google logo

Google’s Artificial Intelligence Masters ‘Most Complex’ Human Game

Google’s artificial intelligence has roundly defeated the European champion of Chinese game Go. Go is widely considered to be the most complex game ever devised by humans, thus making it perfect for the testing of artificial intelligence and learning programs. Google’s artificial intelligence, known as AlphaGo, defeated the European champion Fan Hui in a tournament that stretched over five games. AlphaGo won with a score of five games to nil.

It has long been considered one of the benchmark feats of an artificial intelligence to be able to defeat a professional human player at a game like Go. Because of the nature of the game, Go presents computers with a near impossible set of variables at each move. Traditionally, artificial intelligence players would use a kind of brute force search mechanism to simply search through all the possible contingencies and outcomes arising from each move, something which isn’t really feasible with Go. Google’s artificial intelligence, however, took a different approach.

Instead of attempting to search or derive every possible outcome of every possible move, AlphaGo used two neural networks to narrow the range of its searches. The first of these networks, the policy network, would choose a range of moves based on whether or not they were likely to lead to a win. The other network, called the value network, would then rank these moves according to strength. This would leave a handful of moves to choose from, and a firm set of criteria by which to choose. Instead of “tree” searching to the end of the game, like most artificial intelligence, AlphaGo would only extrapolate sufficiently to calculate the next move. It would then measure the outcome of this move and use this data to learn strategy in realtime.

Artificial Intelligence masters Go
Go has long been considered far too complex for computers to beat. [Photo by China Photos/Getty Images]
Interestingly, the artificial intelligence’s game play was not flawless. AlphaGo did in fact make mistakes from time to time. Fan Hui, in the video of the tournament, pointed out that this was encouraging to him as a player. Obviously it wasn’t encouraging enough, as the hapless human player lost every single game.

Defeating professional human players has been a bit of a holy grail for artificial intelligence developers for many years. Card games and board games have been successively falling to the rise of the machines, with Deep Blue famously conquering the game of chess. Go is considered a major benchmark because of its open-ended nature and complexity. It has long been considered so complex that only a human capacity for single instance learning and directed imagination would be able to consistently achieve victory. Google’s AlphaGo has definitively put paid to this idea.

Recent rapid advances in the development of artificial intelligence have been deeply worrying to many high-profile individuals. Stephen Hawking and Elon Musk are two of the most prominent individuals calling for some kind of ethical framework to be enforced around research into artificial intelligence. The concerns surrounding the science mainly revolve around the possibility that we may create an autonomous and superior intelligence capable of harming humanity. Elon Musk was reported in Inc. as claiming artificial intelligence as humanity’s “biggest existential threat.” Stephen Hawking has said artificial intelligence could spell the end of humanity. While some say this is an exaggeration, there are very real concerns that our very human tendency to optimize and eventually weaponize most technological innovations might potentially cause significant problems for the human race.

In the case of AlphaGo and Deep Mind (the company that developed it), ethical considerations have been paramount. Google developed in parallel an ethics board that closely monitored each development of the AlphaGo intelligence, considering its implications and potential impact on humanity.

[Photo by Chris Jackson/Getty Images]

Comments