close
tech

AI conquers chess: previously imitating humans, now self-taught

ae4b1ffe-1686-11ea-af6a-aaaa00151f43.jpg

Editor’s note: About 20 years ago, the Deep Blue system defeated the world chess champion, but the system did not make much real creative contribution at the time, and it is different now.This article is compiled from a medium article entitled “How Computers Are Reinventing Chess”.This is a standard casual game for players to play in the bedroom. It only takes about 9 minutes from start to checkmate.On one side is Magnus Carlsen, the world champion in chess. He is a well-deserved prodigy and became a chess master at the age of 13.On the other side is the iPhone program Play Magnus App, which mimics Carlson’s chess habits.Carlson reduced the age of the machine to 18, and then faced the program, but Carlson still encountered challenges.In the first few minutes, Carlson was frightened by an unexpected attack, and then he continued to fight, trying to equalize with the App, but eventually surrendered.We seem to see the app condescending and say, “You need to hone your chess skills, let’s try again!” Carlson could only respond with a smile.The event is nothing special.In fact, Carlson has posted multiple videos telling his story against virtual players of different ages.These videos clearly tell us: Whether it’s winning or losing, computers are Carlson’s least favorite opponents.The problem cannot be avoided.Carlson may indeed be the best chess player in human history, but why is such a chess player defeated by the computer again and again, how did humanity get to this point?The story of 1997 may be a chess layman. Let’s review it first: The story of computer conquering chess originates from dark blue. In 1997, the dark blue system defeated the world champion Go Kasparov.Since then, the machine has demonstrated its strengths. It easily chews the beautiful patterns and wonderful strategies that humans have thrown at it.But modern analysts have come to different conclusions: the machine is fragile, Kasparov has made many mistakes, and both sides have made obvious mistakes.The dark blue system won in the first game, but in the second game, the last move changed.At that time, the Deep Blue system had a chance to win a soldier, but it retreated. The Deep Blue system took another measure, which blocked the possibility of Kasparov’s counterattack.The computer’s behavior exceeded Kasparov’s expectations. He was deeply disturbed and eventually missed a chance to draw.After the match, Kasparov accused the Deep Blue system of cheating. He thought that there was a supermaster who helped the computer to make unexpected moves.Controversial moves may be accidents only.A few years later, Murray Campbell, a scientist who helped IBM design the Deep Blue system, explained that the move was caused by a loophole, and the team quietly fixed the loophole before the third round began.Unfortunately, the damage has already been done.In the subsequent games, Kasparov was not so confident.Unable to understand the action of Deep Blue, Kasparov wasted a lot of time. He wanted to use extraordinary human behavior to deceive the computer. As a result, he made mistakes early in the sixth game..In short, although Deep Blue has won, it is not a feat for the computer industry. It won because of human error.This incident tells us that human beings have weaknesses, such as hesitation, fear, guessing, and fatigue, so they are vulnerable to attack.Although the Deep Blue system does not perform well, it is tireless and consistent.When Kasparov’s intuition fails, the computer can easily win.Human Despair Chess may be a very elegant game, but the dark blue game strategy is aimed at ugly brute force.At that time, Deep Blue had not used neural networks and machine learning strategies.Instead, Deep Blue used its powerful raw power to speculate on potential moves at a speed of 200 million steps per second.The dark blue system evaluates each step based on a variety of different parameters, and then assigns a value to each parameter.Researchers analyze the chess games played by nearly one million masters, then determine the weights of the parameters, and then let the chess masters optimize.The dark blue system’s method of playing chess is equivalent to putting together numerous master chess games. Because the system has enough original computing power, it can predict the future and avoid major mistakes.To this day, there are more than ten computer chess engines all over the world. All engines run on standard hardware. They also rely heavily on the chess history accumulated in the past 200 years.In the competition, the chess engine can search a huge database to find the start before the game starts.When it comes to intraday, the system can ensure that it is in a good position.Before the end of the game, the system can use various strategies, which constantly search the database to make each step close to perfection.As for the rules of chess engine evaluation weights, they were developed with the help of a large group of chess masters.Contributors propose algorithm modification suggestions, and then make a test version, and then the new and old versions are duel until the researchers determine which version is better.Chess uses the Elo rating system, which means that the system judges weights based on the probability of defeating opponents.However, comparing the performance of computers and people is difficult, because very few people can compete with computers, and few people are interested in doing so.The machine can easily play 1000 consecutive games, so comparing computers with people can only be estimated.Nevertheless, as long as you look at the data of today’s top humans and top chess engines, you can see a “human despair map”.From the data, the computer is the ruler, but it is not perfect.They can’t predict the end of the game because the outcome is more likely than the atoms in the universe.To beat the human world champion, the engine doesn’t have to be perfect.As long as the computer is consistent, tireless, and doesn’t make obvious mistakes.AlphaZero’s extraordinary chess actually values ​​accumulation, which may be overlooked by laymen.Many chess champions say that the new generation will eventually defeat the older generation, not because they are younger and more energetic, but because they can gain more knowledge.Computers are not as good as humans if they are measured by the criteria used to create moves, but they have changed recently.In 2017, Google-funded company DeepMind demonstrated AlphaZero, the first generation of deep learning systems.At the beginning AlphaZero did not have built-in chess knowledge, there was no opening footwork directory, and there were no games played by millions of masters. It only knew the rules for playing chess, and nothing else.But AlphaZero learns and learns quickly.It plays chess with itself and reaches master level within hours.At the end of the day, AlphaZero already has superb skills to defeat the limited-edition Stockfish chess engine.Last year, Stockfish defeated the full version of Stockfish.When AlphaZero learns, humans can observe its progress, watch it evolve from a beginner to a master, and then continue to age.The hardware used by AlphaZero and Stockfish is basically the same, but the number of steps per second that AlphaZero analyzes is only one thousandth of Stockfish. The advantage of AlphaZero is not analysis speed, but learning.After analyzing the moves, Kasparov lamented that AlphaZero had a dynamic style just like himself.Matthew Sadler said: “AlphaZero has found the secret notebook of the best chess player in the past.” It came to the head of a human like an alien who could play chess.AlphaZero is very different from previous computer chess programs: AlphaZero does not need to simulate humans. It is equipped with a neural network and can understand the game itself.AlphaZero not only defeated humans, it may also help humans understand chess. No computer system could do it before. AlphaZero was the first.Compared to the defeat of Kasparov by the Deep Blue system that year, AlphaZero seems to be more significant.Translator: Minions.