Will computers ever solve chess?

Sort:
troy7915
Elroch wrote:

Another open question is "will computers ever solve football?".

 Computers can only solve closed systems/games, like chess,  not open systems like life or games like football.

Elroch
troy7915 wrote:
Elroch wrote:

Another open question is "will computers ever solve football?".

 Computers can only solve closed systems/games, like chess,  not open systems like life or games like football.

Firstly, did you think my question was serious?

Secondly, you are wrong, in that it is certainly feasible for a computer to seek optimal strategies for complex, interactive, goal-oriented activities like football, and it is even just about conceivable that one would find the optimal strategy that likely exists (some extrapolated theorem of game theory), given careful problem definition. One observation as someone familiar with relevant AI techniques is that although it might seem necessary to have full knowledge of the physical capabilities and behaviour of the players, these are things that could be simultaneously learnt while seeking a good strategy. What is essential is having feedback on results, and receiving enough other observations that are relevant (in principle this could simply be video).

Note that this is certainly a very complex task, like others involving interaction with an open environment (such as autonomous driving in unknown places), but most AI researchers would agree with my statement of feasibility.

Note that robot soccer is a long-established test bed for co-operative AI-powered robots, but it is fair to say that thus far the standard of play is not terribly high.

HorribleTomato
YouFreaking wrote:

Leave the koalas outta this, you sicko!

Hey, you brought up the Koalas. What you should be saying is,

Leave the cheese steaks outta this, you sicko!

troy7915
Elroch wrote:
troy7915 wrote:
Elroch wrote:

Another open question is "will computers ever solve football?".

 Computers can only solve closed systems/games, like chess,  not open systems like life or games like football.

Firstly, did you think my question was serious?

Secondly, you are wrong, in that it is certainly feasible for a computer to seek optimal strategies for complex, interactive, goal-oriented activities like football, and it is even just about conceivable that one would find the optimal strategy that likely exists (some extrapolated theorem of game theory), given careful problem definition. One observation as someone familiar with relevant AI techniques is that although it might seem necessary to have full knowledge of the physical capabilities and behaviour of the players, these are things that could be simultaneously learnt while seeking a good strategy. What is essential is having feedback on results, and receiving enough other observations that are relevant (in principle this could simply be video).

Note that this is certainly a very complex task, like others involving interaction with an open environment (such as autonomous driving in unknown places), but most AI researchers would agree with my statement of feasibility.

Note that robot soccer is a long-established test bed for co-operative AI-powered robots, but it is fair to say that thus far the standard of play is not terribly high.

 

 

 

  Obviously computers can help prepare and enhance one’s skills needed in open systems. Nobody’s disputing that.

 But whereas in chess computers can be undefeated, as they already are, whereas in open systems they can never dominate, like in life, vis-à-vis man’s fear of AI taking over the world. It cannot happen.

  Even for learning a difficult skill, computers are useless, for it depends on muscle memory, which depends on hundreds of thousands of repetitions and discovery, insight into how the best execution feels. No AI can replace this practice.

 

Junebug444

Huh. YouFreaking blocked me. Seems like that’s occurring a lot these days.

Elroch
troy7915 wrote:
Elroch wrote:
troy7915 wrote:
Elroch wrote:

Another open question is "will computers ever solve football?".

 Computers can only solve closed systems/games, like chess,  not open systems like life or games like football.

Firstly, did you think my question was serious?

Secondly, you are wrong, in that it is certainly feasible for a computer to seek optimal strategies for complex, interactive, goal-oriented activities like football, and it is even just about conceivable that one would find the optimal strategy that likely exists (some extrapolated theorem of game theory), given careful problem definition. One observation as someone familiar with relevant AI techniques is that although it might seem necessary to have full knowledge of the physical capabilities and behaviour of the players, these are things that could be simultaneously learnt while seeking a good strategy. What is essential is having feedback on results, and receiving enough other observations that are relevant (in principle this could simply be video).

Note that this is certainly a very complex task, like others involving interaction with an open environment (such as autonomous driving in unknown places), but most AI researchers would agree with my statement of feasibility.

Note that robot soccer is a long-established test bed for co-operative AI-powered robots, but it is fair to say that thus far the standard of play is not terribly high.

 

 

 

  Obviously computers can help prepare and enhance one’s skills needed in open systems. Nobody’s disputing that.

 But whereas in chess computers can be undefeated, as they already are, whereas in open systems they can never dominate, like in life, vis-à-vis man’s fear of AI taking over the world. It cannot happen.

