Google unveils a self-teaching AI that became a board game master in just 3 days
In a development that may send Elon Musk and anyone who’s seen Terminator 2 running for their nearest bunker, Google subsidiary DeepMind has unveiled a new version of its Go-playing AI AlphaGo Zero that can learn skills with almost no human input.
Yep, DeepMind has only gone and built an almost self-teaching AI. While AlphaGo Zero can only play the board game Go at the moment, DeepMind hope to use self-learning AI to tackle bigger problems like climate change, medical science and perhaps even death.
The original version of AlphaGo famously defeated world champion Go player Lee Se-Dol 100-0 after after analysing data from over 100,000 Go games.
Zero, after being programmed with the basics of Go, mastered the game by playing against itself millions of times. Every time it won, it added new tactics and moves to its knowledge base. After just three days, Zero was able to defeat the previous version of AlphaGo to technically become the Go world champion.
Mastering a board game is admittedly a small step for artificial intelligence, but it opens up a world of exciting possibilities. The same processes used to develop AlphaGo Zero could be applied to scientific and medical problems, with the AI able to analyse data and present a solution much faster than us mere humans.
For the time being though, let’s marvel at the fact we live in a time when a computer can be the world champion at a board game (and hope that AlphaGo Zero doesn’t go all Skynet on us…).