After writing the word "death" last time…
This time I drew another dog…
On March 12, he was crushed and ruthlessly crushed. In the 5-set human-machine battle starting from the 9th, he defeated Lee Sedol 3:0. The two sides will play the fourth set on the 13th, then have a day off, and then play the fifth set on the 15th.
After three games, who knows what Alpha Go is thinking?
When Lee Sedol lost in the first game, a reporter once asked him, what kind of existence is Alpha Go to you?
He avoided the question at the time, but during the subsequent game, I'm afraid everyone present was thinking about the same question.
No one knows its rank
no one knows its style
No one knows what it is thinking
When it makes a wrong move, does it know it is wrong?
When it knows that it has made an extremely perfect move, does it know how perfect it is?
When it uses its troops conservatively, is it because it is pessimistic or because it believes that the winning rate is high?
What kind of "intuition" is it trained through so-called "deep learning" and "reinforcement learning"?
Will it eventually become the God of Go, exhausting the infinite possibilities?
No one seems to have answers to these questions, including the engineers at Google Deep Mind.
We can see Li Shishi turning from uneasiness to anxiety, from anxiety to fear, to almost collapse, then barely regaining his composure, and then making a move again. Every minute is torture.
No wonder Fan Hui, who lost to Alpha Go 5:0 five months ago, described it as a "wall" in an interview – "It just doesn't move, and if you put any pressure on it, it will bounce back to you. All the power you exert on it, all the insults you put on it, all the smiles you put on it, are all fed back to yourself in the end, and this is how I feel about it.”
After the game, the on-site commentator used "him" instead of "it" when commenting on Alpha Go's style.
What a strange feeling people have towards machines. And if alpha dogs play checkers, chess, backgammon, and (Deng Ziqi) with humans, what will the result be like?
Artificial intelligence plays checkers and defeated humans in 1994
Checkers: A game that began in 3000 BC, making it older than chess.
Beginning in 1989, computer scientists and colleagues at the University of Albert in the United States started a project called "Chinook". His goal was to defeat the world checkers champion.
With the help of some top chess players, heuristics are used to program the master's experience into a computer program so that it can generate "knowledge" of the success and failure of each move in a specific situation. During the running of the program, researchers continuously monitor, correct and update the program. Through daily intensive calculations (equivalent to 50 personal computers running simultaneously at 200 times the peak computing speed), the "Chinook" program took shape.
In 1990, his program got its first chance to compete with a world champion.
However, until 1992, an artificial intelligence program never defeated the world champion.
Everything changed in 1994. That year, the designed program defeated the human world champion for the first time, and this achievement was also recorded in the Guinness World Records. This undefeated streak lasted until his program "retired" in 1997.
In 2001, guided by the goal of developing a perfect checkers program, Project Chinook was relaunched and ultimately succeeded. Said: "Achieving this achievement makes me tremble with excitement. Solving the checkers problem has been my dream for nearly 20 years. It makes me very satisfied to see the final result."
In 2007, this international checkers program named "Chinook" was continuously improved, and finally completed an information database of the best solutions for each step of checkers, becoming an existence that cannot be defeated. The relevant paper was published in the July 20, 2007 issue of Science.
In other words, after 18 years of hard work and studying 5 trillion checkers positions, an unbeatable checkers artificial intelligence program – Chinook () was built.
Now, the Chinook program no longer requires any expert experience because it has become a database of information that "knows" the best solution for each checkers move. Even if the opponent can make every move perfectly, the final outcome can only be a tie. In practice, it is too difficult to do this relying on human intelligence.
Artificial intelligence plays chess, and humans seem to have never won since 1997
Xiangqi: divided into international chess and Chinese chess. There are many opinions on the difference between the two. Bang Ge quoted a pretentious saying that Chinese chess represents China's Confucianism, and international chess represents Western ideas of freedom and equality. (I feel that this explanation is so weak and powerless, please ignore it. We will focus on the human-machine war)
The history of human chess playing can be roughly divided into three stages.
The first "Othello" human-machine game system to defeat the world champion
While studying at Carnegie Mellon University, Kai-Fu Lee developed the "Othello" human-computer chess system, which became famous for defeating the human world chess champion Brian Rose in 1988. The latter won the Othello World Championship on behalf of the United States in 1987. The battle agreed to win two out of three games, but he gave up after losing one game.
