Artificial intelligence, great and terrible
My boy! I give you this song.
Calculate your strength.
And if you don’t know how to say “oink”,
Squeal, do not hesitate: "AI!" ...
S. Ya. Marshak
This topic has been constantly emerging in various discussions over the years. From the stormy, with fights and tantrums (about once a decade) to more relaxed during times "between the peaks." Now, it seems, another “uptrend” has begun. Although, basically, the same tortured theses are scolding: about the stupid Turing test, about the even more stupid "laws of robotics" Azimov, about emotions and consciousness, about chess and go, plus all kinds of horror films about Skynet. Strangely enough, there is still a lot of talk about chess, although it seemed to me that they should have stopped a long time ago. Well - this topic is closest to me, and we will begin with it.
No, I understand that chess was once associated with AI. This continued as long as the computers were playing poorly, and therefore the sapiens had the opportunity to turn up their noses and boast about their UE, which they allegedly have a “default”. But now? Yes, it is no longer interesting to play chess with a computer, it constantly wins . What about intelligence? Recognize the presence of a piece of iron? Never!
')
Even leaving aside the obvious nonsense about “expert systems,” it is clear that a person will come up with all new, even the most idiotic excuses, just to keep faith in the fairy tale that he has this very I: The definition of reason changes with time. Once it included the ability to multiply - crossed out, even the glands (without electronics) do it. Then they crossed out the ability to play chess, then, as they say, everywhere. The concept of mind narrows as shagreen. As it was rightly said here: No need to hang a fashionable AI label on everything that moves! And I add: and just And do not! The same Turing test assumes that the AI must descend to the intellectual level of the "examiners", who have been credited with intelligence, and lie to their eyes - and all this only so that they deign to admit that they have intelligence? Yes, a normal AI should be above these nonsense! Let us recall: billions of flocks of sheep for thousands of years believe in religious ravings of the lowest possible order, and no one doubts that they have intelligence.
What am I talking about? Oh yes, about chess. Although the people have already switched to go. Not for long, of course - until people got there in the face: But they write that computers still play very poorly in Go - there are many more combinations than in chess. Go game has always been a tough nut to AI. Skeptics argued that go programs would never win, or they would win very soon. Go is more complicated than chess by 10 ^ 100 times — that is, so many times more possible positions of stones on a standard chessboard 19 * 19 than in chess. Something quickly you, gentlemen, have forgotten a similar bleating about chess: The difficulty of analyzing a chess game is that the number of unique chess games is approaching the number 10 ^ 120 and other nonsense in other AI tasks like the Bremmerman limit.
So is a computer chess player stronger than a man? Does anyone still have doubts? He is even stronger because he is stronger, but because people have already surrendered! Kasparov said: If a computer can beat the best of the best in chess, this will mean that the computer is able to compose the best music, write the best books. I can not believe it. If a computer with a rating of 2800, that is equal to mine, is created, I myself would consider it my duty to call him for a match in order to protect the human race. No longer twitches. And Sergey Karjakin expressed himself even more clearly: In a fight against a computer, the feeling of doom does not leave you. And why did the magnificent tournament in the Hague "cars against the people", 50 to 50, wall to wall? Yes, all for the same reason: people have no chance! Although in the early stages of this competition, there were mostly amateurs playing there, and almost the only representative of the chess elite of mankind was David Bronstein, and in the end there involved a whole constellation of outstanding chess players: Vahanyan, Seyravan, sisters Polgar, Kosashvili, Christiansensen, Timoschenko and others. So no need to find out who is stronger: a person or a car - people have already answered this question! Not to mention the fact that no one has yet programmed a computer against a game against a person (which is quite simple to do). Peter Svidler led the line: There is no chance for a man to defeat a supercomputer in chess for a long time. All discussions on this topic are long over .
Svidler, by the way, played with our program during the times of the Kremlin Stars. Although some grandmasters (Short, Nikolic) flatly refused to play with the computer "in public" (in the lobby of the CBC), others (Svidler, Drahomaretsky) were what are called "without complexes". During one of the games, an amusing incident occurred: a boy of ten began to suggest (something like “walk the horse, you can't see the age of will”). His father pulled up: "Do not tell! You have the third rank, and your uncle, probably the second. ” Svidler looked back: "Well, for the second, I play by strong"!
