| AlphaZero - Stockfish (2017) |
|
On December 4th, 2017, Google Headquarters in London applied their DeepMind AI project to the game of chess. The event was more of an experiment than a chess exhibition, and the results are groundbreaking in both the fields of computing and chess. Rather than relying on the classic "alpha-beta algorithm" common to conventional chess software, AlphaZero uses a deep neural network and is trained solely by reinforcement learning from games of self-play. It scans only 80,000 positions per second compared to Stockfish's 70 million. AlphaZero played Stockfish 100 games, winning 28 and drawing the rest.(1) A subset of the match, 10 games that AlphaZero won, was released to the public. (1) Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm - https://arxiv.org/pdf/1712.01815.pdf
|
|
|
|
< Earlier Kibitzing · PAGE 11 OF 12 ·
Later Kibitzing> |
Dec-26-17
 | | AylerKupp: I agree with most of your theories.
1. AlphaZero is definitely just a step in the journey. What I'm particularly excited about is the self-learning aspect of it, it's learning does not use human experiences or "prejudices" to teach itself how to achieve its goals. That's what AlphaGo Zero did. Therefore there is an opportunity for it to come up with approaches different than anything seen before. 2. There may be limits to self-teaching technology (like everything else). But that is just the current state of technology. A long time ago the top chess engines either ran on supercomputers (Cray-1) or had custom hardware to assist it (DeepBlue). But today's chess engines running on very reasonably priced PCs will defeat those older engines easily and convincingly. 3. Yes, software takes years to reach peak performance and many improvements are made over time. AlphaZero is no exception and it will continue to evolve. But whether it enters the market or not I think is questionable. What would Google have to gain? Today's top commercial engines are all less than US $ 100, and the market is fairly limited. The parent company of Google, Alphabet, is currently the world's most valuable company, currently valued at almost $ 500 B. Why would such a company try to market a relatively low-cost, low-volume product? Instead, I think that AlphaZero has shown it's worth many times over for its publicity value. And it is possible that its technology will be incorporated in other Google products. But game playing is a relatively simply endeavor since it has a fairly small set of fixed rules, and it’s approach may not be transferable to real world problems. We'll see. 4. Here we don't agree. Sure, the configuration of Stockfish used in the exhibition took the beating of a lifetime, but not nearly as bad as Alpha Go took from AlphaGo Zero. But, as I've said many times before, without knowledge of Stockfish's configuration used in the exhibition, both hardware and software, whether the results reflect the real relative prowess of the two engines is certainly open to question, at least as far as I'm concerned. So neither is its reputation in tatters nor is its future at stake. |
|
| Dec-26-17 | | LameJokes55: 1 Engine: I agree. This market represents at most a few bread crumbs for google. It will probably save itself the trouble. 2 Hardware: Although, it is accepted that the difference of hardware could have noticeable impact on the result, it is not the bone of contention for average chess fan. It's possible for SF to load itself on super computer. Conversely, it could be equally easier for AZ to load on PC. These two steps do not require path-breaking invention. 3 Software: The subject of real interest revolves around this interesting aspect. Is self-teaching program better than all it's predecessors (chess engines)? Just 8 hours of self-teaching produces such a startling result, what it could do with months or years of teaching? Is it possible for SF, Houdini or Comodo to acquire self-teaching technology? Will self-teaching completely revolutionize the way humans play chess? |
|
| Dec-26-17 | | devere: <LameJokes55: Theory 4: Why was SF along for a ride?> Were they given any choice? SF is free open source software, and Google utilized it for their own commercial purposes. I suppose that the main developers of Stockfish are barred from suing Google, but nothing requires the rest of us to be gullible fools. Since Google made significant modifications to the published version of Stockfish, let's henceforth refer to the match between AlphaZero and Googlefish, the modified form of Stockfish created by Google's developers. |
|
| Dec-26-17 | | markz: <LameJokes55:
