Artificial intelligence (AI) not far from everything, but in the Board game world in any case, you need to be afraid of, apparently, no one at all. “Superhuman” – superhuman – call researchers of the Google deep mind several times, what you present in the current issue of the journal “Science”. It is about to become the new Generation of “alpha zero”, a self – learning machine, the play in the three most complex strategy – chess, Go and Shogi-practically unbeatable. A machine that brought all three games to perfection, and within a few hours, without any Training and with a KI ratio minimum computational effort.
Joachim Müller-Jung
editor in the features section, responsible for the area of “nature and science”.
F. A. Z. Twitter
With the attribute “superhuman” – had triumphed, the Creator of AlphaGo, as their AI machine from the house of Google, played the world champion at the Go Board in the ground. What was with chess and Shogi, the Japanese chess, it’s possible, repeated also with Go: The neural network in the computer is no longer to defeat the force of his very special ability to learn and immense computational capacity. So far, it has shot itself in the choice of Algorithms, however, the complexity of the respective game. In addition, an intensive Training and to measure the game success with the Best of the tray, a prior knowledge about the game strategy was quite necessary.
All of this, the Superhuman-reference shows up mostly in training, such as specialization, all of the müshsame training period is attributable to the new Software Generation, which is why already in the first part of the Science essay in two ways. Alpha zero is basically a continuation of the for Go designed AI, which is now associated with the Algorithms and neuronal networks of the other Board game-KIs clever. Result: alpha zero needs no Training to determine the best moves, it requires no special knowledge. The AI plays rather to his game to perfection against themselves, and gradually learns the best game strategies, without any Input from the outside. Only the rules of the game to be entered to the Computer and the evaluation criteria for successful trains will be defined.
