Teaching a Computer to Play Chess Is Also Teaching It to Decide
In his 1950 paper Shannon noted that chess is an ideal testbed for machine decision-making — its rules are clear and its states enumerable, yet it is complex enough that all possibilities cannot be exhausted, so the machine must learn to "evaluate the merit of a position" and "make trade-offs under limited resources," an idea thereafter proven applicable to decision problems far beyond board games. Forty-seven years later Deep Blue beat Kasparov, and nearly seventy years later AlphaGo beat Lee Sedol, and behind these two human-machine matches nearly half a century apart one can see the shadow of the search-and-evaluation framework Shannon’s early paper proposed.