conversely, it could be equally easier for AZ to load on PC>The real problem of running A0 on a normal PC is speed. Will be very very slow. A0 need at least hours or days to find a good move on PC. This is the real reason they ran A0 on a supercomputer in the match with SF. I call them cheating. <Just 8 hours of self-teaching produces such a startling result> Playing under unfair match conditions, the result is meaningless. |
|
| Dec-26-17 | | LameJokes55: <devere:Were they given any choice? SF is free open source software, and Google utilized it for their own commercial purposes.> I stand corrected. I was unaware of this aspect. All along, I thought the promoters of SF were party to the arrangement. Poor SF! |
|
| Dec-26-17 | | LameJokes55: <markz:The real problem of running A0 on a normal PC is speed.> I mean, may be, not today. A distant date in the future. Who knows what is possible, then? |
|
| Dec-26-17 | | Marmot PFL: <Are computers eternally doomed to do boring binary stuff forever or could they really become creative some day and exercise human-type judgement (as is needed for sophisticated translation (which contrary to popular opinion is really incredibly difficult), creative writing, android-type imitation of human behavior, etc.)> Many video soundtracks are evidently being written by computers. Why hire a composer or pay for the rights to use music when a computer can do the job for pennies on the dollar? https://www.npr.org/sections/alltec... |
|
| Dec-26-17 | | Marmot PFL: So far machine music isn't very impressive, took about 15 seconds to tell the real from the imitation Vivaldi. But in years to come it will no doubt be much harder. |
|
| Dec-26-17 | | ChessHigherCat: <Marmot PFL> Thanks for the link, that's interesting. I remember when recording studios finally succeeding in tricking concert pianists into thinking that they were listening to a real piano (now that was a successful publicity stunt!). The interpretation of classical music is so subtle I think computers would turn it into elevator music, but they might come up with some interesting compositions, so we can finally let them express what it's like to be a computer :-) |
|
Dec-26-17
| | AylerKupp: <<LameJokes55> Conversely, it could be equally easier for AZ to load on PC.> Given that AlphaZero uses specialized hardware (Tensor Processing Units or TPUs), and a proprietary one at that, I doubt that there will be PCs available with these anytime in the near future. This article describes Google's <first> generation TPUs and provides a performance comparison between CPUs, GPUs, and TPUs: https://cloud.google.com/blog/big-d.... Specifically it indicates that a CPU as used in a PC is capable of "a few" operations per cycle and Google's <first> generation TPU is capable of hundreds of thousands, up to 128K operations per second. That makes Google's <first> generation TPU probably about 60,000 times faster than the CPU in a PC (although, of course, the speeds of these vary somewhat, but not more than a factor of 2X – 3X. And for this kind of application the article indicates that the <first> generation TPU provides 15X – 30X more throughput than contemporary CPUs and GPUs, with GPUs in the 15X range and CPUs in the 30K range. And that was Google's <first> generation TPU. AlphaZero used their <second> generation TPU. While the first generation provided for fixed arithmetic operations only, their second generation also provided for floating point operations, and incorporated a faster memory bus for improved efficiency, to approximately 45 TFLOPS. As comparison of the first and second generation TPUs can be found in https://en.wikipedia.org/wiki/Tenso.... Unfortunately, it's not easy to determine the equivalance between TFLOPS and plain old FLOPS. AlphaZero apparently used a single machine with 4 2nd generation TPUs, or 180 TFLOPS. If Stockfish ran on a system with 64 threads as indicated, then it's best configuration would involve 64 cores. If we very conservatively equate TFLOPS with FLOPS and assume that the 2nd generation TPUs are no faster than first generation TPUs, then the AlphaZero configuration used in the exhibition was at least 30,000 * (64 / 4 ) = 1,875 faster than the configuration used for Stockfish (which was crippled by not being allowed to use its opening book and not supported by tablebases). So AlphaZero had very conservatively almost 1900X the computing power than a crippled Stockfish and it was <only> able to win 28 out of 100 games? Doesn't seem to me like a very convincing demonstration of comparable capabilities using current technology. If would be interesting to see what the results would have been if AlphaZero used only 1 TPU so that it was "only" about 500X faster than the system used by a non-crippled Stockfish. |
|
| Dec-26-17 | | Marmot PFL: Unlike chess, bridge is an incomplete information game where teamwork is also very important.