  Even for learning a difficult skill, computers are useless, for it depends on muscle memory, which depends on hundreds of thousands of repetitions and discovery, insight into how the best execution feels. No AI can replace this practice.

 

The notion of open-ended AI is a fascinating and challenging one, but it is fundamentally no more impossible than NI (natural intelligence).

Football certainly does not fall into that category. It is a game with rules based on scoring, and scoring provides feedback on the quality of the strategy used by a competitor (eg a team of robots). While much more cumbersome (impractically so, to be frank), it is possible to imagine a successful attack on the problem of playing the game well using an analogous approach to that of AlphaZero.

Much more general objectives are excellent choices for AI. The ability to move around can be targeted and learnt, as can the ability to manipulate objects, without any deeper purposes. These abilities can form part of a modular AI such as those studied in hierarcharchical reinforcement learning. [Incidentally, babies and toddlers use an approach like this].

troy7915

They can offer strategies, but a spontaneous tactic, a perfect execution in an impossible position, that will always depend on the alertness of the player, there they offer no support.. And it depends on the player: those with a high degree of technique are better left to play free of any strategy, with others serving them balls as often as they can. The rest of the players, though or the whole team if such a player is missing,  must have a clear strategy in mind.

 But individually, nothing can replace repetition and insight. Yes, now they have special balls that offer you feedback, but it’s garbage compared to the actual muscle feel which is irreplaceable and always will be.

 

 As for life in general, sure, AI can accomplish various tasks, but there will always be a lack of answer to the deeper question: why do that? And why do that? And why that? Ultimately, it is meaningless to AI.

 

 After all, it is not a question of survival for them, which is the ultimate motivation for humans and other life forms.

Prometheus_Fuschs
jbent02 escribió:
vickalan wrote:
btickler wrote:

...you should draw the tree as it really exists ...probably far wider...

The diagram isn't to represent the width of the game tree - it shows that it might be possible to solve chess without examining the entire tree. Therefore, any conclusion (such as yours) that chess cannot be solved because the game tree is too large is based on a faulty assumption.

To solve chess, you have to go through every line, leaving one line undone means its not actually solved. 

 

Not really. If you conclude it is a forced checkmate in X amount of moves you can ditch that part of the tree.

Prometheus_Fuschs
troy7915 escribió:

Note to self: ‘math has certainties’? Just earlier in this thread a mathematician agreed that all math is based on beliefs, it’s foundation, something which I pointed out earlier.

 If these ‘certainties’ are born out from beliefs, and develop from there into a sound system, then are they certainties?

 

Well, math is based on axioms which are suppositons but there simply is no other known way to do maths. This is called formalism and in case you wonder, there aren't any better options ATM and probably never will.

troy7915

The point was that math does not have certainties. Regardless of not finding other options, that fact isn’t affected.

Elroch

Nothing is certain because we can't eliminate the possibility that all our thoughts are delusional.

Ignoring that, we have the useful reinforcement that mathematical proofs are computer checkable, if formalised completely (and many have been). Given this, the only way in which it makes sense for maths not to be certain is for it to be inconsistent, and in major cases - eg the maths of the natural numbers - this is something you could safely bet your life on not being so.

As for axioms being "true", we don't have to worry about that - indeed it is not really meaningful - a mathematical theorem applies to any model where the axioms it depends on are true.

So we don't have to worry whether Peano's axioms are really about the natural numbers, or whether the natural numbers exist. rather all theorems derived from Peano's axioms apply to any model where they happen to be true. There isn't really a single thing called the natural numbers, there is a set of properties of a model, and if those properties are satisfied, we consider it a model of the natural numbers.

Elroch
Prometheus_Fuschs wrote:
jbent02 escribió:
vickalan wrote:
btickler wrote:

...you should draw the tree as it really exists ...probably far wider...

The diagram isn't to represent the width of the game tree - it shows that it might be possible to solve chess without examining the entire tree. Therefore, any conclusion (such as yours) that chess cannot be solved because the game tree is too large is based on a faulty assumption.

To solve chess, you have to go through every line, leaving one line undone means its not actually solved. 

 

Not really. If you conclude it is a forced checkmate in X amount of moves you can ditch that part of the tree.

To know that, you need to explicitly deal with every possible way in which the opponent might play from that position. Very few short cuts are to be found (unlike in mathematics, where the axioms are far less arbitrary).

chessspy1

Nothing is certain because we can't eliminate the possibility that all our thoughts are delusional.

