Go, the board game, is perhaps the most famous and complex board game ever created. Nearly 2,500 years ago, Go was invented in China. It involves placing a stone (the pawn) at empty intersections created by a grid, and is played by two opponents — one with white stones and one with black stones. If the player who controls the white stones has a pawn surrounded by the other player’s black stones, the white stone is frozen. The winner is crowned by determining which player enclosed the most territory by freezing the most stones.
The game is used symbolically in TV shows like “Counterpart” and in Rian Johnson’s film, “Knives Out.” Go is also famous for this mind-bending statistic: there are 2.00 x 10170 possible moves. That is more moves than there are atoms in the observable universe.
Lee Sedol, a Go legend and 18-time world champion, squared off against a Go robot in 2016. This game redefined how players and enthusiasts approach the game of Go. Sedol lost to the robot — although, the term “robot” is a stretch. His opponent was AlphaGo, an artificial intelligence (AI) that, unlike robots that interact with their environment via sensors, used machine learning algorithms to annihilate Sedol 4-1. That same year, Sedol retired, saying that the “AI was invincible” and that “best in the world” no longer meant anything.
The creators of AlphaGo went on to build a newer version of the AI that beat Sedol. It was dubbed AlphaGo Zero. This souped-up version was allowed to play against its predecessor, AlphaGo. What happened in these AlphaGo vs. AlphaGo Zero games is hard to fathom and experts from the Go community describe the play style of Zero as “alien.” In an article published in The Atlantic, top-ranked players said that it was as if Zero had beaten you before you knew the game had begun.
The conversation surrounding the prevalence of AI in our society is too complex to truly examine, but the prevalence of AI in international competitions is increasing and provides a miniaturized glimpse into a world where AI does more than play Go. At this year’s International Mathematical Olympiad (IMO) — an Olympics for high school math whizzes — enthusiasts and participants are saying that this could be the last year an AI is not gunning for gold.
An open challenge initiated by Microsoft seeks to develop software, or eventually AI, that can win the IMO. The initiative, known as the IMO Grand Challenge, has already generated a potential computerized contestant for next year. In 2013 Microsoft developed software, known as Lean, which they have been updating for an appearance in the IMO next year. It currently assists mathematicians in proof-writing and can verify mathematical work.
However, Lean is not yet capable of that blinding insight needed to elucidate the secrets of an IMO problem. AI like AlphaGo follow decision trees until they arrive at the best solution for the problem in front of them. An open-ended math problem with an infinite number of ideas that could lead to a correct solution is much more difficult to program. It is like starting from the bottom of a decision tree where there are thousands of possibilities. AlphaGo starts from the top where one move creates a new number of moves as the decision tree unfurls. Microsoft hopes to design a version of Lean intelligent enough to essentially work the decision tree backward until the solution at the top of the tree is determined.
If Lean were to win a gold medal at the IMO, an international competition where one person represents an entire country, what is the purpose of terming it an “Olympiad?” Does the AI represent the home country of the corporation or institution that created it? These questions are reminiscent of the fears many science fiction mediums bring out in all of us. The fear that robots may “take your job” or the fear of our lives are being orchestrated from some control room — these are the premonitions international competitions should keep in mind as they permit artificial intelligence to participate in high-stakes tournaments.