Will computers ever solve chess?

Elo is not a linear scale, and 5000 elo is not even possible...nor can someone be "born" at any rating level.  To get 5000, you would have beat players at 4800 elo with great regularity.  To get to 4800 elo, those players would have to...see the problem yet?

I hereby declare you the least omniscient chess god of all time.

While your claim is highly plausible, it is not entirely clear.

I think the best practical way to get a handle on the Elo of a perfect play would be to derive a really good model for how ratings rise for computers with processor speed (or computing time).

If a doubling in processor speed always gave the same Elo advantage, there would be no maximum Elo. In practice, the gains gradually decrease as the strength of a computer increases, but the shape of the curve is crucial to the question, and the details of this are unknown at present.

Here is some (old) data that hints that with some of the top engines, doubling computing power provides 40-80 Elo points:

Rybka 4.1 6 core (under SCCT Auto232 conditions) performed 65 Elo better than Rybka 4.1 4 core
http://www.sedatcanbaz.com/chess/ratings/scct-auto232/

Note:i7 980X / i7 970 @4.33GHz machines are approx.2 times faster than i7 920 @3.0GHz/QX9650 @3.66GHz

Rank    Name                     Elo    +    -  games  score oppo. draws
    6   Deep Rybka 4.1 x64 6c    3358   15   15  1363   59%  3305   54%
   15   Deep Rybka 4.1 x64 4c    3293   13   14  1602   47%  3314   56%

Rank  Name                     Elo    +    -  games  score oppo. draws
  12  Fire 2.2 xTreme x64 6c   3311   23   22   530   52%  3297   63%
  18  Fire 2.2 xTreme x64 4c   3275   16   16  1165   41%  3328   59%

(from http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=24572)

One theoretically relevant fact is that for a given amount of real time, doubling computing power would eventually lead to the capability to calculate the complete chess tablebase and play perfectly. No more computing power could provide any improvement. The sort of speed involved is not feasible with conventional electronics. There is the usual trade-off between computing power and storage capacity here, but both demands are infeasible.

To get from say 3400 to 5000 would be 40 doublings of speed if Elo continued to rise at 40 points per doubling (we can be rather sure it declines). 40 doublings (an increase in speed of about 1 trillion times) would not be enough to explore the entire tablebase of chess, so this is not excluded by this constraint.

(3400 was a rounded rating for the latest Komodo, running on an out of date 4 core processor, as specified by the CCRL rules. Merely upgrading this to a modern high end CPU would probably add roughly 200 Elo points!]

One possible scenario is that chess could be solved before a complete chess tablebase is created.

If there is only one way found that (for example) white can force a win, then we know the result of perfect play with only a portion of the game tree calculated.

Although unlikely, such a surprise might be discovered at anytime without much of an advance notice (assuming there are researchers working on it).

We won't know the answer to perfect play with "composed" positions, but we would know who wins from the normal starting position in chess.

There will be no surprises here. It is clear that if you can force a result (very likely a draw) in N moves, most of the time the opponent will have a large number of responses. This makes the set of positions the opponent can reach very large. It is necessary to evaluate every single position the opponent can reach: general rules are of very little use for absolute certainty.

Therefore, this simply cannot be achieved with currently feasible technology. Before it can even be plausible, huge advances in computing hardware would be necessary. Note that the problem with Go is far more severe. Typically, the opponent has around 200 moves, and this number is not reduced by having a bad position!

If you find a set of play that leads to draw without calculating the entire game tree, then chess is still not solved. There might be different play where one side can force a win.

But if you find a way that white can FORCE a win, or if black can FORCE a win, then chess is solved. There would be nothing else to be studied. The remainder of the game-tree can remain unknown, but chess would be solved because you know who can force a win.

It is theoretically possible for this to happen at anytime -  if someone would just find that one path to a forced-win.

Finding the path to the forced win requires...you know...traversing the possible positions, and then getting really lucky.  How lucky?

The number is 10^120.  So, let's say we only have to hit a 1 in a billion shot to get lucky and solve chess:  that still means you have to traverse 10^110, which lo and behold is still far more than the number of atoms in the known universe.

