True or false? Chess will never be solved! why?

Sort:
tfdenham

So... let's play!  Laughing

Knightly_News

"Solved"... in a way this reminds me of "Why do we keep having to hassle with finding, cooking and consuming food all our lives?  Why can't we just eat once and for all?"

How can you "solve" chess when each position (except maybe in some end game positions or mate in 1, 2, 3, etc..., can transition to a huge number of positions and possibilities of response, which follows with another huge number of potential responses, where in most cases a win would not be had for many moves so you can't conclude so early that a certain moves leads to a definitive win?  

zborg

Chess has been "solved" for 6 or 7 pieces.

Very long way to go, in order to get to 32 pieces.  Very, Very, unlikely too.

KingSullian
reflectivist wrote:

 "Why do we keep having to hassle with finding, cooking and consuming food all our lives?  Why can't we just eat once and for all?"

 


We can, I do!!

 

 I have a endless keg of beer and hotdogs!Kiss

TheGrobe

It's like the old riddle about getting a penny on the first day of the month, double that in the second, double that again on the third etc. by the end of the month (or for the sake of comparison, 32 days later), you'd receive about $43 million. On day seven, you'd receive $1.28. We're on day seven right now, and the factor is much more than double for each piece added.

Nordlandia

If the perfect chess strenght equals 3800. Should't it be possible to reach that by connecting all processors together.

Engine raiting increase approximately by ~50 doubling speed.

Ziryab
Firestopr wrote:

chess will definitely be solved. I'm predicting around 2825. technology is rapidly increasing right now, and at this rate, we will soon solve chess and all the possibilites!

Fix your math, and maybe you'll find agreement. One teeny problem. The earth no longer will sustain human life in 2825 unless dramatic lifestyle changes are made by so-called First World countries by the year 2000.

qrayons
btickler wrote:

Good luck with that.

Thanks for at least copping to 10^43 instead of the inflated 10^123 ;).

The 99.9% is not nearly a high enough level of culling, though; I would venture to say that every Chess position played in human history falls...well...far short of your 10^40 number :).  It is fairly telling that even with all the analysis accomplished thus far, there are pretty much *no* indications of some giant untapped reservoir of brilliant strategies and tactics out there that can be unlocked with brute force calculation.  There's no major "contributions" that Chess computers are making to fundamentally change the middle game, or shift conventional wisdom about position play, etc.  

Now a lot of that is because we hamstring our computers with our own biases (our opening books, our scoring algorithms in terms of piece values, etc.), but still, you would think by now that computers would have done something more substantial than just busting some known variations.  They haven't though. 

The rules of Chess do allow for a staggering number of moves, but the mechanics of the pieces also make for a very narrow range of truly viable moves.  We just have not applied a lot of thought or effort to quantifying what constitutes a viable move.  We've only really taken two tacks so far...teaching a computer to play like we do as best we can, but really quickly and precisely ;)...or by doing brute force calculations without any real rules or judgements involved at all...and not much in between those two extremes.  Our current methods involve seeding our computers with openings and various predetermined knowledge, and then asking them to apply brute force down certain promising branches.

What we have not done, is back our way into this by using the growing amount of played games stored to start analyzing new values for pieces and positions (kind of like the growing "effective" valuation of a passed pawn has been analyzed, though that would be a horribly simplified example), thus determining some hard numbers for the "guiding principles" chessplayers agree on, and then using that data to start determining what can or can not be culled from the full tree of moves.

If all of human chessplaying history has already culled the number of viable moves far, far, far more than 10^40, and brute force applied to various positions has not produced any wild breakthroughs (how many Chess books out there have some grandmaster saying "playing this computer over time taught me something completely new about Chess that I had never even conceived of being possible, or as a factor of consideration, before"?), it stands to reason that 10^40 is not the number of moves we need to be perusing to determine "best play".  It's a much smaller number, and where brute force calculation can really help us is in determining the exact set of criteria that allow us to cull the tree way, way down.

 

Why would we rely on historic games if we were trying to solve chess? By your own admission, the number of games played throughout history is several orders of magnitude less than what is possible. If we were in 1913 instead of 2013, you’d be eliminating all of the hypermodern openings from analysis.

 

 

The rest of your post seems to be built off of some other misconceptions. If you’re looking for computer inspired breakthroughs in chess, you have to look at the endgame since that’s the only thing computers have solved so far. They’ve refuted several endgame studies, which only strengthens the case for looking at all possibilities rather than relying on humans. I wouldn’t expect breakthroughs in opening analysis from computers today anymore than I would expect the roof of my house to be built before the walls have been put up!

TheGrobe

The biggest problem seems to be the misconception that chess can be solved by making a couple of assumptions to vastly reduce the problem to one that is conceivably manageable. You can't. No matter how certain the assumptions may seem, they are still assumptions and even a one in 10^123 probability that the assumption is mistaken is enough to tank any solution that employs it. Save for the minimal reduction that can be acheived through the equivalence of reflected and rotated positions, brute force is the only option.