No, this was solved a long time ago.

In order for us to doubt our existence, there must be a 'thinker' to formulate that doubt etc.

Surely we can discuss the possibility of computers solving chess without going through all this old stuff which has been so done.

I would think that code-breaking programs would have to deal with a similar number of possibilities,(as are in chess games) and they seem to be able to find solutions. 

PlayerName01

No they won't.  They don't need to.

troy7915
Elroch wrote:

Nothing is certain because we can't eliminate the possibility that all our thoughts are delusional.

 

 

 

  Oh, yes, we can.

 

  But that is beyond the grasp of a mathematician or any specialist in a small ( or big ) field of expertise—the field of knowledge is very, very small, limited, narrow and superficial.

  One must discover another field.

Elroch

It is amusing that someone would seriously think that the possibility of thoughts being mere delusions would not make all conclusions at least to some extent unreliable.

It is perfectly reasonable to ignore this possibility: there is little harm to be done by doing so!

troy7915

Of course, all conclusions are unreliable!

 

 Which is why we said that another field must be discovered. 

  Why assume this is a conclusion?

Elroch

Perhaps I misunderstood. I thought you meant that we could use our (possibly unreliable) interface with the world to construct some alternative approach that would allow to be justifiably 100% confident of our conclusions. This is fundamentally impossible, if there is always the possibility of being fooled.

Note that I am playing devil's advocate here: I personally believe that such things as our understanding of mathematics are generally perfectly correct, because our minds (or, more precisely, the minds of those who have the best knowledge) are reliable enough to give confidence in this. The main reason for this is empirical: it seems more likely that continued success and consistency is the result of being genuinely valid than an elaborate delusion.

troy7915

 No, not conclusions, that part is clear. Math seems to be correct, I have no problem with that, although it seems strange to be empirical at its root. Regardless, it has its applications, which must continue in that field.

  The point being made was being rooted in beliefs, at its root, it is no wonder that a mathematician’s daily life is also rooted in beliefs. Not talking about inevitable, harmless beliefs like making an appointment and believing it can happen—which involves a lot of factors coming together—or boarding a plane, which is based on the belief that the pilot is a professional, and there are no destabilizing factors, human or natural, and so forth.

  We are talking about other beliefs, which by their very nature divide people, start wars and so lead to death. I mean, there is this image about the scientist that he is neutral and acts logically, rationally and looks at the world impartially. Which is utterly false.

He only does that in a very limited field, and in some branches of science, when it comes to interpreting the results of certain experiments and offer suggestions for the real world, his bias as a human being, which is knowledge of psychological nature,  completely wastes the results of his otherwise neutral experiments.

 But as a human being, the scientist is far from being unbiased.

 

I was listening the other day to a theoretical physicist, and he seemed very sharp and neutral, unbiased, in talking about various possibilities of our Universe and beyond that. But as soon as the discussion turned to religion or some other mundane day-to-day aspect, he was back to belief, and not of the inevitable kind.

USArmyParatrooper
troy7915 wrote:
  1. No, not conclusions, that part is clear. Math seems to be correct, I have no problem with that, although it seems strange to be empirical at its root. Regardless, it has its applications, which must continue in that field.

  The point being made was being rooted in beliefs, at its root, it is no wonder that a mathematician’s daily life is also rooted in beliefs. Not talking about inevitable, harmless beliefs like making an appointment and believing it can happen—which involves a lot of factors coming together—or boarding a plane, which is based on the belief that the pilot is a professional, and there are no destabilizing factors, human or natural, and so forth.

  We are talking about other beliefs, which by their very nature divide people, start wars and so lead to death. I mean, there is this image about the scientist that he is neutral and acts logically, rationally and looks at the world impartially. Which is utterly false.

He only does that in a very limited field, and in some branches of science, when it comes to interpreting the results of certain experiments and offer suggestions for the real world, his bias as a human being, which is knowledge of psychological nature,  completely wastes the results of his otherwise neutral experiments.

 But as a human being, the scientist is far from being unbiased.

 

I was listening the other day to a theoretical physicist, and he seemed very sharp and neutral, unbiased, in talking about various possibilities of our Universe and beyond that. But as soon as the discussion turned to religion or some other mundane day-to-day aspect, he was back to belief, and not of the inevitable kind.

This is false. There’s literally nothing I believe in that isn’t proportional to the size of the claim and the size of the evidence that supports it.