So, what you really need is to hit a 1 in a billion times a billion times a billion times a billion times a billion times a billion times a billion times a billion times a billion times a billion shot.

Good luck with that.  I won't be waiting up.

First of all, the number is not 10^120.  That is an estimate for the number of 40-move games.  The relevant number is the number of possible positions, which is about 10^44 - still a huge number, but much less than the number of atoms in the known universe.

Second, you can greatly reduce the number of positions that you need to check without getting particularly lucky.  Let's say, pessimistically, that chess is a win for white and that there is only one play for white that leads to a win.  For the opening move, there are twenty possible moves; if we just started taking moves and random and doing a depth-first search, we would have to go through about 10 first moves on average to find the winning first move.  That still cuts the search space in about half.  Then for the second move we can cut the search space in about half, and so on.

I tend to believe that the game of chess is a draw, and that typically there will be many moves one can make in a given position that preserves the draw.  For example, it seems reasonable that for white's first move any move will lead to a drawn position, so one can just pick a move and not need to get lucky at all.

So the magic number is not 10^120, but some number much less than 10^44 - still probably a long ways away, but can you really assert that it could never be done?

If chess is a draw, then you just need to find forced draws for both sides.  If white has a forced draw, and black has a forced draw, then neither side can force a win.

Do computers want to solve chess?

10^120 is the total number of games, but we don't need to check them all, like Deedlit says. Looking for a forced mate can be done by looking at normal play (ignore ridiculous moves) which reduces the search to 10^40 games (from Wiki "Shannon number").

Then this is reduced more because opening theory has already investigated far into the game. Then from here, things become yet easier because all chess games with seven pieces and less is already solved.

So with work that has already been done, and with very fast alogorithms to decide good chess moves, we may be close to the "cusp" of solving chess.

On the other hand, if best play is a draw, chess won't be solved until the entire tree is checked. So we can only expect an answer in our lifetime if one side can force a win. But if perfect play is a draw, we probably won't know it within our lifetime.

I don't know how anyone can make a guess if chess is a draw or a forced win for white. It's even possible chess is a forced win for black.

afaik, mathemaricians already demonstrated that a strategy to win in chess must exist. but we don't know the strategy nor we know who is going to win: white, black, or depending

so, chess can be solved by super computers using brute force(i.e. listing all possible winning positions or something like that), or solved by finding this 'strategy'. we don't know how much this strategy could be complicated though

The guy that killed me said this...

"Imagine this thought experiment..."

Probably thought he was the next Einstein.

This is true of chess and of all other finite games of perfect information.

Possible, but I would bet my life it is not true, using a probabilistic approach.

Reading the article more carefully, the most important number in the article is the number of reachable positions, bounded above by 10^46.7 (and is probably pretty close to this).

Tablebases don't store legal games, they store legal positions and their evaluations (which are easily recursively generated, working backwards from positions with a result. Solving chess completely cannot rely on anything less than every legally reachable position. Needless to say, there are a LOT more legal games than legal positions!

Even if you subscribe to the 10^46 number...it's effectively just as ridiculous.  Man, people just cannot grasp exponents.  So, 1 in more than a billion billion billion billion...effectively just as impossible as the 10^120 number with today's technology and even tomorrow's imagined technology ...

So, the next claim is that the list can be drastically reduced...but again, exponential immensity eclipses imaginations.  Even if you "prune" 99.9% of the positions (surely impossible), you'd still be at 10^43. 

10,000,000,000,000,000,000,000,000,000,000,000,000,000,000...

You don't need to play every possible move to "solve" chess. All you need to do is prove that there is no way to gain a decisive advantage against best play. 

We're getting close to that level of proof. In fact, I now believe chess could be solved within my lifetime using that definition.

Even so, the definition of a "solved" game is very specific on the criteria required:

https://en.wikipedia.org/wiki/Solved_game

For your definition, SmyslovFan, you could argue it's already solved.  Engines are far better than the best human players, and those best human players are too fearful to even attempt even-odds matches anymore...so effectively, there already is no way for a human player to "gain a decisive advantage against best play".