Will computers ever solve chess? - Chess Forums - Page 187

USArmyParatrooper

Mar 21, 2018

0

#3721

godsofhell1235 wrote:

I'm using EGTB as shorthand for a 32 man EGTB.

Remember a 32 man EGTB has the true evaluation for every single possible position in chess (about 10^42 IIRC).

Of course the EG part of the designation is a little silly, because by the time you have a 32-man EGTB we're not talking about the endgame anymore, but I use EGTB to remind you that the only evaluations really are "mate in __" and "draw" and that every possible position has been evaluated and stored.

---

And yes, now, for the 3rd time, I'm telling you if every single position in the game were evaluated as drawn, then we would say it's a perfect game (under our current definition of perfect).

Yes, under the more usual definition of perfect, a player will play to win, and never make a mistake. This is much trickier to make a definition for since playing for a win often means playing an objectively less-than-best move (because it creates more practical problems for the opponent).

Playing for a win can only mean playing objectively less than perfect moves, if you are playing against something or someone that is fallible.

In terms of the thread topic, which is chess being solved, playing for a win can only objectively mean playing the (objectively) best moves.

Do I think games have been played where current computer assessments from start to finish is near 0, of course. I just think that’s a strange criteria for which to deem a game “perfect.” But if that’s where you’ve been going with us the whole time, I don’t disagree with you.

godsofhell1235

Mar 21, 2018

0

#3722

For example, lets say a players has two options.

One is a mate in 20, but for every single of the 20 moves, the 2nd best move loses (or draws) instantly. Meaning the player must find all 20 moves.

The 2nd choice is a move that objectively makes the game a draw, but now the opponent is forced to play perfectly for 20 moves or they will instantly lose.

We will also say that the position is very difficult, and requires tremendous calculation for either player to find the best moves.

---

The practical choice is clear. Everyone would choose the move that makes the game objectively drawn, and every person who chose this would go on to win the game.

And even further, choosing the move that is a forced mate would be called a horrible blunder.

---

To a lesser extent this happens all the time in top level chess.

Yesterday in the Candidates tournament, Karjakin chose the move h4 vs Kramnik's ridiculously solid semi-tarrasch.

Karjakin won even though h4 very clearly not objectively best.

During that same game when Kramnik chose the move f5, I'm reasonably certain Kramnik knew this move was not objectively best, possibly even fairly bad, but by sharpening the game he purposefully created more chances for both sides to blunder and therefore more chances for himself to win.

---

Even comparatively very low level players like myself make decisions like this, if only in our opening repertoire.

USArmyParatrooper

Mar 21, 2018

0

#3723

godsofhell1235 wrote:

USArmyParatrooper wrote:

And even if it determined every position from start to finish was drawn, let’s say on move 20 White had a candidate move that would’ve been assessed +1.16, but instead chose a move that was assessed 0.08. Did he play the perfect move?

Taking into account conventional engine evaluations this would lead to a much more pleasing definition of perfect.

But it's also a lot harder to properly define.

First of all, moves that are difficult for one opponent may be easier for another. How do we quantify this?

Secondly if we only go by "difficulty" then there's the problem of games won due to purposefully giving yourself a losing position (like some king's gambit variations) but for the purpose of greatly increasing the chance of a blunder from your opponent.

I’m speaking from the standpoint of objectivity. The objectively best, most perfect moves. The thread topic, chest being solved. Who sits on the other side becomes irrelevant. Psychology becomes irrelevant. Mathematically, what is the objectively best move, which will achieve the most desirable outcome of the game.

godsofhell1235

Mar 21, 2018

0

#3724

USArmyParatrooper wrote:

I just think that’s a strange criteria for which to deem a game “perfect.” But if that’s where you’ve been going with us the whole time, I don’t disagree with you.

My first foray into this topic I started off by not using this definition of perfect.

You politely informed me that only two real evaluations exist (I agreed) and if it never changes the whole game then it's a perfect game (I agreed).

