[ot-caption title= “A typical 19×19 Go board like the one played by Lee Sedol and AlphaGo in the recent match. (via wikimedia commons/xchen27)”]
Go is an ancient Chinese board game that offers a mind-boggling number of possible moves that would seemingly crush modern computing capabilities, but is that still the case? In a widely televised and media covered event, Lee Sedol, Korean famed Go master, lost to Deepmind’s AlphaGo artificial intelligence. This story has gained much coverage. This is because many anticipated a win for AI against a human in a game as complex as Go to still be decades away. However, winning 4 games to Lee Sedol‘s singular win in a 5 game match, AlphaGo proved that AI is rapidly improving at a rate that is set to replace humans in varying industries; this is leading some people to speculate that humans will face widespread unemployment in the near future.
Looking into the Go master beating program itself, AlphaGo combines various algorithms that allow it to narrow its search of possible moves; Go lets each player place pieces on any intersection within the board, lacking the rigidity of placement that other complex board games like chess have. Multiple algorithms divide up the board to reduce the quantity of outcomes that the program has to compute leads to its efficiency and effectiveness. Using machine learning that combines deep learning and reinforcement learning techniques, the Google acquired AI was able to essentially have a backlog of all of humanity’s greatest Go moves at its disposal. But, given the sheer number of possibilities in Go, it is more important for the program to demonstrate creativity.
AlphaGo exhibited the strengths of machine learning, which can be summarized as combining the power of evolutionary improvement with modern computational power. While Lee Sedol is recognized as an incredible Go master, other masters of Go are claiming that they will be able to beat the program as of the recently publicized match proving the competitiveness of an AI against the top human players in the world. However, Harvard Business Review notes some of the limitations of an AI as restrictions must be made to ensure it does not compute indefinitely and properly conforms to the rules of the game. AlphaGo was programmed to have a time cap per move to ensure that it meets the regulation quantity of time to play Go; interestingly, Lee Sedol uses his time less sparingly in times of critical moves than he would otherwise, which shows the human nature of plasticity and adaptability that may still be lacking in AI.
Ultimately, machines are proving ever more impressive in their ability to mimic and even surpass humans. Increasingly, humans will have to come up with solutions and preparations for what appears to be an inevitability of machines taking a greater share of the human job market as AI expands in a more general fashion and as AlphaGo goes toe-to-toe with more world leading Go masters.
Sources: Harvard Business Review, Google, Wired Photo Source: Wikimedia