Nevertheless, attempts to save human intellect from disgrace in chess are still ongoing. About the middle game, however, no one stutters, but there is also a debut and an endgame! It looks like this: Prohibit a computer from using directories for debuts, they are also compiled by protein chess players. This is unacceptable playing along with computers, in my opinion.
It will not help, gentlemen! The computer is quite capable of playing without a reference book. For example, our program (depending on the time on the course and speed of the computer) quite in theory played the Scottish, Italian, Spanish, Russian parties, the Queen's Gambit, the orthodox defense, Karo-Kann, the French ... only my beloved Pirc completely refused to play, even if This option came across in the debut directory (and we forbade playing it for black), and in general (unlike, say, from Rybka), she did not like the fianchetto. But this computer is capable of playing without a reference book. Is man capable of that? Oh, doubtful! Rather, I agree with the statement that several centuries of experience and the growth of opening theory from the beginning of the twentieth century have given man a considerable head start . And what a man looks like without a debut reference book, Valery Murahveri perfectly demonstrated. He suggested an idea to swap the black queen and the king, and when I programmed this thing and brought him, his enthusiastic exclamations: “Look! Look The Gambit of Muzio in his left hand! ”Was quickly replaced by much less enthusiastic:“ No, this is not the Gambit of Muzio. ” He played four or five games, but never managed to save a single one.
The last hope for the endgame remains: Making a draw in chess is much easier than winning. There are a lot of opening options that have been fully analyzed before a draw endgame. No supercomputer will win a draw from a grandmaster.
First, it is far from a fact neither one nor the other. Secondly, even in unfortunate ending directories, which have nothing to do with chess, let alone AI, Kasparov managed to “see God”! By the way, what is this “Nalimov endgame table”? What kind of Nalimov? Not Zhenya, by chance? Exactly, Zhenya: Evgeny Viktorovich Nalimov, Novosibirsk! A unique person! A very strong programmer, who from the first attempt became the champion of Russia. Rather, he took second place, but later it turned out that the winner of the tournament used Chessmaster, which he gave out as his own program. But that's not the point: Zhenya could not play chess at all! We just laughed when he literally calculated on his fingers where e2 is and where e4 is (there were no vertical and horizontal signs on the boards)! True, according to him, he tested his program and compiled the debut reference book after all some familiar KMS.
Now let's run through some theses, most of which are not true.
The debut theory in chess is very advanced - Wiki + Database. At the moment, the debut theory is almost ossified, since all major debuts have already been calculated for 20 moves ahead.
Yes, there is nowhere more advanced! I remember once we got the debut guide to the thick book titled “Encyclopedia of Chess Debuts” (one of the “crazy options” in the French defense) - some kind of strange victim, the counter-sacrifice ... they gave to our Mirage, it looked - MAT!
As far as I know, this approach to chess is outdated, the strongest modern programs are based on multi-million bases of the parties played by people.
Complete nonsense! From the same opera as the infallibility of debuts. Nonsense if only because the programs are stronger than a person - so why do they need these “multi-million bases” performed by notorious teapots?
A person looks at the board and “sees” the best move without an obvious search of options (well, maybe analyzing several moves).
Yes, damn, he does not "see"! The same Karjakin in the match with Carlsen made mistakes of the level of the second category. The same Karpov went to the option where he got mated in two moves and won. And in one of the games of the “match of titans” Karpov-Kasparov on the board for three moves in a row (!!!) was the position of simple forced gain (for most programs it takes 10-15 seconds) - one of them did not use this opportunity three times, and the second did not use the opportunity to avoid immediate death three times. The party ended in a draw.
Traditional dispute: what is more effective, accurately assess the position or achieve greater depth of search. Experience shows that too “heavy” evaluation functions are ineffective. On the other hand, a more detailed assessment that takes into account more factors usually leads to a more “beautiful” and “aggressive” game. The speed of work will mainly depend on the evaluation function.