Why you suddenly want to use a different definition of perfect I don't know, but I agree that the one we've been using isn't anything like the colloquial way we talk about a perfect game.

USArmyParatrooper

Mar 21, 2018

0

#3725

It’s your turn. You have mate in 92 moves, the first of which is pushing Harry the H-pawn. But you miss it and play something else that humans and current computers approve of.

You have not played a perfect game.

godsofhell1235

Mar 21, 2018

0

#3726

USArmyParatrooper wrote:

godsofhell1235 wrote:

USArmyParatrooper wrote:

And even if it determined every position from start to finish was drawn, let’s say on move 20 White had a candidate move that would’ve been assessed +1.16, but instead chose a move that was assessed 0.08. Did he play the perfect move?

Taking into account conventional engine evaluations this would lead to a much more pleasing definition of perfect.

But it's also a lot harder to properly define.

First of all, moves that are difficult for one opponent may be easier for another. How do we quantify this?

Secondly if we only go by "difficulty" then there's the problem of games won due to purposefully giving yourself a losing position (like some king's gambit variations) but for the purpose of greatly increasing the chance of a blunder from your opponent.

I’m speaking from the standpoint of objectivity. The objectively best, most perfect moves. The thread topic, chest being solved. Who sits on the other side becomes irrelevant. Psychology becomes irrelevant. Mathematically, what is the objectively best move, which will achieve the most desirable outcome of the game.

Mathematically, and without any regard for the opponent or for psychology, there are no such evaluations as 0.08 or 1.16 (see your post #4994)

If both moves are drawn, then both moves are equally good.

godsofhell1235

Mar 21, 2018

0

#3727

USArmyParatrooper wrote:

It’s your turn. You have mate in 92 moves, the first of which is pushing Harry the H-pawn. But you miss it and play something else that humans and current computers approve of.

You have not played a perfect game.

I disagree.

You didn't say that the move I played was not also a mate in 92.

In any case, this definition of perfect also runs into seeming contradictions when applied to practical chess, so I wonder if I start using it, if you wont start to disagree with this definition 20 posts from now.

USArmyParatrooper

Mar 21, 2018

0

#3728

godsofhell1235 wrote:

USArmyParatrooper wrote:

I just think that’s a strange criteria for which to deem a game “perfect.” But if that’s where you’ve been going with us the whole time, I don’t disagree with you.

My first foray into this topic I started off by not using this definition of perfect.

You politely informed me that only two real evaluations exist (I agreed) and if it never changes the whole game then it's a perfect game (I agreed).

Why you suddenly want to use a different definition of perfect I don't know, but I agree that the one we've been using isn't anything like the colloquial way we talk about a perfect game.

Well, you see I’m not changing it. You would still have to establish that at no point during the game, did either side have an unfathomably long win by force that no human or current computer can find. Which we absolutely don’t know. Nor do I think we have nearly enough information to even claim it’s likely.

USArmyParatrooper

Mar 21, 2018

0

#3729

godsofhell1235 wrote:

USArmyParatrooper wrote:

It’s your turn. You have mate in 92 moves, the first of which is pushing Harry the H-pawn. But you miss it and play something else that humans and current computers approve of.

You have not played a perfect game.

I disagree.

You didn't say that the move I played was not also a mate in 92.

In any case, this definition of perfect also runs into seeming contradictions when applied to practical chess, so I wonder if I start using it, if you wont start to disagree with this definition 20 posts from now.

By any definition, how can you say any game is perfect if they miss forced mate?

godsofhell1235

Mar 21, 2018

0

#3730

USArmyParatrooper wrote:

godsofhell1235 wrote:

USArmyParatrooper wrote:

I just think that’s a strange criteria for which to deem a game “perfect.” But if that’s where you’ve been going with us the whole time, I don’t disagree with you.

My first foray into this topic I started off by not using this definition of perfect.

