Most checkers games end in a draw…


Photograph by kennysarmy

From The Atlantic:

At the highest levels, checkers is a game of mental attrition. Most games are draws. In serious matches, players don’t begin with the standard initial starting position. Instead, a three-move opening is drawn from a stack of approved beginnings, which give some tiny advantage to one or the other player. They play that out, then switch colors. The primary way to lose is to make a mistake that your opponent can jump on.

This would seem to make checkers a game amenable to computer play. That was certainly the idea back in the mid-1950s, when an IBM research scientist named Arthur Samuel began to experiment with getting a checkers-playing program to run on an IBM 704. He worked on the problem for the next 15 years or so, publishing several important papers on what he called—and what we all would now call—“machine learning.”

Machine learning is the underlying concept for the current wave of artificial intelligence. The descendants of that early work now promise to revolutionize whole industries and labor markets. But Samuel’s programs never had much success against actual humans. In May 1958, several members of the Endicott Johnson Corporation Chess and Checkers Club trounced the computer, much to the delight of the Binghamton Press and Sun-Bulletin.

“The human brain, sometimes lost sight of in an age of satellites, frozen foods, and electronic data processing machines, returned to former glories early today,” the paper said. “The 704, Dr. Samuel explained, does not think. What it does, he said, is to search its ‘memory,’ stored on tape, of checkerboard situations it has encountered previously. Then it rejects choices which have turned out badly in the past and makes moves which turned out well.”

“How Checkers Was Solved”, Alexis C. Madrigal, The Atlantic