After beating three time European Go champion Fan Hui five games straight in October 2015, Google's AI division DeepMind has just taken the accomplishment one step further by defeating the world's number one champion Lee Se-Dol in the first of their five game matchup.
Considered significantly more complicated computationally than Chess, Go is a game where there are more possible moves than a computer can reasonably calculate, so some level of "intuition" is required. Also, Go champions typically use the unconscious reactions to plays from their opponents, such as breathing changes and eye movement, to determine their next move much like poker. This means that a human playing against a computer has an inherent disadvantage compared to their normal games.
DeepMind, purchased by Google in 2014, has programmed their AI AlphaGo using machine learning, where it first studied a large number of previously played games and then ran simulations against itself with slightly altered combinations in order to determine what plays were best in which situation. This machine learning process means that AlphaGo is even more advanced than when it beat Fan Hui in October.
Source: BBC | Image: Artificial Intelligence via Profit_Image / Shutterstock
7 Comments - Add comment