Excuse me, WHAT “experience shows”? AND IN WHAT SIDE “will depend”? I, for example, argue that a complex, integrated assessment function speeds up the search! But in literature there is usually the opposite opinion.
The idea is used first to make a move in which there will be a capture of the opponent’s piece that made the last move. Also known is the heuristics of history or the service of the best moves.
The service of the best moves really noticeably speeds up the search. But “taking the piece that made the last move” is not used at all. There is a “exchange at a point” for this, which refers rather to the evaluation function, but not to brute force - it is too expensive!
Even now I can dig out the chess program of 30 years ago and not beat it, but this intelligence will not appear in it, but I will not lose it.
Here the slightest objection! You can neither subtract nor add what is not.
The Komodo project was launched in 2007 as a result of collaboration between programmer Don Daley and grandmaster Larry Kaufman. Unfortunately, Don Daily passed away, but Mark Lefler, the author of the chess program Now took over the further development.
Daly and Kaufman had a very interesting Rex program, and then a stronger, but much more boring Socrates. I never intersected with them, but with Lefler it happened: in London we shared 3-4 places with him (after Hiarcs and M-Chess), and in Munich at the World Blitz Championship Mirage lost to Now, having lost the rook out of the blue. Does Lefler have eyes on his forehead: “Is this some kind of program crash?” I say: "No, this is grief from the mind - the Mirage saw the mat itself and chose the lesser of the evils." It seems that everything is correct, but I didn’t see Now Mata! So I agree - it also happens like this: If a computer assesses its position as clearly losing and sees that it can’t be saved, then it chooses not the most persistent continuation, but just plays something.
By happy coincidence, there are 64 squares on a chessboard. So, if for each cell to use one bit, we can store the entire board in a 64-bit integer.
Marasmus. Even if you do not read further explanation about the "advantages of this option."
Most of the chess engines are based on the minimax search algorithm or its modification non-max.
Already tired of this "alpha beta"! Not to mention the type of nonsense "this method requires a predetermined depth." Cascade busting is not at all “first we carry out the analysis to a depth of 1, then to a depth of 2, etc.” - the depth of the search is torn! In particular, our maximum depth was limited to 21 half-shots. And the forced search is far from always “just taking and the shahs” - there are still at least attacks and dull threats, and sometimes the movement of passed pawns. And in general, I believe that none of the chess programs (at least, of the strong chess programs) use negomax.
Due to the high-quality sorting of moves and clipping, modern engines manage to achieve a branching factor below 2.
Ha ha ha! At Mirage, this ratio rarely exceeded one and a half, but it is very much. By the way, there were no “cuts”.
The computer always expects the strongest move from his point of view. He does not set incorrect traps, in the expectation that the enemy will not notice the combination or attack. People do it all the time.
Well and rightly so! People do not need incorrect traps, people need correct traps!
Deep Sot at the time of the match was the strongest electronic chess player in the world - the current world champion among computer programs.
And after the match, if I remember correctly, I took only third place, losing to Fritz, who played on the "stump" with a frequency of 90 MHz.
There is also a classic book by ex-champion M. Botvinnik, On the cybernetic goal of the game.
Did you read this book yourself? As Leonid Filatov said in a film: “Seven mistakes: three in theory and four in numerical evaluations”. I personally knew 3-4 dozen authors of chess programs, including 5-7 world champions, and affirm that not one of them has ever considered his work as an AI. However, I do not think that Botvinnik himself was so stupid as to believe in this nonsense. Most likely, he simply hung noodles on the ears of the suckers, getting very tangible material dividends. In any case, his “Pioneer” is anything but a non-chess program.
Modern programs are too far ahead, even from programs a decade ago. Over the past 15 years, about 5 or 6 generations of chess programs have changed, the algorithms have been extremely improved, and what was considered the strongest in 1997 - now every newbie will write a program for a week more simply by reading the finished articles on specialized sites. This was not due to the growth of iron power, but due to the development of algorithms.
Oh really?! Announce the entire list, please! And where is “any newcomer” hiding there? Well, turn around, son! Did you know that one of the first world champions (in my opinion, it was Ken Thompson) once said: “Over the past decades nothing new has been introduced into computer chess, except for the zero move”? I really do not agree: zero move is also not needed! Here are also the tales of the form: My students after the 2nd year, three of them, in three weeks, made these chess. Lord Just take my word for it: the chess program is a really complicated thing! And the ability to play chess is not at all synonymous with the ability to move chips.
