I think the main problem is that in many positions multiple moves are good enough so, while a program could prune out some lines from specific positions, there will be a number of good enough moves still.
To solve chess you need to be able to say that you can force a win from a specific opening move, or potentially moves, or prove that the game is a draw from most initial moves.
Just having a program that is capable of beating humans isn't solving chess. Solving chess is proving forced outcomes.
For example, if such a program could solve chess, and we posit white does have a forced win (unlikely but possible), then the program would have to be able to show that none of black's replies lead to anything but a win for white and that requires looking at a large number of variations still.
If it is instead a draw, as mostly is assumed, it would have to show that the draw is forced also and have to validate every variation where the draw holds, which again is likely a vast number of positions.
I'm not saying a method similar to the DeepMind/AlphaGo program couldn't do it, just that I still think the complexity level and search tree is too large for current computing. The program hasn't solved Go, but can play better than humans. We pretty much already have that with modern engines for chess.
I think most of us know about google deepmind. they produced an artificial intelligence which can learn on its own, can develop on its own. Its name is AlphaGo. that was designed for GO but if it is done for chess, cant it solve chess?
In computers solving chess, the main obstacle was the number of possibilities. There are millions or billions of possible ways of playing 5 or 6 moves or may be 10 moves. ordinary Computers cant handle it because they need to try and see every variation. but can not Alpha go handle it? Alphago knows what is wrong in GO (in game 4 against Lee Seedol it detected, discovered its own error/mistake). So, since Alphago knows which move is losing, since it can omit unneeded possibilities it can ease solving chess. after the first move of white, there are 20 possible replies by black. This is the problem for us(humans) and ordinary computers; we need to check every variation and we can not do it so deeply. since AlphaGo can think like human, it can omit unneeded variations. ordinary computers dont know it since they are just calculation machines, they just do what is taught to them, what they are programmed for but Alphago is not like them. Let me explain it more clearly:
if we take 2005 version of stockfish or komodo and make them play against each other over a billion timez, can they reach their 2015 version? can they learn something new? can they detect their own weaknesses? NO! as long as programmers do not do that, they wont improve. but AlhpaGO is not like that, it can do that. Alphago does not need to be taught in order to improve. so, if goofle deepmind composes a AlphaChess, after playing many games against itself, it can solve chess at least in a weak sense I guess. The theoretical perfect result can be known I guess. or am I wrong? ex:
Alphachess plays itself, it tries this variation: 1. e4-e5 2. Nf3-f6 3. Ne5-fxe5 4. Qh5+ in this position, white can force a win, after trying this, Alphachess will omit this possibility. this is just an example, many variations can be omitted in this way.
2. solving chess cant be that difficult:
for a single CPU it is too difficult as long as we dont have quantum computing computers but we have millions of computers in the world. what if we make an Internet server and connect thousands of computers and make an alhorithm or program or software in which they can divide number of possibilities among themselves? with this way, I guess we can solve openings. for ex: we will try to solve 1.e4-e5 2.f4-exf5 3.Nf3 or Kings Gambit.
100 computers(group A) will focus on 3.g5 and another 100(group B) will focus on 3.d6 and another 100(group C) will focus on 3.d5 ad so on.
Each group will also divide possibilities among themselves and calculate variations. for ex: Group A works on 3.g5 and 10 computers from grup A will try 4.d4 and another 100 will focus on 4. Bc4 and so on. at the end, they will connect their results.
you may: even though we have billions of computers working on Kings Gambit, it is still not enough and possibilities are far more than the number of computers. again: if we have 1000s of AlphaChess computers they can omit unneeded possibilities, they will select best 2 or 3 options among themselves and focus on that. thus, we can solve chess.
p.S:these are just my predixtions, I am programmer or comouter engineer, I may be wrong (which is probably rigjt). if you know I am missing something or do not know something in programming which refutes my predixtions, share please. you are welcome.