When Google creates software that wins the most complex game … – Atlantico.fr

The game of Go is the most complex game in the world, far more complex than chess. Until recently, most experts thought that it would take decades for software to be able to beat a human at the game of Go. But yesterday, Google announced a software developed by one of its teams artificial intelligence, Deepmind, named AlphaGo, beat the European champion of Go, 5 to 0.

What is the GB and why so difficult?

The Go is a board game abstract, which is played on a grid of 19 intersections by 19. Players place black and white stones on each in turn intersection. The goal is to surround the stones from each other to capture and, ultimately, to control at least 50% of the surface of the game.

The game of Go is extremely complex.

<- partner tags: videoad (top) -> <- pwbreak ->

There are 10 to the power 761 possible games of Go, more than there are atoms! in the universe, against 10 power 120 possible games in chess.

To be good at Go, one must be able to detect subtle patterns on the ground, and think strategically. The first software to have defeated a champion chess was a “brutal” approach: he had stored a large quantity of chess in his memory, and statistically determined to play the next round, depending on what was indicated in its database. This approach is simply not possible in the game of Go, because of the astronomical number of possible combinations. Most major experts of Go are often unable to explain why this or that choice works – the game requires a importnate amount of intuition, exactly what is missing computers

L. original approach AlphaGo

The usual approach, statistic, “brutal”, could not walk. But the team of Google, after a British startup, Deepmind, recently acquired by Google, uses the latest, most advanced artificial intelligence technique, the “deep learning”, as reported by the MIT Technology Review. The deep learning just trying to reproduce the level of abstraction of which the human mind is capable. How to “design” a picture? We just think of it as a series of pixels, or as color gradients – what a computer can do very well. Or you can map the as a series of forms that can be compared to other forms. It is this more conceptual approach which is represented by the deep learning. By a series of conceptual multiple elevations (for “Color gradients” to “forms” to “face” to “Mona Lisa”), the deep learning software allows to approach the level of conceptualization of a human mind.

AlphaGo has two simulated neural systems, which operate in parallel. Instead of trying to assess all possible contingencies statistically, like a chess software, which is impossible to Go, AlphaGo will focus on the next few possible moves that can be evaluated. Equipped with this conceptual approach which reconciles the human, AlphaGo can then compete with human players. And unlike human players, AlphaGo is never tired or upset – and unlike a human, it can cause by playing several million games per day

<- partner tags.! videoad ( bot) -> <- partner tags: mediabong ->