Arthur Samuel (computer scientist)

• 1. Introduction

  ›

  Arthur Lee Samuel coined the term "machine learning" in 1959, and the computers of his era had almost no memory to spare. Working at IBM in Poughkeepsie, New York, Samuel faced a problem that would define the rest of the century: how do you teach a machine to get better at something it has never been told how to do? His answer was a checkers-playing program that could improve through experience, remember positions it had already encountered, and eventually beat a respectable amateur. The program caused IBM's stock to jump 15 points overnight. What drove a man trained in vacuum tubes to spend decades teaching a machine to play a board game? And what does that obsession reveal about the birth of artificial intelligence as a field?

• 2. Emporia to Poughkeepsie

  ›

  Samuel was born on the 5th of December 1901 in Emporia, Kansas, and graduated from the College of Emporia in 1923. His path then led him to MIT, where he earned a master's degree in Electrical Engineering in 1926 and stayed on for two years as an instructor. Bell Laboratories came next, in 1928, and there his work turned largely to vacuum tubes. During World War II he developed a gas-discharge transmit-receive switch, known as a TR tube, that let a single antenna both send and receive signals. After the war, the University of Illinois at Urbana-Champaign brought him in as a Professor of Electrical Engineering. He helped initiate the ILLIAC computer project there, though he left before that first machine was finished. In 1949 he arrived at IBM in Poughkeepsie, and the decade that followed would carry him far beyond radar and vacuum tubes.

• 3. Teaching the Machine to Win

  ›

  Samuel chose checkers deliberately. He believed games were an exceptionally productive way to develop tactics that could apply to general problems, and checkers offered both relative simplicity and genuine strategic depth. Memory was the first hard constraint: with very limited storage available, Samuel implemented what is now called alpha-beta pruning, cutting off search paths that could not possibly improve on the best option already found. Rather than tracing every game to its conclusion, he built a scoring function that estimated each side's chances at any given board position, weighing factors like piece counts, the number of kings, and how close pieces were to promotion. The program then chose moves using a minimax strategy, picking whatever move optimized that score while assuming the opponent was doing the same from the other side. To deepen its learning, Samuel introduced what he called rote learning: the program stored every position it had seen along with the resulting reward value, effectively extending its search depth each time a familiar position reappeared. Later versions studied professional games to refine the reward function further, and the program also played thousands of games against itself. By the mid-1970s, after more than two decades of refinement, it could challenge a respectable amateur.

• 4. IBM, Stock Prices, and the First Hash Tables

  ›

  Samuel made his first checkers program on the IBM 701, which was IBM's first commercial computer. The public demonstration was striking enough to move IBM's stock up 15 points in a single night. His work at IBM stretched well beyond checkers. He is credited with one of the first software hash tables, a data structure now fundamental to computer science. His influence also shaped early research at IBM into using transistors for computers. Because Samuel worked on projects that were neither calculation nor numerical processing, his non-numerical programming helped define what instructions processors would need to handle. He was also known for writing clearly about difficult subjects; in 1953 he was chosen to write an introduction for one of the earliest journals devoted to computing, and that same year he published a piece titled "Computing bit by bit, or Digital computers made easy" in the Proceedings of the Institute of Radio Engineers.

• 5. Stanford, TeX, and a Final Chapter

  ›

  Samuel retired from IBM in 1966 and joined Stanford University, where he spent the rest of his life. At Stanford he worked alongside Donald Knuth on the TeX typesetting project, contributing to its documentation. He was recognized within the TeX community as a senior member who gave personal attention to individual users. In 1983 he wrote "First Grade TeX: A Beginner's TeX Manual," published as Stanford Computer Science Report STAN-CS-83-985. He kept writing software past his 88th birthday. The IEEE Computer Society gave him the Computer Pioneer Award in 1987, recognizing his adaptive non-numerical processing work. In 1990, the year he died, he was named a Founding Fellow of the Association for the Advancement of Artificial Intelligence. Samuel died of complications from Parkinson's disease on the 29th of July 1990. The term he invented in 1959 now names one of the most active fields in computing.