How An AI Made Better Decisions Than The World’s Best Go Players
Artificial intelligence has become a core part of many businesses, helping to assist in fields such as analysis, image reconstruction, automated decision making and diagnosis.
However, perhaps the best illustration of the rapid advances in AI comes in the form of strategic board games, and the finest example of AI’s development may come in a five-game series between professional Go player Lee Sedol and Deepmind’s AlphaGo AI system.
Go is one of the most complex games ever made despite (or possibly because of) its relatively simple ruleset, allowing for incredible creativity and a lot of moves to choose per turn.
Whilst the more famous AI challenge was the battle between chess grandmaster Garry Kasparov and Deep Blue, chess is a game that can (and in Deep Blue’s case was) easily brute-forced through sheer computation.
With more legal moves in go than there are atoms in the universe, a computerised Go player has to think more like a human being and make careful judgements.
Before the 2016 five-game series, AI Go players could barely beat amateur dan players and struggled on a full-sized board.
Deepmind, which was initially designed to learn to play computer games and has since moved on to predicting protein folds, worked differently by analysing the entire online library of Go matches, literature and play styles before using its neural networks and tree search system.
This helps Alphago calculate moves several turns ahead, something few other Go machines were capable of at the time.
After a flawed but dominant win over 2 dan player Fan Hui, Lee Sedol, a 9 dan professional and seen as one of the best players alive, proceeded to lose the first three games before winning an inspired game four with a move described as “divine” by other players.
Whilst this and a very close fifth game made the result closer than it may originally appear, it was clear that AI had made vast strides and had the capacity to make better decisions than some of the most intelligent people in the world.
What has been seen in the five years since is this technology being used to shape decisions in a variety of different ways and AI being part of many new industries.