A proof is a proof is a proof. Anything else is an educated guess.

DiogenesDue

If it had been up to some of the people on this thread, Geometry would never have been developed :).  After all, you can't possibly determine anything mathematically by relying on a long series of smaller proofs built upon each other...you just have to measure every angle by brute force to determine what it is...using anything else is not "proof". 

TheGrobe

This proof is no exception, consider the seven price table base an early proof that can be built upon.

Unfortunately, there are no shortcuts to the 32 piece table base that represents a solution. Next step is 8, then 9 piece table bases. I'm not sure where we'll peter out exactly, but it will be long, long before 32, that is certain.

DiogenesDue

The 32 piece table is nowhere close to the Geometry analogy.

Unfortunately, there are no shortcuts to the 32 piece table base that represents a solution.

Nice wording :).  That sentence is true.  However, the 32 piece table base only represents the the "let's count on our fingers and toes" solution, and is far from the only conceivable way to get there.

TheGrobe

Any other solution would be mathematically equivalent and equally impossible. There's no getting around it.

DiogenesDue

There's no getting around it.

Ok.  Sure.  A true scientist would not make that statement without any qualifer, but maybe I'll stop trying to be Copernicus to your Ptolemy. 

This is the problem with many Chess players' point of view regarding computers.  They don't understand them, and they fear that the game will be destroyed by them.  So they post inflated numbers, and talk about how it is impossible that this or that will happen.  Kasparov and his team of GMs made some horrific errors in judgment and lost to Deep Blue by trying to play one program as if it would suffer from the weaknesses displayed by other programs.  In doing so, they showed a complete lack of understanding of how Deep Blue was put together.

It's interesting though:  Deep Blue was basically "mentored" by GM Benjamin ("The move 26.f4 was a result of my laboratory work. I spent a great deal of time forcing Deep Blue to play closed positions and trying to get it to utilize "levers" along the lines of the move in game two. We may have set the value for levers unusually high because we wanted to make sure Deep Blue would play them.

I think a number of moves from that game (especially those associated with building up with the rooks on the a-file before opening it) would not have been played by the commercial programs of that day. It is a shame that Kasparov met Deep Blue's groundbreaking play with suspicion rather than admiration. Programmers were able to make many of their subsequent advances because they learned from Deep Blue's accomplishments.")

As I mentioned before...the problem of "solving" chess, which I will define as being able to produce "best play" in every situation, and not as traversing a 32 piece tablebase, has only been attacked in two very rudimentary ways:  by "teaching" the computer to play like a human grandmaster by rote, and by raw computational force applied against those "teachings".  We did this via our own basic assumptions (piece valuations, etc.).

Now we have a huge and ever growing base of games that can be used to analyze what really is "best play", and statistically allow us to gather data that can be used so that raw game results will be used to tweak the computer engines, not a GM consultant hand-tweaking table values.  

The reason computers fail to play well in even positions is because the "burn the candle at both ends" approach we have taken is inadequate, and this is where it falls down, and the fact that the computer's processing power is flailing around trying to deal with with our poorly coded valuations of various postional advantages et al becomes very obvious ;).  There's nothing for the computer to do.

The whole reason we went the route of trying to exhaustively traverse and evaluate every possible move/position is because of our own failings in trying to understand and quantify things like valuing pawn structure in the opening that will only come into play 30-60 moves later, etc.  But we don't need to do that anymore.  We can let systems crunch the growing database of games and figure out it's own valuations without any human tweaking...and by the end, computers will know things that we don't about the game, like, say, exactly how much more valuable two bishops are in X, Y, and Z situations down to a thousandth of a point, and those values will change on the fly as new games flow in (computer and human).

It is pretty easy to imagine the Internet a hundred years from now, and a SETI-like distributed processing effort to advance Chess analysis, and it is pretty easy to imagine Chess playing programs that are not so rudimentary...

Not that Grobe will care, but for anyone else:

Here's a free class from Stanford that represents another route right here...(not one that I personally think will bear fruit for advancing Chess analysis as-is, but I would not just dismiss the possbility, either).

https://www.coursera.org/course/ggp

Irontiger

I am pretty sure theGrobe is delighted to be told he has no clue by someone who demonstrate how this statement applies to him.

What btickler is suggesting is that somehow (throw a few buzzwords like "statistical analysis" here) we will manage to teach the computer to play the best move in every situation without having to actually calculate the continuation until the mate (or the 7-men database for that matter), by managing some very clever positional analysis.

Don't you think that if such a thing were even remotely possible, Houdini and others would already be perfect players ?

 

And your link has absolutely nothing to do with your argumentation - the underlying algorithm will still be a minimax (maybe refined) thingie.