Kasparov couldn't sleep during the battle with "Deep Blue"
In 1997, the contest between the world's top chess master Garry Kasparov and the computer "Deep Blue" produced by IBM was, like Lee Sedol this time, the greatest contest in history. After several rounds of fierce competition, "Deep Blue" finally defeated Kasparov 2:1. This is an important sign of the rapid development of artificial intelligence.
Chinese chess masters also lose when playing games with machines
On September 21, 1995, in order to celebrate the completion of the IBM China Research Center, IBM invited the international chess champion Xie Jun to compete with the challenger, a computer produced by IBM. After a two-hour contest, Xie Jun won The army shook hands with the computer 1:1 to make peace.
In 2004, in China's first national chess man-machine battle, Chinese chess queen Zhu Chen lost two games in a row and finally lost to Ziguang Star. Since then, it has become commonplace for Chinese athletes to lose in the human-machine battle of China's national elephants.
In short, it seems that humans have not won since 1997.
In 2009, Fritz 4's chess ability reached the Grand Master level, with a rating of 2898. Magnus Carlsen, currently ranked number one, is only 2882.
If artificial intelligence plays backgammon, humans might still have a chance to win?
Backgammon: One of the black and white chess games that originated in ancient China. It is typically easy to learn but difficult to master.
Currently the best program for playing chess is Yixin.
Let’s talk about the results of the competition first (it is currently the highest level computer backgammon competition and is held once a year).
In 2006, in the first human-machine battle, the top Czech players played against the top AI at the time. The results were one win, one draw and one loss (these were all balanced starts).
In 2011, the result of the second human-machine battle between the top Czech players and the top AI at the time was 5:3, with the AI winning (both games had a balanced start).
In 2012, China's Yixin appeared on the Internet, beating all other AIs (that is, the software that defeated humans back then), and took the lead for three consecutive years.
In 2014, Yixin defeated the other AIs with a disparity of 153:9. According to this, AI should be able to beat top human chess players.
On Dan Yixin’s official website, author Sun Kai said:
Compared with computer chess, the development of computer backgammon is slow. Top backgammon programs still cannot surpass the strongest human players.
This is confusing. Later I guessed that there might be other hidden meanings behind this:
Compared with computer chess, the development of computer backgammon is slow. (Before Yi Xin came out) the top backgammon program still couldn't surpass the strongest human chess players.
But in fact, after research, the true meaning is:
Compared with computer chess, the development of computer backgammon is slow. The top backgammon program still cannot surpass the strongest human chess players (the Czech players are too bad. It is said that the Czech champion came to China to participate in the national championship and would never make it into the top 30).
It is said that Yi Xin's current level (in 2014) cannot beat the top 100 chess players in the world, but there has been no corresponding competition yet.
Sun Kai said that he could not surpass human chess players. He was not being modest. Yi Xin is now completely unable to compete with top chess players. This is recognized by the industry, so there is no need to organize any competitions.
(Part of the above content is quoted from Zhihu@hutianshuo)
There is so little information on the introduction of Yi Xin’s master, so let’s just quote the inspirational introduction from Baidu Encyclopedia.
When Sun Kai, who has been obsessed with Chinese chess since childhood, came across the "Chu-Han Chess", the idea of "making a chess program" was born. A small program that can only perform four-layer calculations, which seems very simple today, started Sun Kai's wonderful relationship with computer science. In the end, after continuous modifications, this chess program also achieved satisfactory results – defeating Sun Kai and his father.
However, Sun Kai’s story of man-machine game does not end there. In communicating with friends, he became very interested in the backgammon program. At first, due to the influence of multiple factors, Sun Kai underestimated the difficulty of human-machine backgammon to a large extent. But in fact, the backgammon game itself has characteristics such as no end under professional rules, many branching factors, difficulty in reviewing games, and the lack of data, which undoubtedly posed a problem for Sun Kai. However, it was never his style to give up at the first sign of failure. After experiencing the depression of meditating hard but making no progress, and the ecstasy of making a huge leap due to occasional inspiration, Sun Kai's backgammon "Yi Xin" finally became one of the best backgammon artificial intelligence programs.
Artificial intelligence plays Deng Ziqi
This combination of subject, predicate and object cannot be more dirty! The picture is simply beyond imagination! And do you think Brother Bang can really speak? Save it. Don't dirty goddess wuli.