1

Topic: The machine intelligence became the best chess player of the world for 4

AI from Google in some hours itself learned to play chess and  and won "champions" https://rb.ru/news/alpha-zero/Machine intelligence there was the best chess player of the world for 4 hours https://vivalacloud.ru/2017/12/alphazero-chess/ One more revolution. The algorithm is developed, allowing to play chess at space level http://chess-news.ru/node/24010 DeepMind presented new algorithm, for  hours mastered chess https://hitech.newsru.com/article/06dec2017/alphazero for the present did not read, but briefly "the white begin and benefit" - it appeared true (in that sense that at white there is a certain advantage, statistically on an example among the strongest programs AlphaZero and Stockfish), "Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm" (the scientific article if to whom it is interesting) https://arxiv.org/pdf/1712.01815.pdf

2

Re: The machine intelligence became the best chess player of the world for 4

It seems, here there is especially nothing to rejoice.

3

Re: The machine intelligence became the best chess player of the world for 4

Hello, De-Bill, you wrote: DB> It seems, here there is especially nothing to rejoice. Why? I so understand that soon such approach can be applied more actively to the theorem proving. Is not present?

4

Re: The machine intelligence became the best chess player of the world for 4

Hello, itmanager85, you wrote: I> AI from Google in some hours itself learned to play chess and  and won "champions" Perfectly. Now it it is necessary to teach to select tariff plans of cellular operators.

5

Re: The machine intelligence became the best chess player of the world for 4

Hello, kov_serg, you wrote: I>> AI from Google in some hours itself learned to play chess and  and won "champions" _> Perfectly. Now it it is necessary to teach to select tariff plans of cellular operators. Cellular operators will invent new tariff plans. AI on AI AI sends on errands. The main thing to pump up energy it is a closed circuit.

6

Re: The machine intelligence became the best chess player of the world for 4

I> AI from Google in some hours itself learned to play chess and  and won "champions" I> https://rb.ru/news/alpha-zero/Unlike the previous programs DeepMind, AlphaZero formed as the algorithm, capable to learn at once to several tasks-games, instead of one. For this purpose algorithm did not train to win, and gave only basic knowledge of game rules. AlphaZero then played with itself and independently worked out tactics. I transport on  language: They  rules, and then launched search to find all successful variants. AI?

7

Re: The machine intelligence became the best chess player of the world for 4

Hello, namespace, you wrote: N> N> Unlike the previous programs DeepMind, AlphaZero formed as the algorithm, capable to learn at once to several tasks-games, instead of one. For this purpose algorithm did not train to win, and gave only basic knowledge of game rules. AlphaZero then played with itself and independently worked out tactics. N> I transport on  language: They  rules, and then launched search to find all successful variants. N> AI? N> you are valid facepalm certainly in course what to sort out all variants in chess up to the end - it is impossible, and there is a problem of an objective estimation " positions" so that in absence of the unambiguous definiteness to make their most exact estimation? And so, AlphaZero searches just 80 thousand positions per second in chess and 40 thousand in shogi, compared to 70 million for Stockfish and 35 million for Elmo. I.e. Having computing productivity of depth of courses almost in 1000 times less (at AlphaZero against Stockfish), AI nevertheless thanks to more exact estimations of "the relative positions" benefits against is deeper-reboric variant in the form of Stockfish. At some forums specify that played pier Stockfish not at the full capacity since "64 threads" - twice it is less from capacity. (  it is a comment) "with 40ms thinking time" is twice more from the minimum the Full power is thinking 5000! And also that Speak and  selected a little (1) for such amount of flows. To include on the declared reference capacity, there it is necessary terabyte and that Stockfish could not use debut basis. And without the debut book,  bases, they are forced from zero to count a position, unlike AlphaZero. But and AlphaZero studied only 4 hours, and can after all and month and if hallows does not suffice even after that (what I doubt) - that could be fed also in addition "the debut book,  bases" but it is clearly finite that here counter AlphaZero in that that it (in few hours) independently reaches results of the best than all mankind with all chess achievements, including the best programs written to it. Well and so to convergence, AlphaZero and the previous AlphaGo Zero used a single machine with 4 TPUs. And at training, Training proceeded for 700,000 steps (mini-batches of size 4,096) starting from randomly initialised parameters, using 5,000 first-generation TPUs (15) to generate self-play games and 64 second-generation TPUs to train the neural networks.

8

Re: The machine intelligence became the best chess player of the world for 4

Hello, namespace, you wrote: N> I Transport on  language: They  rules, and then launched search to find all successful variants. N> AI? They used not all variants, and Monte Carlo Tree Search and achieved that that training activity became converges.