In this implementation there is something to improve, but do not forget - everything was done in 48 hours. We also plan to open API for chess AI. When the AI becomes strong enough, we will call the real French grandmaster and the representative of Guinness to hold the most exciting chess game! We are also looking for talented JavaScript developers to handle everything described above.
Here I can only say one thing: what kind of shit will the AI be called! In general, "Petriks rule" ...
The learning process of a chess program took only 24 hours. Only the rules of the game were entered into the system - that's all. No debut libraries or batch databases. Only rules. And 24 hours the program played with itself. After 24 hours and 44 million games played with itself, AlphaZero became the strongest chess player in the history of the game. Elo AlphaZero rating is about 3500 points, although according to various data it reaches 5000.
Well or similar: In 72 hours the Giraffe neural network learned how to play chess at the level of the international master FIDE . And what, then the electricity was cut down? I would have learned further, at least to the level of a grandmaster. Personally, I never took neural networks seriously - at least, as an AI. For this can not be because there can never be! But this is a topic for another conversation.
So, listened - they decided: there is no AI in computer chess, there wasn’t, and never will be. Intelligence was also not found among protein chess players.
In conclusion, I want to offer one of the most spectacular combinations of our Mirage, which he won in the tournament game against the current world champion. At that time, none of the known programs could not find it. You can try to feed it to modern programs - for example, the same neural network, which in a day reached the superhuman quality of the game of Go, chess and Segou . If someone does this, do not take the trouble to report the result.
Black's move Win.
Calculate your strength.
And if you don’t know how to say “oink”,
Squeal, do not hesitate: "AI!" ...
S. Ya. Marshak
This topic has been constantly emerging in various discussions over the years. From the stormy, with fights and tantrums (about once a decade) to more relaxed during times "between the peaks." Now, it seems, another “uptrend” has begun. Although, basically, the same tortured theses are scolding: about the stupid Turing test, about the even more stupid "laws of robotics" Azimov, about emotions and consciousness, about chess and go, plus all kinds of horror films about Skynet. Strangely enough, there is still a lot of talk about chess, although it seemed to me that they should have stopped a long time ago. Well - this topic is closest to me, and we will begin with it.
Chess
No, I understand that chess was once associated with AI. This continued as long as the computers were playing poorly, and therefore the sapiens had the opportunity to turn up their noses and boast about their UE, which they allegedly have a “default”. But now? Yes, it is no longer interesting to play chess with a computer, it constantly wins . What about intelligence? Recognize the presence of a piece of iron? Never!
- What does the artificial intelligence? There is just going through all the options.
- The same chess and other games are still taken mainly in the "ass", i.e. brute force
- The operations that we do while playing chess are not at all like simple arithmetic. If you take a more complex task, such as pattern recognition, it will be clear that the machine is not able to perform the necessary calculations as quickly as the human brain.
- About twenty years ago it seemed to many that the creation of AI would be marked by the fact that a computer would beat a man in chess. Beaten. Where is AI?
- We have already built a computer that can beat the best chess players. But to build a computer that could understand the children's book, and not just recognize the words, is still very difficult.
- The chess program is not an AI. This is an expert system (according to a strict classification).
')
Even leaving aside the obvious nonsense about “expert systems,” it is clear that a person will come up with all new, even the most idiotic excuses, just to keep faith in the fairy tale that he has this very I: The definition of reason changes with time. Once it included the ability to multiply - crossed out, even the glands (without electronics) do it. Then they crossed out the ability to play chess, then, as they say, everywhere. The concept of mind narrows as shagreen. As it was rightly said here: No need to hang a fashionable AI label on everything that moves! And I add: and just And do not! The same Turing test assumes that the AI must descend to the intellectual level of the "examiners", who have been credited with intelligence, and lie to their eyes - and all this only so that they deign to admit that they have intelligence? Yes, a normal AI should be above these nonsense! Let us recall: billions of flocks of sheep for thousands of years believe in religious ravings of the lowest possible order, and no one doubts that they have intelligence.
