 1946.^[1] He then spent three years as professor of electrical engineering at the University of Illinois, from 1946 to 1949, and in 1946 also received an honorary doctorate from the College of Emporia.^[1]

The checkers program

In 1949 Samuel joined IBM's Poughkeepsie Laboratory, where he remained until 1966.^[1] There he wrote the first checkers program for the IBM 701, IBM's first commercial computer.^[2] Getting a game-playing program to run at all on a machine of that era was an engineering achievement; getting it to play well was a research problem.

Two techniques in the program matter for the history of the field. The first is alpha-beta pruning, a search method for two-player games: the program explores possible moves and replies, but abandons a branch as soon as it can show that the branch cannot change the decision at hand. Pruning lets a program look further ahead in the same amount of computing time. The second is rote learning: the program stored the positions it had already seen and could recall them later instead of working them out again.^[2] That memory of past play is what made the program more than a fixed set of hand-written rules, and it is why the work is treated as an early demonstration that a machine could pick up something from experience.

In 1959 Samuel published "Some Studies in Machine Learning Using the Game of Checkers" in the IBM Journal of Research and Development, and he is credited with coining the term "machine learning" in that year.^[2] The game served as a controlled testbed: the rules are simple and fixed, wins and losses give an unambiguous score, and a program's strength can be measured by playing it. Those properties made checkers and similar games a standard proving ground for learning methods for decades afterwards.

Stanford years

Samuel retired from IBM in 1966 and joined Stanford University as a lecturer and research associate, becoming a research professor there in 1974.^[1] At Stanford he continued to work in an academic setting at a time when computer science was separating from electrical engineering and mathematics into a discipline of its own.

He received the IEEE Computer Society Pioneer Award in 1987.^[1] He died on 29 July 1990 at Stanford Hospital.^[1]

Why he matters

Samuel's contribution is easy to state and easy to underestimate. He took a machine that filled a room, gave it a game with clear rules, and asked whether it could be made to play better by accumulating experience rather than by following instructions written out in advance. The name he gave that question in 1959 is the name the field still uses.^[2] His checkers work also set a pattern that outlasted the hardware: games as benchmarks, search plus memory as the basic machinery, and published, reproducible experiments as the way to show that a learning method works.^[2]