9

Re: The machine intelligence became the best chess player of the world for 4

DB>> It seems, here there is especially nothing to rejoice. N> why? N> I so understand that soon such approach can be applied more actively to the theorem proving. Is not present? Soon this approach (or modified) will be we apply to all.

10

Re: The machine intelligence became the best chess player of the world for 4

Hello, De-Bill, you wrote: DB> Soon this approach (or modified) will be applicable to all. Allow not to talk in riddles. What causes in you anxiety?

11

Re: The machine intelligence became the best chess player of the world for 4

1) Rules change in due course under the known law for a neural network (probably special case of static rules) 2) Rules change in due course on not to the known law for a neural network 3) Rules change in due course on not to partially casual law known for a neural network

12

Re: The machine intelligence became the best chess player of the world for 4

Hello, Nuzhny, you wrote: N> I so understand that soon such approach can be applied more actively to the theorem proving. Is not present? Theorems and without neural networks are perfectly proved. Wong's algorithm and other. (We at school transited).

13

Re: The machine intelligence became the best chess player of the world for 4

Hello, Nuzhny, you wrote: DB>> Soon this approach (or modified) will be applicable to all. N> allow not to talk in riddles. What causes in you anxiety? I wanted to result here the citation from Green , but in a direct type it is inappropriate, the poet I paraphrase it: "you will be, the programmer to put a potato?".

14

Re: The machine intelligence became the best chess player of the world for 4

Hello, itmanager85, you wrote: I> you certainly in course what to sort out all variants in chess up to the end - are impossible, and there is a problem of an objective estimation " positions" so that in absence of the unambiguous definiteness to make their most exact estimation? I> i.e. having computing productivity of depth of courses almost in 1000 times less (at AlphaZero against Stockfish), AI nevertheless thanks to more exact estimations of "the relative positions" benefits against is deeper-reboric variant in the form of Stockfish. I> but and AlphaZero studied only 4 hours, and can after all and month and if hallows does not suffice even after that (what I doubt) - that could be fed also in addition "the debut book,  bases" Well generally the intelligence is a correctness, instead of speed. Therefore I am not surprised to that  in certain cases benefit. I suspect, AlphaZero studied only 4 hours just because on the fifth hour anything new does not happen. After a certain limit it gives not so important, what estimation of a position 17.748 or 17.749. It does courses just the same and it becomes cleverer"not. And to make Stockfish it is cleverer" to give simply enough to it of more resources. And what to do with  - it is not clear. It is impossible for it to feed "the debut book,  bases". If it was possible, would feed and did not buzz. Therefore  yes, this time Stockfish struggled with the bound hands.

15

Re: The machine intelligence became the best chess player of the world for 4

I> AI from Google in some hours itself learned to play chess and  and won "champions" As far as I understand, the most interesting there "the same algorithm settings, network architecture, and hyper-parameters were used for all three games". That is the same grid mastered three games.

16

Re: The machine intelligence became the best chess player of the world for 4

Hello, Glory, you wrote: I wanted to result here the citation from Green , but in a direct type it is inappropriate, the poet I paraphrase it: "you will be, the programmer to put a potato?". Why is not present? For example, I do not have desire to consider on a piece of paper or the adding machine, to use tables of Bradisa etc.

17

Re: The machine intelligence became the best chess player of the world for 4

Hello, igor-booch, you wrote: IB> 1) Rules change in due course under the known law for a neural network (probably special case of static rules) IB> 2) Rules change in due course on not to the known law for neural network IB> 3) Rules change in due course on not to partially casual law known for a neural network Not. The following stage here

18

Re: The machine intelligence became the best chess player of the world for 4

Hello, alpha21264, you wrote: A> I suspect, AlphaZero studied only 4 hours just because on the fifth hour anything new does not happen. A> after a certain limit it gives not so important, what estimation of a position 17.748 or 17.749. A> it does Courses just the same and it becomes cleverer"not. Interesting theory A> And to make Stockfish it is cleverer" to give simply enough to it of more resources. A> and what to do with  - it is not clear. A> it is impossible for it to feed "the debut book,  bases". A> If it was possible, would feed and did not buzz. In AlphaGo fed and AlphaZero which generally fed nothing - it  won. A> Therefore  yes, this time Stockfish struggled with the bound hands.