What am I talking about? Oh yes, about chess. Although the people have already switched to go. Not for long, of course - until people got there in the face: But they write that computers still play very poorly in Go - there are many more combinations than in chess. Go game has always been a tough nut to AI. Skeptics argued that go programs would never win, or they would win very soon. Go is more complicated than chess by 10 ^ 100 times — that is, so many times more possible positions of stones on a standard chessboard 19 * 19 than in chess. Something quickly you, gentlemen, have forgotten a similar bleating about chess: The difficulty of analyzing a chess game is that the number of unique chess games is approaching the number 10 ^ 120 and other nonsense in other AI tasks like the Bremmerman limit.
So is a computer chess player stronger than a man? Does anyone still have doubts? He is even stronger because he is stronger, but because people have already surrendered! Kasparov said: If a computer can beat the best of the best in chess, this will mean that the computer is able to compose the best music, write the best books. I can not believe it. If a computer with a rating of 2800, that is equal to mine, is created, I myself would consider it my duty to call him for a match in order to protect the human race. No longer twitches. And Sergey Karjakin expressed himself even more clearly: In a fight against a computer, the feeling of doom does not leave you. And why did the magnificent tournament in the Hague "cars against the people", 50 to 50, wall to wall? Yes, all for the same reason: people have no chance! Although in the early stages of this competition, there were mostly amateurs playing there, and almost the only representative of the chess elite of mankind was David Bronstein, and in the end there involved a whole constellation of outstanding chess players: Vahanyan, Seyravan, sisters Polgar, Kosashvili, Christiansensen, Timoschenko and others. So no need to find out who is stronger: a person or a car - people have already answered this question! Not to mention the fact that no one has yet programmed a computer against a game against a person (which is quite simple to do). Peter Svidler led the line: There is no chance for a man to defeat a supercomputer in chess for a long time. All discussions on this topic are long over .
Svidler, by the way, played with our program during the times of the Kremlin Stars. Although some grandmasters (Short, Nikolic) flatly refused to play with the computer "in public" (in the lobby of the CBC), others (Svidler, Drahomaretsky) were what are called "without complexes". During one of the games, an amusing incident occurred: a boy of ten began to suggest (something like “walk the horse, you can't see the age of will”). His father pulled up: "Do not tell! You have the third rank, and your uncle, probably the second. ” Svidler looked back: "Well, for the second, I play by strong"!
Nevertheless, attempts to save human intellect from disgrace in chess are still ongoing. About the middle game, however, no one stutters, but there is also a debut and an endgame! It looks like this: Prohibit a computer from using directories for debuts, they are also compiled by protein chess players. This is unacceptable playing along with computers, in my opinion.
It will not help, gentlemen! The computer is quite capable of playing without a reference book. For example, our program (depending on the time on the course and speed of the computer) quite in theory played the Scottish, Italian, Spanish, Russian parties, the Queen's Gambit, the orthodox defense, Karo-Kann, the French ... only my beloved Pirc completely refused to play, even if This option came across in the debut directory (and we forbade playing it for black), and in general (unlike, say, from Rybka), she did not like the fianchetto. But this computer is capable of playing without a reference book. Is man capable of that? Oh, doubtful! Rather, I agree with the statement that several centuries of experience and the growth of opening theory from the beginning of the twentieth century have given man a considerable head start . And what a man looks like without a debut reference book, Valery Murahveri perfectly demonstrated. He suggested an idea to swap the black queen and the king, and when I programmed this thing and brought him, his enthusiastic exclamations: “Look! Look The Gambit of Muzio in his left hand! ”Was quickly replaced by much less enthusiastic:“ No, this is not the Gambit of Muzio. ” He played four or five games, but never managed to save a single one.
The last hope for the endgame remains: Making a draw in chess is much easier than winning. There are a lot of opening options that have been fully analyzed before a draw endgame. No supercomputer will win a draw from a grandmaster.
