Tuesday , November 12 2019
Home / chile / "AlphaZero surpasses us in a deep way": a computer that produces its own knowledge and plays like a "superhuman"

"AlphaZero surpasses us in a deep way": a computer that produces its own knowledge and plays like a "superhuman"

robotImage copyright


There are machines that behave like supermen.

It's been a long time since the engine stopped being a challenge to overcome humans.

In many ways, robots are proven "Harder, better, faster, stronger", as said by the song by Daft Punk.

Now, the challenge is to become more stylish, or at least that's what AlphaZero's computer shows, one of DeepMind's latest developments, a Google company dedicated to artificial intelligence.

A year ago, AlphaZero defeated other machines that were trained to play chess; shogi, known as Japanese chess; and go, one of the most complicated board games available.

  • Magnus Carlsen, world chess champion: who is Norway who makes this sport fashionable

Now, a year later, the creator of AlphaZero reveals the key to the engine way destroyed for his opponents, playing at the level of "superhuman", according to its creator.

Image copyright


Shogi is considered as Japanese chess.

"AlphaZero learns to play three challenging games at the level highest that has been seen, "he wrote in a magazine science Murray Campbell, IBM researcher.

There is no need to mention that he defeated humans, because the barrier was overcome 20 years ago, when the Deep Blue computer defeated Garry Kasparov, maybe the best chess player of all time.

  • What can we learn from the feminine wonder of chess, male dominated sports

But the engine, although more powerful, has so far failed to imitate human characteristics such as instincts or imagination

"Just as planes don't flap their wings like birds, machines don't produce chess movements like human", Kasparov wrote in the latest article on science.

But it seems like that change with the arrival of AlphaZero.

Image copyright


Gary Kasparov was "pleased" by the way AlphaZero played.


Armenian chess player Tigran Petrosian, who is a world champion, is credited with the phrase: "the best trainer chess is yourself ".

And that's what AlphaZero did.

  • Why did Armenians succeed in chess?

Until now, the most powerful artificial intelligence programs have been based on a combination of search techniques, adaptations to the environment presented to them and functions that enable them to evaluate the situation.

All this, obviously, comes from that human being train and they program it according to what they think is the most optimal.

But that didn't happen with AlphaZero, who was only told the rules of the game and from there he strengthened the learning by playing against himself.

Image copyright


AlphaZero is his own chess coach.

Within a few hours, AlphaZero plays more games against itself than has been registered throughout the world story chess

"Programs usually reflect priorities and prejudice programmer, "Kasparov wrote." But like the AlphaZero program itself, I would say that his style reflects the truth. "

As such, AlphaZero keeps away from the way humans play and starts to show the way extraordinary to understand chess.

In his article, Kasparov admitted that he was "happy" to see AlphaZero playing in a "dynamic and open" style, preferring a position that for humans like he would appear "Risky and aggressive".

According to Kasparov, he could see that AlphaZero "reevaluate basic concepts like chip value "and" prioritizing chip activities on matter ".

Image copyright


Artificial intelligence has applications in many fields.

Part of AlphaZero's success is not only to rely on its ability to analyze thousands of possibilities in one second, just as calculator, but focusing its search on drama that looks superior.

  • The "Turkish trick" that destroyed the reputation of expert chess players in 18th century Europe

In other words, AlphaZero doesn't work harder, if not more intelligent.

Although for now AlphaZero has no function other than playing chess, shogi, and leaving, its creators hope that this technology can be applied to other areas such as science and drugs

"AlphaZero surpasses us in a profound and useful way," said Kasparov, "a model that can be duplicated in tasks or other fields where it can be produced. knowledge virtual ".

This is what DeepMind says, which confirms that their goal is "to develop programs that can learn to solve complex problems without the need to be teach How to do it. "

Now you can receive notifications from BBC News World. Download the new version of our application and enable it to not miss our best content.

  • Do you already know our YouTube channel? Subscribe!

Source link