 

It's very popular among non-scientists to dismiss mainstream science because they believe they are going to be the next Copernicus or Marconi. But Copernicus had some solid evidence for his theory (and additionally was very acquainted with the previous mainstream theory, which you obviously are not), and Marconi demonstrated radio transmission in his room at 12. You just have the intuition that all others have got it wrong.

waffllemaster
FirebrandX wrote:

Statistics are already showing in centaur correspondence chess that you don't need to brute-force every possible move in chess in order to solve it. Computer software and hardware are getting to the point where the tactics in centaur CC games are so incredibly deep that your only chance to win is in the opening preparation. When two centaur CC opponents are prepared with databases of previous centaur games, the outcome lately is almost always a draw.

Another situation to consider is certain defenses by black being pretty much impossible for white to reach a complex enough position to get winning chances from. Two examples are the Petroff and the Marshall gambit in the Ruy Lopez. If a prepared centaur cc player uses either one of these lines, white is not going to win. Yet another line in the Ruy that is fast becoming rock-solid for black in centaur cc play is the Archangelsk variation.

So yes, chess will be solved in the future in terms of computers and cc time controls, but live OTB games will never get there for humans anyway. There is simply too much to memorize.

I'm not sure what you mean.  How can cc statistics support this?

If the only chance to win is in opening preparation due to tactics wouldn't that simply show that humans aren't contributing to the middlegame rather than showing something related to solving chess?

I'm sure you have to work with the engine to make sure it's seeing all the tactics, but it's still the engine doing all the verifying.  There is still the problem of the horizon effect, so I don't see how this can be taken to mean e.g. certain openings have been solved as a draw.

waffllemaster
FirebrandX wrote:
waffllemaster wrote:

I'm not sure what you mean.  How can cc statistics support this?

I actually had edited my post with more information after you quoted it. I've been playing on ICCF for five years now, and recently became the current USA cc champion. I can tell you from experience that a major portion of viable chess lines have been worked to dead draws on ICCF. As I already stated, black can effectively force a draw with defenses like the Petroff on ICCF. There's nothing white can do to improve that situation except to no longer play 1.e4. Basically you have to hope the other ICCF player wants more than just a draw, or you won't get any winning chances.

I edited my post and added some stuff too :p

I guess my question is, if you say it's a draw can you claim that a centaur today could easily draw a centaur with technology e.g. from 20 years in the future?  Or do you mean when equally experienced and equipped ICCF players face it's a draw?

ChiseledChessy

its defineltly got an ultimate truth, i think since chess is limited to 64 squares computers in time can solve it like checkers but then again u can create more 40 move chess games than there are electrons in the universe ive heard so theres a lot to account for but of course including unrealistic games but thats still a huge number, its more about knowing what not to do ulimately because infinity is humbling, at least infinity to my brain

tmodel66

Sorry to beat a dead horse here, but if computers evolve to a level where they can defeat or draw a human 100% of the time, cannot we consider chess "solved"?  I understand the definitions provided by TheGrobe and others requiring every variation to be exhaustively covered and saved to "prove" the solution.  However, if we get to a place where a computer will never lose against a human, what greater solution is required?  Ultimately, you could only improve play by matching one computer against another.  While this might lead to a better understanding of "perfect" play, the exercise at that point becomes purely academic.  

I do not think we are more than 10 years from this being reality.

Does this mean that chess between humans becomes an anchronism?  I don't think so.  But, for practical purposes, chess will be solved.

If you don't think so, I would just ask you to play that computer, where you would never win.  

If there were a human that played and could never lose, we would consider him/her to be the perfect chess player.  Why would we require more of a computer opponent?  Winning is the object of the game.  When you cannot lose, the game is solved.

tmodel66
Victor-Servranckx wrote:

Nothing to do with solving chess. Suppose nobody can beat carlsen is chess then solved?

Yes - if no one or computer could ever beat Carlsen, he would have solved chess.  Any reply you made to his moves would not beat him.  (This is the goal of chess. ) 

But, we know that isn't true.  Carlsen - great as he is - is "human".  But, when a human could never win against a strong computer - and I believe that will happen, chess will be solved.

Here's an analogy, when you pressed send on your previous message, you knew that it would be emailed unless you had computer issues or the grid went down.  While you didn't (and couldn't) know every router and switch that the message might take, you were confident that the message would be sent.  Do we consider email "solved"?  Yes, because you understand in an ideal setting what the outcome will be.  Similarly, when you can predict the outcome with certainty - even without mapping every variation - chess will be solved.

Anothter anology...Is a King vs. King (only) endgame solved?  Yes, even without the 50 move limit.  Do you need me to exhaustively demonstrate every move to you?  Of course not, you will accept that it is a draw.  

Similarly, once computers can reliably guarantee a win or a draw, the game is solved.