First, it is far from a fact neither one nor the other. Secondly, even in unfortunate ending directories, which have nothing to do with chess, let alone AI, Kasparov managed to “see God”! By the way, what is this “Nalimov endgame table”? What kind of Nalimov? Not Zhenya, by chance? Exactly, Zhenya: Evgeny Viktorovich Nalimov, Novosibirsk! A unique person! A very strong programmer, who from the first attempt became the champion of Russia. Rather, he took second place, but later it turned out that the winner of the tournament used Chessmaster, which he gave out as his own program. But that's not the point: Zhenya could not play chess at all! We just laughed when he literally calculated on his fingers where e2 is and where e4 is (there were no vertical and horizontal signs on the boards)! True, according to him, he tested his program and compiled the debut reference book after all some familiar KMS.
Now let's run through some theses, most of which are not true.
The debut theory in chess is very advanced - Wiki + Database. At the moment, the debut theory is almost ossified, since all major debuts have already been calculated for 20 moves ahead.
Yes, there is nowhere more advanced! I remember once we got the debut guide to the thick book titled “Encyclopedia of Chess Debuts” (one of the “crazy options” in the French defense) - some kind of strange victim, the counter-sacrifice ... they gave to our Mirage, it looked - MAT!
As far as I know, this approach to chess is outdated, the strongest modern programs are based on multi-million bases of the parties played by people.
Complete nonsense! From the same opera as the infallibility of debuts. Nonsense if only because the programs are stronger than a person - so why do they need these “multi-million bases” performed by notorious teapots?
A person looks at the board and “sees” the best move without an obvious search of options (well, maybe analyzing several moves).
Yes, damn, he does not "see"! The same Karjakin in the match with Carlsen made mistakes of the level of the second category. The same Karpov went to the option where he got mated in two moves and won. And in one of the games of the “match of titans” Karpov-Kasparov on the board for three moves in a row (!!!) was the position of simple forced gain (for most programs it takes 10-15 seconds) - one of them did not use this opportunity three times, and the second did not use the opportunity to avoid immediate death three times. The party ended in a draw.
Traditional dispute: what is more effective, accurately assess the position or achieve greater depth of search. Experience shows that too “heavy” evaluation functions are ineffective. On the other hand, a more detailed assessment that takes into account more factors usually leads to a more “beautiful” and “aggressive” game. The speed of work will mainly depend on the evaluation function.
Excuse me, WHAT “experience shows”? AND IN WHAT SIDE “will depend”? I, for example, argue that a complex, integrated assessment function speeds up the search! But in literature there is usually the opposite opinion.
The idea is used first to make a move in which there will be a capture of the opponent’s piece that made the last move. Also known is the heuristics of history or the service of the best moves.
The service of the best moves really noticeably speeds up the search. But “taking the piece that made the last move” is not used at all. There is a “exchange at a point” for this, which refers rather to the evaluation function, but not to brute force - it is too expensive!
Even now I can dig out the chess program of 30 years ago and not beat it, but this intelligence will not appear in it, but I will not lose it.
Here the slightest objection! You can neither subtract nor add what is not.
The Komodo project was launched in 2007 as a result of collaboration between programmer Don Daley and grandmaster Larry Kaufman. Unfortunately, Don Daily passed away, but Mark Lefler, the author of the chess program Now took over the further development.
Daly and Kaufman had a very interesting Rex program, and then a stronger, but much more boring Socrates. I never intersected with them, but with Lefler it happened: in London we shared 3-4 places with him (after Hiarcs and M-Chess), and in Munich at the World Blitz Championship Mirage lost to Now, having lost the rook out of the blue. Does Lefler have eyes on his forehead: “Is this some kind of program crash?” I say: "No, this is grief from the mind - the Mirage saw the mat itself and chose the lesser of the evils." It seems that everything is correct, but I didn’t see Now Mata! So I agree - it also happens like this: If a computer assesses its position as clearly losing and sees that it can’t be saved, then it chooses not the most persistent continuation, but just plays something.
By happy coincidence, there are 64 squares on a chessboard. So, if for each cell to use one bit, we can store the entire board in a 64-bit integer.
Marasmus. Even if you do not read further explanation about the "advantages of this option."
Most of the chess engines are based on the minimax search algorithm or its modification non-max.