<The question whether bridge-playing programs will reach world-class level in the foreseeable future is not easy to answer. Computer bridge has not attracted an amount of interest anywhere near to that of computer chess. On the other hand, researchers working in the field have accomplished most of the current progress in the last decade.>
https://en.wikipedia.org/wiki/Compu... |
|
| Dec-26-17 | | ChessHigherCat: <Marmot PFL> computers are definitely gifted at remaining poker-faced but you would need special peripherals so they could give their partners footsy signals under the table. Every poker player I know claims that the psychology of gestures and bluffing are a big part of the game. Computers could be programmed to bluff based on the expectations raised by their past playing patterns but they would be a bit limited when it comes to gestures. Anyway, it would be interesting to let a computer player "sit down" at the table with professionals to see how important the psychological aspects really are. |
|
Dec-26-17
 | | OhioChessFan: <CHC: Why on earth would you take it for granted that that I don't know anything about heuristics or probabilities, that's like some kind of stupid racial prejudice. > Looking at your picture, I can't tell if you think <AK> is prejudiced against your black or your yellow skin. |
|
Dec-26-17
| | OhioChessFan: <Marmot: Why hire a composer or pay for the rights to use music when a computer can do the job for pennies on the dollar?> Unemployed composers don't have the money to buy computers. Just saying. But high tech professionals are hopelessly clueless in discussing such things. I have a brother who is an electrical engineer who is stunningly-perhaps purposely-oblivious of the employment ramifications of technological innovation. |
|
| Dec-26-17 | | ChessHigherCat: <OhioChessFan: Looking at your picture, I can't tell if you think <AK> is prejudiced against your black or your yellow skin.> Most people find all my skin colors a'peeling. |
|
| Dec-26-17 | | WorstPlayerEver: Yeah, I wrote this song back in 1952 with CSIRAC https://m.youtube.com/watch?v=FH2Eg...
https://en.m.wikipedia.org/wiki/CSI... |
|
| Dec-27-17 | | LameJokes55: <AylerKupp:Given that AlphaZero uses specialized hardware (Tensor Processing Units or TPUs), and a proprietary one at that, I doubt that there will be PCs available with these anytime in the near future.> 1 Fair Match: If that's the case, then how do we create a level-playing field for both opponents (AZ and SF)? By loading original (and not truncated) version of SF on a super computer. Then, let AZ evolve itself with months of self-teaching. Both giants then square up for a 100-game rematch. A repeat performance by AZ in the second match would lend legitimacy to the current result. Of course, this is a wishful thinking on my part. There is no guarantee of the rematch in near future. 2 Application of Self-teaching: Google does not plan to enter chess or go market. They have conducted their media communication in a hush-hush manner. Mystery and suspense could give Harlan Coben a run for the money! Where could they try this (self-teaching) program, next? Self-teaching is applied to human sport where rules are few and evolution manageable. I don't think, it could be applied to something like language or writing, being too indefinite. As one kibitzer pointed out, music might represent such an opportunity. Human ear hears sound frequency in the range of 20 to 20,00 hz. In addition, there are just 12 musical notes. Google program could use this information to work out various permutations and combinations. Thus, creating a number of musical compositions. How melodious that ends up on human ear is altogether another matter! |
|
| Dec-27-17 | | ChessHigherCat: <Lame Joke 55: Where could they try this (self-teaching) program, next? I don't think, it could be applied to something like language or writing, being too indefinite. As one kibitzer pointed out, music might represent such an opportunity. Google program could use this information to work out various permutations and combinations. Thus, creating a number of musical compositions. How melodious that ends up on human ear is altogether another matter!> Right, that's why human coaching would be necessary for anything beyond binary decisions (like winning move vs. losing move). Computers can generate beautiful fractal images based on mathematical formulas but that's a "no-brainer", pure calculation. There seems to be an insurmountable barrier between quantitative and qualitative thought requiring judgement and imagination, but who knows, maybe if the computing power reaches a certain level on a "neural network", computers will really be able to think like humans (for what it's worth :-) All the science fiction writers thought there would be "thinking androids" by this time. Maybe that's completely unrealistic, though, just as no computer programs will probably ever be able to predict cloud formation, there are just too many variables. Some people were so optimistic that they thought it would be possible to hear original performances by Mozart by retracing the vibrations of the air molecules produced by the instruments back to the 18th Century :-) Quantum theory has pretty much destroyed that kind of absolute determinism. |
|
| Dec-27-17 | | markz: <LameJokes55:
1 Fair Match: If that's the case, then how do we create a level-playing field for both opponents (AZ and SF)? By loading original (and not truncated) version of SF on a super computer. Then, let AZ evolve itself with months of self-teaching. Both giants then square up for a 100-game rematch. A repeat performance by AZ in the second match would lend legitimacy to the current result. Of course, this is a wishful thinking on my part. There is no guarantee of the rematch in near future.>Even I think SF on a supercomputer can beat AZ, it still isn't a fair match. Because AZ has home court advantage on supercomputer, SF has home court advantage on PC. A real fair match should be something like 100 games on supercomputer, and 100 games on PC. I am very sure SF can beat AZ easily in any real fair matchs. Also, I believe most people would like to use chess engines on their PC instead of supercomputer. Developing and running chess engines on supercomputer aren't very useful. Increasing the training time may not improve the neural network. As shown in figure 1 of the paper, after 300K training steps, further training may not be very helpful, sometimes (very often) even harmful. |
|
| Dec-28-17 | | refutor: well at least it beat the French twice, makes me happy |
|
| Dec-28-17 | | 50movesaheadofyou: What will AZ tell us in the future if it learns and masters other disciplines? Human sexuality, the true story of jesus, its opinion on abortion, origins of the universe, etc. etc.
