This morning, my wife (who has never played chess) asked me the following question after I had told her a few things about AlphaZero.
"If AlphaZero played against itself in some form, what would the result be?"
I replied that if both sides played perfectly (or even simply to an identical very high standard), the result would almost always be a draw because it is considered that the advantage of white having the first move is insufficient to gain a victory. Perhaps in a long enough series of games, there might be a very small number of white wins.
However, I then realised that I was simply stating the long-accepted view, and that I have no idea if there is nowadays clear evidence of this as a fact. Perhaps there is and perhaps there isn't: I had to admit that I don't actually know. For all I know, it might even turn out that, right at the start, white is technically in zugzwang!
So... What is the actual situation?
Chess is probably a draw with perfect play.
But if you put AZ to play current opening theory (they did it in the full recent paper) the number of decisive games increases a lot. Which only means chess is quite complex and AZ is still far from perfect play. JMO
By the way AZ even didn't see the mate in WC game 6. So very far from perfect play, inhuman good but not yet perfect.
white wins, so simple :-)
Seriously, to many draws, who more play whites will win.
is "chess is a draw" the most widely accepted view among strong players? yes
is it a proven fact? no
I believe that Black has a slight advantage over white ( objectively speaking ) and over the course of 1000 games between 'perfected A.I.' Black would win the tournament by 1 or 2 games... of course most of the games would be drawn. Of course in 'human play' white has the advantage because of the subjective nature of 'human play'... initiative provokes, initiative conflates... humans are capable of hallucinations and 'initiative' is a tool in the magicians bag of tricks. Perfected A.I. plays chess in a hyper-objective and a non-emotional reactive or reflexive manner... ' every move becomes an expression of weakness to be exploited... there is no perfect move but rather the exploitation of an imperfect move. The true nature of chess is reactive in the realm of perfected A.I. - This is just speculation... perhaps I'm in the mood to play at devils' advocate. White having an advantage may yet to be proven a fallacy. Who knows?
> If AlphaZero played against itself in some form, what would the result be?
AZ actually trained by playing against itself with all the results. This doesn't have anything to do with the question about the perfect result - AZ is still a practical real-life entity using clever euristics of its NN and by no means full solution like endgame TBs.
There is very clear evidence to the fact that chess is a draw. The level of play is rising and rising with time, some days this was the level of the best humans, and now it's the level of the best engines on the best hardware. And the percentage of draws between higher level opponents gets higher and higher. This has one realistic explanation: the chess is a draw, and because the opponents make less mistakes, more and more games end without ever stepping aside from the true result of the game.
However, the chess is a well-defined mathematical game, so people usually want to get the proper proof. This is understandable, but it's likely that chess will never be solved in any sense due to the enourmous complexity of the game tree. There are many other things in this Universe that are also cannot be realistically mathematically proven, but we treat them as facts because the evidence for them in enourmous and despite they can be theoretically proven wrong. I think we should treat the statement "chess is a draw" as such naturalistic fact.
P.S. It is NOT *proven* that the initial position is not losing for white, being a reciprocal zugzwang. However, all our chess experience says that this would be absurd. Will you really doubt this fact just because it's not mathematically proven?
There is no mathematical proof that in all cases, a perfectly played game is a draw. it is more speculation on people's part. The tree that would be needed would be enormous and might be beyond any current computer.
Imagine in the Hitchhiker's Guide to the Galaxy movie if they had asked Deep Thought to play chess against itself for 7 million years with no other tasks to focus on during that time period and then ask Deep Thought to report what would be the result of perfect play on both parties.
Suppose after 7 million years of playing every possible combination and it can be shown that with perfect play, the game of chess is a draw but only if say one specific move is played (say e4, d4 or c4 or some other move).
Would you want to wait for 7 million years just to find out the result? Not to mention that the tree could be several branches long and even the slightest variation from perfection (say at move 34) it could change the end result of the match.
So it is just a guess what the ultimate result will be.
Researchers solved checkers. Maybe one day it will be chess.
Humans will likely be extinct or we'll be considered ancient ancestors before chess is solved.
However the argument "you can't prove chess isn't a draw because there isn't a complete 32 piece table base" is technically true but you'd have to be a moron to think chess isn't a draw. I'm not talking to your internet troll personality here, it just isn't a forced win. You need to look into current table bases and start to think of it based on odds and not "oh maybe there's a tactic that forces this or that". If that's the case I'm wrong, but I'm certain it isn't because I tend to think about things logically and don't let my internet troll out to play very often.
we have complete 7 piece table base where many likely positions are forced draws. You can't tell me with perfect play white or black can't achieve one of these forced drawn positions. There is no tactic at the start that forces your opponent to move into a series of moves that isn't one of these drawn positions. There are too many of them to forcefully avoid them.
The more we complete the table base the more we will see drawn middle games and endgames that a perfect engine would be able to aim toward and achieve. It's absurd to believe that from move 1 these countless positions are avoided by force by either side.
Going by numbers alone the odds of chess being a forced win approaches zero.
You don't have a very good grasp of tactics if you think there is a tactic at the start that avoids all of these forced draws. Once table bases reach 32 (they wont) I believe we will see every first move combo from both sides are a draw, even 1.f4 simply because the advantage it gives isn't enough to avoid all of the drawn positions that it would transpose into with perfect play.