Already tired of this "alpha beta"! Not to mention the type of nonsense "this method requires a predetermined depth." Cascade busting is not at all “first we carry out the analysis to a depth of 1, then to a depth of 2, etc.” - the depth of the search is torn! In particular, our maximum depth was limited to 21 half-shots. And the forced search is far from always “just taking and the shahs” - there are still at least attacks and dull threats, and sometimes the movement of passed pawns. And in general, I believe that none of the chess programs (at least, of the strong chess programs) use negomax.
Due to the high-quality sorting of moves and clipping, modern engines manage to achieve a branching factor below 2.
Ha ha ha! At Mirage, this ratio rarely exceeded one and a half, but it is very much. By the way, there were no “cuts”.
The computer always expects the strongest move from his point of view. He does not set incorrect traps, in the expectation that the enemy will not notice the combination or attack. People do it all the time.
Well and rightly so! People do not need incorrect traps, people need correct traps!
Deep Sot at the time of the match was the strongest electronic chess player in the world - the current world champion among computer programs.
And after the match, if I remember correctly, I took only third place, losing to Fritz, who played on the "stump" with a frequency of 90 MHz.
There is also a classic book by ex-champion M. Botvinnik, On the cybernetic goal of the game.
Did you read this book yourself? As Leonid Filatov said in a film: “Seven mistakes: three in theory and four in numerical evaluations”. I personally knew 3-4 dozen authors of chess programs, including 5-7 world champions, and affirm that not one of them has ever considered his work as an AI. However, I do not think that Botvinnik himself was so stupid as to believe in this nonsense. Most likely, he simply hung noodles on the ears of the suckers, getting very tangible material dividends. In any case, his “Pioneer” is anything but a non-chess program.
Modern programs are too far ahead, even from programs a decade ago. Over the past 15 years, about 5 or 6 generations of chess programs have changed, the algorithms have been extremely improved, and what was considered the strongest in 1997 - now every newbie will write a program for a week more simply by reading the finished articles on specialized sites. This was not due to the growth of iron power, but due to the development of algorithms.
Oh really?! Announce the entire list, please! And where is “any newcomer” hiding there? Well, turn around, son! Did you know that one of the first world champions (in my opinion, it was Ken Thompson) once said: “Over the past decades nothing new has been introduced into computer chess, except for the zero move”? I really do not agree: zero move is also not needed! Here are also the tales of the form: My students after the 2nd year, three of them, in three weeks, made these chess. Lord Just take my word for it: the chess program is a really complicated thing! And the ability to play chess is not at all synonymous with the ability to move chips.
In this implementation there is something to improve, but do not forget - everything was done in 48 hours. We also plan to open API for chess AI. When the AI becomes strong enough, we will call the real French grandmaster and the representative of Guinness to hold the most exciting chess game! We are also looking for talented JavaScript developers to handle everything described above.
Here I can only say one thing: what kind of shit will the AI be called! In general, "Petriks rule" ...
The learning process of a chess program took only 24 hours. Only the rules of the game were entered into the system - that's all. No debut libraries or batch databases. Only rules. And 24 hours the program played with itself. After 24 hours and 44 million games played with itself, AlphaZero became the strongest chess player in the history of the game. Elo AlphaZero rating is about 3500 points, although according to various data it reaches 5000.
Well or similar: In 72 hours the Giraffe neural network learned how to play chess at the level of the international master FIDE . And what, then the electricity was cut down? I would have learned further, at least to the level of a grandmaster. Personally, I never took neural networks seriously - at least, as an AI. For this can not be because there can never be! But this is a topic for another conversation.
So, listened - they decided: there is no AI in computer chess, there wasn’t, and never will be. Intelligence was also not found among protein chess players.
In conclusion, I want to offer one of the most spectacular combinations of our Mirage, which he won in the tournament game against the current world champion. At that time, none of the known programs could not find it. You can try to feed it to modern programs - for example, the same neural network, which in a day reached the superhuman quality of the game of Go, chess and Segou . If someone does this, do not take the trouble to report the result.
Black's move Win.
Source: https://habr.com/ru/post/448398/
All Articles