When asked about religion, what if it insults Christianity or othere faiths?
Many authorities will want it banned, even destroyed. It can be a dangerous device.
I say continue the research, let it show us its max potential. The same they said of the internet, that it could destroy the world. It changed the world for sure it occupies a dominant place in modern society but hasn't destroyed it. Maybe in the future some highly advanced self-taught version of it will though. |
|
| Dec-28-17 | | ChessHigherCat: It's dubious whether anything self-taught based on self-consistent principles could ever become controversial because it can never encounter any outside influences, but if it contains contradictory principles to begin with then it could become "heretical". For example in medieval Scholasticism, the professors at the University of Paris tried to reconcile Aristotle with the Gospels and the brightest professors who picked up on all the contradictions, like Pierre Abelard, were criticized as heretical for "trying to look God in the face". |
|
| Jan-10-18 | | zborris8: I've read comments that members aren't able to replicate the mistakes by Stockfish on their own systems. |
|
| Jan-10-18 | | WorstPlayerEver: A theory of conditions. Let's elaborate.
There could be an application which plays perfect chess. Theoretically. Let's imagine there is such program. There are 20 moves, so there are 20 best variations. Without question. Another theory; since we do not have such program, it could be there is a program, at some point, but how do we know it found the best moves? It could be that some point the best variations are played and the engine will recognize them as such. Which in practice means it eventually 'stumbles upon' the best variations. However, it could be also the case that a program gets eventually an update and will predict the best variation from the start position and recognize it as such. |
|
| Jan-11-18 | | coxscorner: I wonder why they didn't release any of the draws? Surely some of those were interesting. |
|
 |
|
|
< Earlier Kibitzing · PAGE 11 OF 12 ·
Later Kibitzing> |
|
|
|
NOTE: Create an account today
to post replies and access other powerful features which are available only to registered users.
Becoming a member is free, anonymous, and takes less than 1 minute! If you already have a username,
then simply login login under your username now to join the discussion.
|
Please observe our posting guidelines:
- No obscene, racist, sexist, or profane language.
- No spamming, advertising, duplicate, or gibberish posts.
- No vitriolic or systematic personal attacks against other members.
- Nothing in violation of United States law.
- No cyberstalking or malicious posting of negative or private information (doxing/doxxing) of members.
- No trolling.
- The use of "sock puppet" accounts to circumvent disciplinary action taken by moderators, create a false impression of consensus or support, or stage conversations, is prohibited.
- Do not degrade Chessgames or any of it's staff/volunteers.
Please try to maintain a semblance of civility at all times.
NOTE: Please keep all discussion on-topic.
This forum is for this specific tournament only. To discuss chess or this site in general,
visit the Kibitzer's Café.
|
Messages posted by Chessgames members
do not necessarily represent the views of Chessgames.com, its employees, or sponsors.
All moderator actions taken are ultimately at the sole discretion of the administration. |
Spot an error? Please suggest your correction and help us eliminate database mistakes!
Copyright 2001-2025, Chessgames Services LLC
|
Kibitzing
