Ever since 1997, when the DeepBlue computer first defeated the legendary chess player Gary Kasparov, there has been no excitement, harassment and, finally, disbelief like this March, when Deep Manndo. The new artificial grandmaster, among them, possesses something that DeepBlue did not have: a program of ways to develop some kind of intuition and perfect it. The application of such software can be wide - in climatology, economics, and even in healthcare.
"If we had played a full computer at the beginning of the game, we would have been a good duck for our teachers,"
was heard among young players analyzing games played by AlfaGo and game grandmaster Lee Sedol. Unlike Kaspareva, who ruled out one of five computer games while three were unresolved, Sadle is basically one, losing four in the first man fight of the 21st century. To speak more about the machine than about man.
But before that, something about the game itself. Unlike chess, in which two players have two sets of different pieces at their disposal, which are, again, different table crests, the rules are much bigger: two players have 180 black or other identical pieces at their disposal, and the task is to determine which " Territory "which is counted by uninhabited sections of 19 horizontal and vertical lines. Also, unlike chess, players on the move face a far greater number of directions in which the game can develop. Therefore, until last year, it was considered unattainable for artificial intelligence - something, with knowledge of the rules and a large base of built games, would be a program that must prepare human intuition to recognize the potentially greatest conditions for the situation ... develop artificial interests. Since its founding in 2010, it has managed to attract investors of the caliber of Ilona Moscow, and part of Google is in 2014 for 500 million dollars. Their most important resource is the ability of the neural network to "learn" to play games in a similar way to humans, the network networks in AlfaGo. This program can not only learn to lose tens of millions of built games, but also play games with itself so that it can sharpen its ability to choose the right move. And that this move by people may seem irrational or wrong is confirmed by the influence of "kibbutz" in the initial text. and part of Google is in 2014 for $ 500 million. Their most important resource is the ability of the neural network to "learn" to play games in a similar way to humans, the network networks in AlfaGo. This program can not only learn to lose tens of millions of built games, but also play the game with itself so that it can sharpen its ability to choose the right move. And the fact that people can look at this move irrationally or wrongly is confirmed by the influence of "kibbutz" in the initial text. and part of Google is in 2014 for $ 500 million. Their most important resource is the ability of the neural network to "learn" to play games in a similar way to humans, the network networks in AlfaGo. This program can not only learn to lose tens of millions of built games, but also play games with itself so that it can sharpen its ability to choose the right move. And the fact that people can look at this move irrationally or wrongly is confirmed by the influence of "kibbutz" in the initial text. but also play games with yourself so you can sharpen your ability to choose the right move. And that this move by people may seem irrational or wrong is confirmed by the influence of "kibbutz" in the initial text. but also play games with yourself so you can sharpen your ability to choose the right move.
WHAT DID THE PROGRAM MEAN?
Is there a skill that AlfhaGoista can measure with human intuition? In an article for atlanticss.com, Michael Helsen needs a vivid development path for this program, which can be compared to the development of a quality player of modern computer games: from beginner's troubles, when persistence means much more than imagination, to creating the game itself no longer hides behind the player.
AlfhaGo has taken the same path as a man, which makes him close to us, and many days since DeepBlue, his proud ancestor with Kaspar’s scalp behind his belt. The key to success with the machine lay in the fact that it could be assessed not only which next move they could regain, but also to achieve that quality move and bring them closer to the winners. The "old" logic behind DeepBlue means that it will be possible to look for a huge party base and choose the right move based on that. Maybe Kasparov was furious back in 1997 - he saw that the machine was not more intelligent than him, but only mechanically faster in "reviewing" the built-in games in his village.
The creators of AlphaGo started in a completely different way: the first steps of the "artificial champion" were for his neural network to analyze 160,000 games played by quality players from the world list, and thus learn the basic mechanism of the game. He then begins “self-training,” playing games with himself where the program begins to “learn” which moves are useful and desirable, using the neural network property to adjust game parameters until the game’s outcome is “desirable.” Interestingly, the program creators themselves they know, nor can they predict what move the program will make during the "training", which they admitted to VIRED magazine:
"Even though we programmed this machine to play it, we never know what it will do next. The moves that the game derives from the situation in which it currently finds itself. All we do is provide data and a training algorithm. The moves he decides on are his work, and we have nothing to do with them - even those moves are better than we, who also play him, could think of. "APPLICATION OF NEW ARTIFICIAL INTELLIGENCE
Even before the tournament in which their program was won by Lee Sedol, Google DeepMind signed a contract with the British health system NHS. The department, called DeepMind Health, will use the neural network, which is able to learn from available data, to create applications that will help doctors and medical staff to even better assess which patients are at risk of complications, according to a company statement. Their first product for that purpose should be the Streams application, which they are currently working on.
One of the co-founders of Google DeepMind, Demis Hasabis, sees the use value of neural networks and learning-capable programs almost everywhere: in the fight against climate change, cancer treatment or macroeconomics. As he told the British Guardian, "General Artificial Intelligence" (AGI), unlike Artificial Intelligence (AI), will be able to learn as the human brain does, but also to meet challenges that would take a lot to solve. more time than a human life lasts.
DeepMind-Alphabet
Until a new kind of artificial intelligence takes over solving the world's troubles, we lay people should start to distinguish "smart" machines from those that only imitate that mind: all those situations when it seems to you that Google has chosen the right ad for you and just because you really need gardening equipment or cheap airline tickets doesn't mean that the search engine has become smarter and knows what you need - it's programmed to track the terms you're searching for and offer you the same or related topics. The same goes for services like YouTube or the Siri app on the iPhone, whose function is to be something between an assistant and an advisor.
When machines present themselves to us as intelligent one day, we should keep in mind that they have to prove such a claim. The most well-known and widely accepted test of intelligence in that sense is the Turing test, which implies a conversation through the exchange of text messages between the examiner, a machine whose intelligence is proven, and a third participant-human. The machine and the man are, of course, hidden from the examiner during the test, and the machine will prove its intelligence if it manages to answer in writing as a man would answer. It is not the only test of intelligence. Scientist Ben Herzel suggests a far simpler approach: to order the machine to enter the average American house, with an order to make coffee. If she manages to find a coffee machine, water, coffee and cooks a black drink properly - here are her confessions that she is smart!
The fact that they chose an ad for us is remembered somewhere that we were once interested in this term or this product, for example... I think it works that way, it doesn't have to mean I'm right