Scientists have just created AI to defeat the world's best

We humans have almost accepted a harsh truth that we will never play chess as well as robots. But now, even robots have to admit that they will never be as good as . other robots.

A new platform, known as AlphaZero, can learn to play Go, chess and Shogi flags without any human intervention. Taking advantage of deep neural networks, AlphaZero quickly learned to play these flags "to become the strongest player in history".

AlphaZero is a product of DeepMind Technologies . This is the British AI company - and the subsidiary of Alphabet - specializing in AI research for many years. In 2017, DeepMind let the AI ​​champion at that time "retire" , but continued his AI research work. With AlphaZero, DeepMind's research has entered the peak period.

AlphaZero has been challenged with the best " AI " in the world in 3 chess games:

• Stockfish, AI playing chess world champion.
• elmo, the 27th Annual World Computer Shogi Championship Champion in 2017.
• AlphaGo Zero, DeepMind's AI is the most powerful Go player in history.

AlphaZero does not need people to guide it anything.

In all 3 matches, AlphaZero is only provided with knowledge of basic game rules. Before the battle with the AI, it quickly played with himself millions of matches, initially trying many random tactics to win, then gradually learned which strategies are most effective. through a trial and error process called "reinforcement learning".

Training and learning process takes 9 hours with chess, 12 hours with Shogi flag, and 13 days with Go Go, using 5,000 tensor processing units (TPU). For those of you who don't know, only a single TPU can handle more than 100 million photos on Google Photos every day, so AlphaZero is really a monster in terms of handling ability. Once the learning process has ended, AlphaZero is "released" to battle other opponents.

Of course, AlphaZero crushes all of them.

The unique thing about this research is that the learning algorithm is combined with a "search method" called the "Monte Carlo tree search" (MCTS). This is a method that go Go AI programs use to determine which moves to follow next. The DeepMind team used this system to help AlphaZero learn to play chess and Shogi flags, for the first time showing that MCTS can be adapted into other complex games.

Perhaps what makes chess players most excited is that AlphaZero does not need people to guide it anything. Study the tactics themselves and come up with crazy ideas that people have never seen before. Its highly dynamic and dynamic playing style surprised the chess player Matthew Sadler.

The unique tactics and abilities mentioned above turned AlphaZero into a great teaching tool for chess players - encouraging the creation of unprecedented tactical gameplay.

Perhaps the "AI kill man" story so far is not imaginative in the game world, with robots constantly beating us at stand-by games, or massively multiplayer online games like DOTA 2, and of course, the Go flag.

Does that mean that the AI ​​will defeat us in every fighting game invented? Probably not. Although the 3 games DeepMind tried above are quite complicated, they actually have some advantages for AI, such as the 2 players and all the information needed to carry out the next move. already available on the table.

Therefore, even if the AI ​​is sure to usurp us in chess games, they won't have a door when they fight or hang around with humans!