19

Re: The machine intelligence became the best chess player of the world for 4

Hello, itmanager85, you wrote: I> Hello, alpha21264, you wrote: A>> I suspect, AlphaZero studied only 4 hours just because on the fifth hour anything new does not happen. A>> after a certain limit it gives not so important, what estimation of a position 17.748 or 17.749. A>> it does Courses just the same and it becomes cleverer"not. I> the interesting theory Obvious. On Google Translate look. Why it becomes cleverer not? Here . A>> And what to do with  - it is not clear. A>> it is impossible for it to feed" the debut book,  bases ". A>> If it was possible, would feed and did not buzz. I> in AlphaGo fed and AlphaZero which generally fed nothing - it  won. Where? You told that: Stockfish could not use debut basis. And without the debut book,  bases, they are forced from zero to count a position, unlike AlphaZero. On mine, it makes it clear that on fair AlphaZero cannot struggle against Stockfish.

20

Re: The machine intelligence became the best chess player of the world for 4

Hello, alpha21264, you wrote: A> Obvious. Can because course time restricted? A> on Google Translate look. Why it becomes cleverer not? On me so transfer in Google  on the order became more qualitative (after implementation  for transfer) I>> in AlphaGo fed and AlphaZero which generally fed nothing - it  won. A> Where? What where? AlphaGo it is AI for Go, trained including on human batches. And AlphaZero is AI trained only on game with itself. A> You told that: A> A> Stockfish could not use debut basis. A> and without the debut book,  bases, they are forced from zero to count a position, unlike AlphaZero. A> On mine, it makes it clear that on fair AlphaZero cannot struggle against Stockfish.  on water  though certainly causes certain interest the further succession of events

21

Re: The machine intelligence became the best chess player of the world for 4

I> you are valid facepalm certainly in course what to sort out all variants in chess up to the end - it is impossible, and there is a problem of an objective estimation " positions" so that in absence of the unambiguous definiteness to make their most exact estimation? I did not write about stupid search of all courses. I think, the concept of etudes is not necessary . And if in this case the fitting algorithm of etudes won algorithm of simple search one lines it is algorithms all easier, and to AI they have no relation. From the point of view of the machine, chess - the extremely simple task with a scanty amount of rules. Intelligence of level of any cockroach. A problem of present AI in understanding/acquisition of new rules, but these rules such amount that adding their such a method (manually encoding), we on I.Q. of the three-year child even this century do not quit. For comparing if in autonavigators to build in possibility of saving of all possible routes (on the basis of rules of traffic regulations and a map of the city), then as it can be checked up what of them more '  '.

22

Re: The machine intelligence became the best chess player of the world for 4

Hello, itmanager85, you wrote: I> at some forums specify that played pier Stockfish not at the full capacity since I> I> "with 40ms thinking time" is twice more from the minimum the Full power is thinking 5000! Not so - I looked at it to dock and about a miracle that I saw there Figure 2 shows the scalability of each player with respect to thinking time, measured on an Elo scale, relative to Stockfish or Elmo with 40ms thinking time. The schedule on Figure 2 (on 7 page) - on it it is visible that since a floor of second (on a course) AlphaZero starts to soil StockFish - and the more time for a course (the schedule up to 10 seconds) - the more lift-off at AlphaZero. We evaluated the fully trained instances of AlphaZero against Stockfish, Elmo and the previous version of AlphaGo Zero (trained for 3 days) in chess, shogi and Go respectively, playing 100 game matches at tournament time controls of one minute per move. https://arxiv.org/pdf/1712.01815.pdf Table 1. (5th page). . Each program was given 1 minute of thinking time per move. And on each course in competitions on 100 batches - minute was given generally. I> also that I> I> Stockfish could not use debut basis. I> and without the debut book,  bases, they are forced from zero to count a position, unlike AlphaZero. It too , who that  at a forum.

23

Re: The machine intelligence became the best chess player of the world for 4

Hello, De-Bill, you wrote: DB> Soon this approach (or modified) will be applicable to all. Glory to robots! To kill all !... <<RSDN@Home 1.0.0 alpha 5 rev. 0>>

24

Re: The machine intelligence became the best chess player of the world for 4

Why they select games instead of we tell train its compilations and  programs? Hello, igor-booch, you wrote: IB> 1) Rules change in due course under the known law for a neural network (probably special case of static rules) IB> 2) Rules change in due course on not to the known law for neural network IB> 3) Rules change in due course on not to partially casual law known for a neural network

25

Re: The machine intelligence became the best chess player of the world for 4

S> Why they select games instead of we tell train its compilations and  programs? I will try to shift roughly the task of optimization of programs on chess. We admit arrangement of chess on a board it is the program. It is a lot of variants of arrangement. Exactly as well as programs can be much. Courses it is program optimization. Now a neural network trained to benefit from one initial position (initial arrangement of chess according to rules). Whether not the same neural network can turn on in the game when game is played half and current arrangement of chess differs from the initial? For example, to solve a chess etude? If is not present, it is one more following stage.