Chess

12 réponses [Dernière contribution]
Masaru Suzuqi
Hors ligne
A rejoint: 06/06/2018

e4

Masaru Suzuqi
Hors ligne
A rejoint: 06/06/2018

Roman ping.

nadebula.1984
Hors ligne
A rejoint: 05/01/2018

I haven't played board games (go, chess, shogi, etc.) any more since the release of AlphaGo.

Also I trashed my entire collection of board game tutorials. Everything becomes meaningless.

zapper (non vérifié)
zapper

I have played chess, I am what you call an intermediate player. But tell me, what is AlphaGo like?

Magic Banana

I am a member!

Hors ligne
A rejoint: 07/24/2010

AlphaZero is even more impressive than AlphaGo. Only taking the rules at input (no opening book or endgame table), it mastered chess, shogi and go. And, yes, it defeated AlphaGo. Both algorithms are convolutional neural networks trained by reinforcement learning, where each game is played by running a Monte-Carlo Tree Search: https://science.sciencemag.org/content/362/6419/1140

And, to answer Masaru Suzuqi, it does require capital to train AlphaGo or AlphaZero. Google develops and uses its own circuits dedicated to training neural networks: https://en.wikipedia.org/wiki/Tensor_processing_unit

In this way, it took AlphaZero 13 days to master go, rather than years.

nadebula.1984
Hors ligne
A rejoint: 05/01/2018

I heard that Facebook made its ELF go engine somehow "open source". If it's permissively licensed, I wonder whether we could re-license it under GPL v3 or later and then apply its algorithms to GnuGo. Even if the "search depth" is set to (as low as) making 1 move per minute on an ordinary legacy notebook (like X60), it would still be strong enough for any amateur go players.

zapper (non vérifié)
zapper

I tend to doubt, facebook would do that, but if they did, it would probably be mit licensed or apache... :/

More likely the latter...

They would probably do some sneaky crap so that it would depend on non-free stuff. (like Google does. ;P)

Masaru Suzuqi
Hors ligne
A rejoint: 06/06/2018

I do not understand almost at all but it seems that after this move, an AI for analysis showed every square of the board as a candidate move. Then that tsuke was said as a move that breaks even a computer.

The 43 masters challenge, the fifth match, 井山裕太(Iyama Yuta)名人 vs 張栩(Cho U)九段. Oct. 15-16, 2018

PS: The AI seems to be ELF.

37F15208-4AA5-4848-AF79-3B85E17A7300.jpeg
Masaru Suzuqi
Hors ligne
A rejoint: 06/06/2018

Why could not the free software world make an AI like DQN before Google achieved? It seems that that was an ideal opportunity for the free software world to show the superiority of ideology over the superiority of capital. Some of you would have known that if an AI would defeat a professional go player, it must have a valuable impact. It does not seem to cost so much. It seems that it just needed an original new idea/method rather than capital. Because the stones of go have not class unlike the pieces of chess. It is reterritorialization and deterritorialization like G.Deleuze says. After all it played a part in the superiority of capital. It seems that Deleuze liked go. I think female go players might be able to have better fights with Alphago than male go players because of the nature of go. A female go player might make "the move of God" against the AI (not a move though). It seems that Google did not think such a possibility. By the way, you are supposed to say Roman pong.

Masaru Suzuqi
Hors ligne
A rejoint: 06/06/2018

There seem to be a lot of such opportunities like Alphago, though they might not be able to have as much impact as it. I think that we should be able to secure manpower by actively recruiting especially students (by facebook etc). They basically lack experience, so there is a risk of especially instability, but they have physical strength and could have passion. Those would make up for the defects. I don't think that there is no such endeavor in this world though... if you think that students are basically idiots who do not understand the sublime purpose, you are underestimating. Even if you are right, the prime mover of revolution is always young people. If you deign some things they want to know, they desire their freedom of their free will, passionately. Okay, you serve. I don't mind.

Masaru Suzuqi
Hors ligne
A rejoint: 06/06/2018

OK, I would look forward to watching a poor developer or developers pwning Google's AI thoroughly. How fun it is. Roman ping poong!

Masaru Suzuqi
Hors ligne
A rejoint: 06/06/2018

I checked the two scores of the fights against Kasparov that Deep Blue made both of the first moves on e4. Perhaps even modern AIs would make e4 first with a good probability. So, if an AI specializes in a3, or an AI specializes in strategies such as returning a knight once to b1 or b8, the chances that the poor AI beat a strong AI might increase. If the algorithm like AlphaGo learns chess by playing with itself, it seems that it is a low possibility that the algorithm learns a tactic that starts with a3 as much as it learns a tactic that starts with e4. It might lack the experience of Anderssen's opening compared to King's pawn opening. They would make a move by probability, I guess. After they learned that a3 is a weak move, they might not learn derivative of a3 anymore. Besides the opponent, it is itself, would have the same preference that is based on percentage. I guess that a modern chess AI also almost always selects e4 first. I mean, it seems that there is always an unique way and in the unique way, maybe there is sublime something that is beyond calculation that is between a rock and a being (I am not sure if a rock is a being, though) and the possibility that a poor man will beat Google's AI that MB denied. However, I think that generally, the move of God is born in the royal road, currently e4. But there might be possibility that a3 becomes the standard of the first move. By the way, even Deep Blue could calculate 2 hundred million moves per second. I wonder if they AIs have not finished calculation of all the moves of chess yet.

Masaru Suzuqi
Hors ligne
A rejoint: 06/06/2018

It seems that only people who have understood some degree of H.Bergson's philosophy can have the image what I will state now, though. (It does not seem that only having read or having studied it is always sufficient to understand his philosophy. Because, again, I have not seen a normal doctor or master or whatever who are supposed to be professionals of philosophy because they have the diplomas that seem to be useful to make loans to buy a mansion or Mercedes or something and also who could have a normal argument about philosophy with me. They always run away from the arguments. But there are 9 or 10 years olds French kids who can understand what Bergson is saying without the diploma unfortunately)
I was thinking about how many moves chess has. zetta? yotta? Then I noticed when I had thought about something between a number and a number, I was always thinking about supposed smallest numbers. Was it the other way around? I should have thought about the largest number? or the largest numbers, too? There seems to be a clue to our origin there too. Computers can have preference? And I also noticed that their (our enemies's) moves are at the almost opposite end of the efficiency of the computers. They move by.... stubbornness? rather than preference. There would be one of the decisive differences between professional board games players and the idiots (our enemies). Can we call a fickle move of an electron which affected a move which the AI took in between perfect 50-50 a preference?