AlphaGo (2017)

The ancient Chinese game "Go" has always been considered a major challenge for artificial intelligence, as more board configurations are more than observable atoms in the universe. March 9, 2016, Go and Artificial Intelligence competed in Korea, five best game competitions were held, and Google Deep Mind Challenge was born. Hundreds of millions of people all over the world saw legendary Go Master for the first time in history to accept unproven AI challengers. AlphaGo is a director of Oscar candidate Hauschka, who recorded the way from Cambridge Hall to the back street of Bordeaux, passed through Deep Mind Code Terminal in London and finally took part in a 7-day tournament. In Seoul. As the drama is unfolded, more questions arise: Can artificial intelligence reveal games of 3,000 years ago? What can we teach about human nature? Written by AlphaGo

Garry Kasparov, an excellent chess master, was defeated by IBM's Deep Blue in 1997, Chin Sedol was defeated to Chinese board game Go champion, in 2016 lost to Google AlphaGo. In 2017, AlphaGo - Zero (Variant of AlphaGo) got support for playing on its own. Kill world champion with games such as chess and go - these games are widely regarded as a microcosm of human intelligence by society - not necessarily clever artificial intelligence than humans. Currently, the expertise of artificial intelligence is limited to a few fields, which limits its universality. For example, a cow may give milk, but you can not learn to fly an airplane. In the defense of cattle, people may argue that flying planes are not born. Therefore, AlphaGo (which is designed to run Go) can not even defeat young children in chess until you learn to play chess (but until later to be called AlphaChess). Application AI may use the domain expert too much, but it has domain specific features

In 2017, DeepMind released the next generation AlphaGo. AlphaGo Zero is built on simplified hardware and only learns Go to play. After 3 days of training, AlphaGo Zero surpassed Lee Sedol more than AlphaGo, breaking his young version 100-0. After 40 days of training, it also beat his previous version of AlphaGo Zero. So far, there are several expert artificial intelligence systems, for example AlphaGo (Zero) is mastering Go and is superior to a human master. However, there is no artificial intelligence (AGI) yet. AGI refers to artificial intelligence that can successfully execute all intellectual tasks human beings can perform. The Turing test is an example of testing whether AI is AGI.

In October 2017, DeepMind released AlphaGo Zero. As the latest version of AlphaGo, AlphaGo Zero is built from scratch. AlphaGo Zero does not train millions of mobile data in race, but trains by starting a random game to compete with your copy.

The AlphaGo team has released an article introducing AlphaGo Zero in Nature's October 19, 2017 issue. This is more powerful than the defeat version of the human champion ever. By playing his own game AlphaGo Zero break through AlphaGo Lee's strength in three days, win 100 games 0, reach AlphaGo Master level within 21 days, and in 40 days all old Version exceeded. In a paper published in arXiv on 5th December 2017, DeepMind claims to extend the AlphaGo Zero approach to a single AlphaZero algorithm that breaks world champion in chess, chess, Go game in 24 hours I will. Superman's game level has been reached. In either case, the program, Stockfish, Elmo, and the 3-day version of AlphaGo Zero