The game of Weiqi has been conquered! Artificial intelligence has finally beaten a top professional Weiqi player. Chess was much easier to conquer, falling in 1997, but WeiQi has been mysteriously elusive to the slicing-and-dicing of cold mechanical computers calculations. But the interesting news is how “slicing-and-dicing” still doesn’t work for WeiQi. Instead of using old possible-move tree searching algorithms (what I am calling slice-and-dice), AlphaGo (Google’s victorious program) adds two more elements: data from expert systems and the final coup de grâce of 12 deep neural networks.
Now I won’t pretend to understand any of these three elements used in AlphaGo, but apparently, Google made AlphaGo’s neural networks play against themselves to discover its own new strategies.
Sure while I am excited about the WeiQi story in itself, this may also point toward new ways of using AI to address other apparently highly complex systems like disease, climate, political policy and more.
Caveat: When I say WeiQi (or any other game) is “conquered”, I just mean by AI. The games will still be a great delight to all us humans. Just because anyone (computers included) can do something better than us, does not have to take away any of our joy.
- See Googlelblog for my source: “AlphaGo: using machine learning to master the ancient game of Go.”
- The last half of this short video shows the reigning European Go champion, Fan Hui’s humble reaction after losing 0-5 to AlphaGo test.