It is of historical significance. Experts say that this not only proves that computers can surpass the greatest brains in some challenges, but also shows the limitations and shortcomings of these smart metal blocks.
Deep blue also emphasizes that if scientists want to make intelligent machines that can think, they must decide the meaning of "intelligence" and "thinking". [Super Intelligent Machines: Seven Robots in the Future]
In many games, computers have their limits for a few days. At the Fair Center in lower Manhattan, Deep Blue beat Kasparov 2-/Kloc-0-0, and tied three games. This machine approaches chess by predicting many steps and through possible combinations-this strategy is called "decision tree" (think about describing every decision of a branch of the tree). Deep Blue "pruned" some of these decisions to reduce the number of "branches" and speed up the calculation, and still be able to "think" about 200 million movements per second.
Despite these incredible calculations, the machine still has shortcomings in other aspects.
"Although they are good, (computers) are quite poor at other types of decision-making," said Murray Campbell, a research scientist at IBM research. Some people doubt whether computers can work like humans.
"More interestingly, there is more than one way to look at a complex problem," Campbell told Life Science. You can look at it in a human way, with experience and intuition, or more like a computer. "These methods complement each other," he said.
Although the victory of Deep Blue proves that human beings can make great chess player machines, it emphasizes the complexity and difficulty of building a computer that can handle board games. Campbell said that it took IBM scientists several years to build Deep Blue, and all it could do was play chess. He added that it has proved more difficult to build a machine that can handle different tasks or learn how to complete new tasks. When was Deep Blue filmed?
Learning machines, the field of machine learning has not yet developed, and most of the computing power is not available, Campbell said. For example, IBM's next intelligent machine is called Watson, which works very differently from Deep Blue and is more like a search engine. Watson proved that it can understand and deal with the long-term "Danger 20 1 1" champion. The "KDSP" and "KDSP" machine learning systems developed in the past 20 years also use a lot of data, which did not exist in the era when the Internet was still in its infancy, 1997. Programmed design has also made progress. For example, AlphaGo, an artificial intelligence computer program that defeated the world chess and Go champion, works differently from Deep Blue. AlphaGo plays many board games and uses these modes to learn the best strategies. This kind of learning is carried out through neural networks or programs similar to human brain neurons. Campbell said that when Deep Blue was built in the 1990s, the hardware used to make Deep Blue was not practical. Thomas Haig, an associate professor at the University of Wisconsin-Milwaukee, has written a lot about computational history. He said that the deep blue hardware was a demonstration of IBM engineering at that time. This machine combined some custom chips with other high-end versions of PowerPC processors and was used in personal computers at that time. [CIA history. : artificial intelligence (infographic)]
What is intelligence Haig said: "Deep blue" also shows that the intelligence of computers may have little to do with human intelligence. "Deep blue" is different from the traditional symbols of classical artificial intelligence. It tries to copy the functions of human intelligence and understanding through a machine capable of general reasoning. So we strive to make a better board game machine. However, this strategy is more based on the computer manufacturer's idea of what is intelligence than on what is intelligence. As early as 1950s, chess was considered as something that smart people were good at. Because mathematicians and programmers are often particularly good at playing chess, they think it is a good way to test whether machines can display intelligence.
It changed in the 1970s. "Obviously, the technology that makes computer programs more and more powerful players has nothing to do with general intelligence," Haig said. Therefore, we don't think that computers are smart just because they are good at playing chess. We believe that playing chess is not a test of intelligence after all. "kdspe" kdsps "How scientists define changes in intelligence also shows the complexity of some artificial intelligence tasks," Campbell said. Deep blue may be one of the most advanced computers at that time, but it was built for playing chess, that's all. Even now, computers are still struggling with "common sense"-contextual information that people usually don't consider because it is obvious.
"People over a certain age know how the world works," Campbell said. Campbell added that computers have been trying to complete some pattern recognition tasks that humans think are easy to complete. "He said that in the past five years, a lot of progress has been made in perception issues, such as face and pattern recognition.
Campbell pointed out that another thing that computers can't do is self-explanation. A person can describe her thinking process and how to learn. Computers can't really do this yet. "Artificial intelligence and machine learning systems are a bit like black boxes," he said.
Haig pointed out that even Watson was "in danger!" If you win, don't "think" about hurting like a person [Watson] who uses the next generation processor to implement a statistical violence method (rather than a knowledge-based logical method)! In an email to Life Science, he wrote that this proves once again that being the winner of a quiz has nothing to do with intelligence, as most people think. However, as computers become better and better than us, "we either leave a very specific definition of intelligence or have to admit that computers are actually intelligent, but they are different from us," Haig said.
What is the next step of artificial intelligence? Campbell said, "Because humans and computers have different ways of thinking," it takes a long time for computers to make medical diagnosis, such as completely diagnosing themselves or dealing with a problem, such as designing a residence for elderly people who want to stay at home. Deep blue shows the ability of a computer to adapt to a certain task, but so far, no one has been able to make a universal machine learning system that can work like a specially made computer.
For example, computers can handle a large amount of data well and find patterns that humans will miss. Then, they can provide this information for human beings to make decisions. "A complementary system is better than a person or a machine," Campbell said.
Maybe it's time to solve different problems, he said. Board games, such as chess or Go, let players know the position of their opponents, which is called complete information games. This is not the case in the real world. The lesson we should learn now is that … we don't learn much from board games. Elligent's computer program Libratus defeated the best human poker player in the 20-day Texas Hold 'em Infinite Championship, which was considered as a game with incomplete information. )
As for the fate of Deep Blue, after the historic competition with Kasparov, the computer was dismantled; Its components are on display at the National Museum of American History in Washington, D.C. and the Computer History Museum in Mountain View, California.
Original articles on life sciences.