The Korean Chess Academy announced that Li Shishi Jiuduan resigned as a professional chess player and officially announced his retirement. In the first man-machine war with alphago, Li Shishi’s magical excavation achieved the only victory between mankind and alphago.

Maybe it’s providence. Just yesterday, Google’s deepmind published a paper again and announced the launch of a new algorithm muzero. Through this algorithm, the training speed of artificial intelligence can be accelerated. At present, the learning performance of go, chess, general chess and dozens of Atari games is better than all current algorithms. Specifically, go can surpass the previous alphazero.

The traditional algorithm is the search tree, that is, to search for possible branches. But in fact, the problems to be solved by artificial intelligence are often very complex, and it is inefficient to construct the algorithm only by using search tree. Deepmind proposed an algorithm combining search tree and learned artificial intelligence, called muzero.

Deepmind announced that it will develop a smarter alphago algorithm

The above figure shows the power of the muzero algorithm: after running 1 million steps, the level of muzero obviously exceeds the orange line, that is, the level of alphazero before. It can be seen that muzero’s go ability exceeds elo5000, and there is room for improvement. It’s no problem to win chess, general chess and Atari games.

I just don’t know when Google will release an available muzero for go enthusiasts. At least there is a super artificial intelligence alphastar to accompany players in the field of E-sports.


Leave a Reply

Your email address will not be published. Required fields are marked *