While the collective tech world was agog at the recent Go championship match between a world master and Google DeepMind’s AlphaGo software, other AI innovations have quietly been making waves of their own. It might not be as flashy as a computer beating a human champion in four of the five games–or the human turning around and beating the computer, for that matter–but there are no doubt serious implications involved.
Microsoft, for example, announced the news that it was teaming up with Minecraft to create an AI interface. Long perceived as the go-to world building game for eight-year-olds everywhere, Minecraft is actually an ageless and highly-sophisticated gaming world of its own. With the advent of the open-source AIX interface, Microsoft is inviting AI developers to use Minecraft as their own personal proving grounds, largely due to the high-volume of unforeseeable game play ramifications.
Another software research lab has been working on developing an AI that can win at poker, a seemingly basic card game on the surface but one that is actually filled with nuance and pitfalls. As Thor Oluvsrud points out for CIO.com, poker–specifically Texas Hold ‘Em–is even more difficult for a computer to play than loftier games of strategy like Go or chess. In those games, the players (and therefore, the computer) can see the board at all times. They can develop predictive strategies in order to prepare their moves. But in card games, the players don’t know what all of the other players are holding, and therefore have a much harder time predicting what moves they’ll make based on the moves of other players.
But what is all of this effort for? Surely there’s more to AI development than being able to play against the computer on your tablet. As always, the goal for AI technology is to develop robots that don’t have to be told every single “if/then” scenario in order to make a sound decision. Coding for every possible outcome is hard enough when it’s chess, but it’s nearly impossible when it’s life-altering and emergency situations. For now, though, the tools to develop this type of understanding of outcome-based predictive decision making lie in games.