4
$\begingroup$

In the circumstances of two perfect AI's playing each other, will white have an inherent advantage? Or can black always play for a stalemate by countering every white strategy?

$\endgroup$
7
$\begingroup$

This relates to the concept of "solved games". In general, two player turn-based games with perfect information - of which chess is an example - can result in all three possible outcomes: a forced win for white, a forced win for black, or a forced draw.

The short, although unsatisfactory answer is that chess is not solved, and it is not clear whether it can be. There is generally thought to be an advantage to white for the first move, so likely results are considered to be a forced win for white, or a forced draw.

No current AI attempts to "solve" chess, although some of the techniques such as MCTS might be adapted theoretically to find a solution, the available computing power to run that search to completion from the start positions is too low by a few orders of magnitude.

| improve this answer | |
$\endgroup$
  • $\begingroup$ There are more possibilities, like an endless match. Moreover, the possibility of win by timeout, that means that winner will be the faster AI. More interesting from the point of view of AI ¿why two perfect AI chess gamers will play chess if they know there other one is also perfect? Both AIs known from start the final result, they can jump directly to it. $\endgroup$ – pasaba por aqui Oct 1 '18 at 14:23
  • $\begingroup$ @pasabaporaqui: I'd like to avoid expanding definitions to include chess rules that the OP did not specify. In addition, the OP specified "perfect" but not "general" as in AGI. Yes if there was an AGI that could also play chess perfectly, it might choose to do things such as not play. However, the concepts of "general" and "perfect" are completely orthogonal here. IMO we are more likely to build a general AI than one that can play chess perfectly. We have functioning examples of general intelligences (ourselves) but no functioning examples of perfect chess players. $\endgroup$ – Neil Slater Oct 1 '18 at 16:33
  • $\begingroup$ @pasabaporaqui Chess can't go on forever. $\endgroup$ – PyRulez Oct 2 '18 at 1:11
1
$\begingroup$

No that doesnt have to be the case because from each machine's point of view the game is different so they will have different possibilities like saying different paths to follow, now something different would be if two machines with the same brain play against the same opponent with same parameters then we can say theboutcome will problably be the same, if they are "programed in a deterministic way"

| improve this answer | |
$\endgroup$
0
$\begingroup$

I'm no expert chess player. Some specialist chess forums also discussed this issue and there's no clear answer.

Due to the high amount of possible moves, I would suggest, that most outcome would probably be a remis. White would have the advantage of starting, but as black as a perfect AI would know which possible strategies are in play it could block every try for a clear winning strategy but probably always trying to avoid a loss and therefore probably reaching just a remis.

Basis for this thought is that black as a reaction can always make the choice with fewest loss possibility so probably most strategies of white would stall.

But now comes the interesting point: I am not quite sure how these AIs would look like. They obviously would have to lean on a strategy to choose between all the possible moves. A pure statistical best outcome algorithm even then would leave tooo much options open, specially in the beginning. So you would be forced to prefer certain strategy choices, and with this you would clearly implementing a bias (what to do if two moves have the same win/loss outcome?). This would make the experiment more random or human and thus no outcome could be predicted. So if both AIs are perfect (and thus identical) the outcome is most probably a draw, but I assume that statistically some games would be won/loss... I think...

EDIT: I just read about googles AlphaZero (a AI which has taught itself Chess, Go and Shogi) and excells in most games against other AIs. But apparently by precalculating not so much possible outcomes... So it could be that there still is no "absolute perfect AI" for this game(s)

| improve this answer | |
$\endgroup$
-1
$\begingroup$

Since both AI know the best possible moves at each step, black would never win as white already knows all games that leads to black's victory and would easily avoid it, but since the black is also a perfect AI, the optimal reponses to white's moves are fixed. So the logical reasoning is these two perfect AI would not make any moves at all if they know their opponent is a perfect AI as well. Both of them know the outcome of the match even before it begins, which my gut feeling says is always a draw. The correct question is does the nth move matter?

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.