His contributions were significant, but he was not done, for after the War, closely tracking the locus of new technology, Samuel arrived at the University of Illinois to work on the giant ILLIAC scientific computer, and after being recruited by IBM, he was immersed in the design the world's first commercial mainframe computers that were soon to disrupt industry and education all over the world. But it was then that he came up with an audacious idea that would be the precursor of a transformative technology that would further disrupt industry and the lives of almost all the peoples of the world. His idea was inspired by a return to his everyman roots as the conjurer of a “thinking machine” playing a trivial game known and played by almost everyone in America.
The IBM 701
Working in Poughkeepsie in 1949, Arthur Samuel noted that the storage and display matrices of the new IBM 701 computers, looked just like a checkerboard, and like almost everybody in America at the time, he played that common game of jumping about a board to capture an opponents’ pieces and “kinging” your own pieces.
In his uncommon mind however, Samuel saw a clever way to create an image for IBM as the creator of a wondrous checkers-playing computer that could challenge and defeat humans at their own game, garnering publicity for the marvelous capabilities of IBM computers.
It was clearly a creative marketing scheme, but his company was not in the least supportive, for its venerable chairman Thomas Watson, Sr., like Samuel was conscious of IBM's image, but in the opposite sense of avoiding the spectre of a menacing “IBM thinking machine” going around beating up on humans all over the country.
The idea of the homey IBM Selectric typewriters and office machines souped-up like Frankenstein's monster to triumph over small-town residents at checkers was anathema to the friendly helpmate that was Watson Sr.'s marketing vision for IBM products.
Like Victor Frankenstein the earnest scientist, the dogged Samuel followed his checkered muse, jumping over the IBM powers by developing the 701 checkers machine on his own time. He encoded the rules of the game and programmed a decision tree move generator, evaluating moves by attaching numerical values to the move options that would most likely lead to a desired advantage. Included in the evaluations were nuggets of checkers wisdom familiar to any checkers player, for instance depleting the opponent's checkers by even challenges when ahead, simple gambits, and guile that today are called heuristics.
But this rote learning from rules, simple decision evaluators, and well-known heuristics would take the 701 checkers machine only to the level of its creator, which in Samuel's case, despite his technical brilliance, was below average, and so he consulted expert players for more advanced winning strategies and tactics.
Samuel found it difficult however to incorporate their checkers skill, being mostly based on knowledge of the proclivities and idiosyncrasies of opponents, and no little hard-to-codify “feel” and outright guessing.
Since this “top-down” encoding of rules, tree search results, heuristics, and learning from some expert players, although now “expert”, the 701 could only as well as those players, it was clear to Samuel that the 701 expert system needed to learn from playing against other expert players, and like a human player, 701 could accumulate playing skills “bottom-up” and from its own actual game experiences, gain knowledge on how to win by reinforcing its good move choices and degrading the bad moves.
Samuel now evaluated moves based upon their ultimate success or failure in training sessions based on the recorded matches of expert players, and the reward and punishment of good and bad moves in actual games against human players, and even in games played against itself. These were three learning schemes that would later be called respectively supervised learning on training sets, reinforcement learning in game situations, and unsupervised learning by playing progressively improving versions of itself, the basic methods of today's “machine learning”, a term Samuel himself coined.
By means of machine learning, the IBM 701 checkers machine slowly improved, and in 1962 after 13 years of part-time development, Samuel's machine challenged and easily defeated the Connecticut state champion Robert W. Nealey.
After the match, the former champion said he had not had such competition from anyone since 1954, when he lost his last game, but in the rueful pride characteristic of accomplished human beings, he also circumspectly exhibited a clear approbation of the IBM 701's “intelligence”.
Despite its successes, the checkers-playing IBM 701 had no such baneful pride, much less circumspection, but if pride instigates the will to succeed, thereby producing greater effort, a machine lacking such pride might make it inferior to humans in determination, but it has no equal in effort as it needs only electricity to practice tirelessly 24/7 against not only humans but other machines and itself.
And pride although motivating, if once broken can devolve to paranoia, as the great chess champion Garry Kasparov would later reveal in his acrimonious duel with IBM's Deep Blue.
Samuel's cool and collected IBM 701 in the ensuing years would have every reason to be proud, for it remained undefeated for 15 years until 1977 when it finally lost, not to a human, but to a rival checkers program developed at Duke University.
How prideful humans internalize defeat has been analyzed, but how a machine internalizes a defeat may never be known, for deep within the hidden layers of the artificial neural networks of today's AI machines, the germination and processing of a “thought” is largely unfathomable even to the learning algorithm's creator, and that unknown, contrary to the coldly logical computer ethos that usually makes it superior to an emotional human being, could conceivably leave room for machine attitude, willfulness, and even emotion.
The blessings of computers and artificial intelligence have been the machine's potential to increase production while freeing people from the drudgery of everyday work, allowing them to think about the work rather than just enduring the tedium of doing it, thus improving efficiency and leaving more time to pursue lofty goals and enjoy life. The bane of the machine is its potential to take over almost all human occupations, relegating humans’ activities to the care and feeding of the machines.