|
« TAP TO LOG IN
|
Chess related: Artificial Intelligence invades Chess
« Back to forum| From | Message | ||
|---|---|---|---|
|
 Artificial Intelligence invades ChessAs the years passed and the chess engines progressed well past what present day grandmasters could accomplish over the board, chess engine tournaments became all the rage. I was less interested in them as they always seemed to be set at quick time controls. And the emphasis seemed to be more on competition rather than finding the very best move in a position or even in understanding the position. In the last few years the big three in this area were Stockfish, Komodo and Houdini. Now in December 2017 a new kid has appeared on the block as per the link below. www.youtube.com It’s name is AlphaZero, a piece of software using artificial intelligence to learn how to play chess. Apparently after mastering the game of GO, the brains behind this new software phenomenon decided to unleash it on the world of chess. Apparently it taught itself to play in 4 hours, then proceeded to beat Stockfish in a chess match. I’m unsure of the time controls used (always an issue for me) but nevertheless it strikes me as a mind blowing achievement. This is Stockfish we’re talking about here. Anyway I welcome anybodies thoughts on this… |
||
|
deeper_insight 16-Dec-17, 13:03 |
Yes,but baby is still learning and papa still holds the crown... So,this new AlphaZero so far,is not impressing me.Let me know if/when AlphaZero defeats the NEWEST versions of Komodo on longer time controls(not blitz or one hour games,ect)...like an average of 40/2(or longer) and then and ONLY THEN will I be impressed.I'm sure Alpha is in its learning mode yet,but again,i'm not impressed yet,since stockfish has been defeated by not only Komodo but other chess engines.The problem with Stockfish is that it has NOT been improved or upgraded(like Komodo is improved on a perpetual basis),thus,some programs on longer time controls can analyze on better improved ply depth searches as opposed to Stockfish. I have been post game analyzing with about 20 different chess programs over the last 10 years or so.Some programs are better at faster time controls,some with end games over others,ect.and others are stronger at midgame levels,even if the overall rating is a bit lower than its program vs program competitor. For example,Stockfish is a bit stronger than Komodo with just a few pieces left on the board,but Stockfish is a bit WEAKER than Komodo in midgame(after book) where Komodo many times will edge out a better positional deeper strategy than Stockfish.So switching post game engines DURING game analysis is one of the secret keys to deeper/more efficient analysis success. Again,i'm keeping my eye on any new engines or other chess avenues like Alpha,but i'm only impressed with solid results against the best.And the best to date(at least above Stockfish)is Komodo,with the constant program improvements they make several times a year.The flip side of the coin is that the best opening books,not engines can play a MORE important role in winning games this day with engine vs engine or program tourneys.Some books run over 50 moves deep automatically and can be tuned to play only on narrow band top tourney levels with no weaker variant moves or ? moves,thus,when a top notch program sets in like Komodo or stockfish(after book),it does not matter anymore.The game is almost always = and the draw rate is 98% or higher.Fischer once was quoted as saying that chess is all memory now(opening long memory book)and is not played from self invented and gut moves like it was decades ago.Its all just copycat book moves for 75% of the game many times.And after that,one can memorize endgame Komodo/Stockfish move variations to complete the game. |
||
|
A few issues have since been highlighted since Alphazero's Victory There seems to be an emphasis on engine competition whereas I think they are more valuable for game analysis. The Komodo engines seem to put more an emphasis on evaluating a given position and less on brute force making it an ideal aid for deciphering your own games. Anyway we'll see where all this leads. I wonder what the old masters would think about all of this? |
||
|
deeper_insight 18-Dec-17, 07:51 |
sixofspades... So time controls are of course,a big factor in determining superiority.I generally ignore blitz time controls or even one hour game time controls to make my personal decision on the best of the best.Also,of course,the amount of "cores"can make a huge difference.I have a 10 core system at home.Its steps on a quadcore in search depth and speed like a big foot on an ant. Every few years,someone creates an engine that will outdo its past competitors.Years ago Fritz,Shredder,then Rybka,then Houdini,ect.Now Stockfish kills all 4 of these and Komodo edges out Stockfish. So who knows?Maybe Alpha Zero will end up holding the crown in a year or so?Its going to happen by some future engine,either way,given time.But at software rated over 3000...it really has very little significance anyway to players below master level,(in my opinion).Players in the B class or below need to study many other basics before even trying to study with a 3000 ELO rated program(again,my opinion). I find this "Alpha" interesting and will keep an eye on it. Best, TA |
||
|
Thanks for your input |
||
|
www.google.com Seems to me now,with all the older reports about the famous Alpha-Zero becoming the crown jewel of chess program dominance,that Stockfish still rules the board.I agree,but there is a catch.Once upon a time Komodo seemed to dominate at some tourneys.It all boils down many or most times to the level and degree of massive hardware used compared to the other program....well,at least in part as will be explained in the link below.It also depends upon the Hash,how an engine can be overclocked,intercoolers and the time settings which might be set at,for example: 20/4 instead of 40/2.Setting at infinite analysis are good for analysis,say for example 24 hour analysis per move(after book)but that does not mean that stockfish or Alpha-Zero will come up with the utmost best move in 24 hours,only what its analysis window shows at that exact 24 hour deadline....for in 26 hours,it might have something "better".So,infinite analysis is misleading.A true test is regular tournament setting whether it be 40/2 or even perhaps any other slow settings far from a blitz setting.Keep in mind that all this chat past or present about Alpha-Zero is a moot point.Why?Because the average chess player cannot afford to have Alpha-Zero in their playroom obviously: www.google.com So getting down to Earth.... Stockfish just unveiled its new Stockfish 18 version very recently on Jan 31,2026.Downloads are free and are usually placed in ones zip file before loading into your GUI.So,it is facinating that a free Stockfish with average hardware can defeat a massive monster like Alpha Zero.And if one "tweaks"their own personal software with overclocking,higher Ram and the right setting....ALONG with a top opening book like Hiarcs....you have an instant "Genie In A Bottle". www.google.com Over the years I found it facinating to test computer vs computer which can be set up in some GUI's.I have used infinite analysis but only rarely,since I do not want to wait hours per move. With todays hardware quality and ply depth speed,one does not need infinite analysis anymore. Tuning stockfish beyond "default mode"is the key,but that is something I will not get into in the forums only my old club which is gone now. So Stockfish wins for POST game analysis as well as engine vs engine tests. GK has Stockfish analysis of course,but may not necessarily have Stockfish 18 or be able to tweak Stockfish to ones individual needs of critical analysis in a more customized way. TA Joe |
||
|
Above in those links you will notice at first how Alpha-Zero killed Stockfish but then the tables turned a few years later and Stockfish took over the reigns of power.In a few years,some new chess program with super AI will defeat Stockfish 18 or Stockfish 30 in a decade.Chess programs have always improved,all the way back to...."Boris in a box".An antique chess chess playing program that I still own in my massive collection,along with Tasc R-40 and the famed Phantom Chesster Eyeball.Stockfish would crush them all,but those wooden and plastic beauties are worth some big money to the right collectors and have a deep beauty and memory attached to them that no cold Stockfish will ever have. |