You politely informed me that only two real evaluations exist (I agreed) and if it never changes the whole game then it's a perfect game (I agreed).

Why you suddenly want to use a different definition of perfect I don't know, but I agree that the one we've been using isn't anything like the colloquial way we talk about a perfect game.

Well, you see I’m not changing it. You would still have to establish that at no point during the game, did either side have an unfathomably long win by force that no human or current computer can find. Which we absolutely don’t know. Nor do I think we have nearly enough information to even claim it’s likely.

I gave criteria for how we can argue for likeliness.

I also challenged you to offer a counter example by using this criteria to set up a position that is not a draw using the EGTBs we currently have.

godsofhell1235

Mar 21, 2018

0

#3731

USArmyParatrooper wrote:

godsofhell1235 wrote:

USArmyParatrooper wrote:

It’s your turn. You have mate in 92 moves, the first of which is pushing Harry the H-pawn. But you miss it and play something else that humans and current computers approve of.

You have not played a perfect game.

I disagree.

You didn't say that the move I played was not also a mate in 92.

In any case, this definition of perfect also runs into seeming contradictions when applied to practical chess, so I wonder if I start using it, if you wont start to disagree with this definition 20 posts from now.

By any definition, how can you say any game is perfect if they miss forced mate?

Not many people would call this game perfect, for example.

USArmyParatrooper

Mar 21, 2018

0

#3732

godsofhell1235 wrote:

USArmyParatrooper wrote:

godsofhell1235 wrote:

USArmyParatrooper wrote:

And even if it determined every position from start to finish was drawn, let’s say on move 20 White had a candidate move that would’ve been assessed +1.16, but instead chose a move that was assessed 0.08. Did he play the perfect move?

Taking into account conventional engine evaluations this would lead to a much more pleasing definition of perfect.

But it's also a lot harder to properly define.

First of all, moves that are difficult for one opponent may be easier for another. How do we quantify this?

Secondly if we only go by "difficulty" then there's the problem of games won due to purposefully giving yourself a losing position (like some king's gambit variations) but for the purpose of greatly increasing the chance of a blunder from your opponent.

I’m speaking from the standpoint of objectivity. The objectively best, most perfect moves. The thread topic, chest being solved. Who sits on the other side becomes irrelevant. Psychology becomes irrelevant. Mathematically, what is the objectively best move, which will achieve the most desirable outcome of the game.

Mathematically, and without any regard for the opponent or for psychology, there are no such evaluations as 0.08 or 1.16 (see your post #4994)

If both moves are drawn, then both moves are equally good.

I was using your current evaluation tool, a system that hasn’t solved chess and still uses subjective positional analysis.

Hal would only move between literally drawn 0.00, and forced mate.

godsofhell1235

Mar 21, 2018

0

#3733

Ok, in #4944 I thought you were implying that 1.16 was better than 0.08 when determining whether a game is perfect or not. I guess I misunderstood you.

So you agree with me that if the eval never changes from draw, that it's a perfect game?

USArmyParatrooper

Mar 21, 2018

0

#3734

godsofhell1235 wrote:

USArmyParatrooper wrote:

godsofhell1235 wrote:

USArmyParatrooper wrote:

I just think that’s a strange criteria for which to deem a game “perfect.” But if that’s where you’ve been going with us the whole time, I don’t disagree with you.

My first foray into this topic I started off by not using this definition of perfect.

You politely informed me that only two real evaluations exist (I agreed) and if it never changes the whole game then it's a perfect game (I agreed).

Why you suddenly want to use a different definition of perfect I don't know, but I agree that the one we've been using isn't anything like the colloquial way we talk about a perfect game.

Well, you see I’m not changing it. You would still have to establish that at no point during the game, did either side have an unfathomably long win by force that no human or current computer can find. Which we absolutely don’t know. Nor do I think we have nearly enough information to even claim it’s likely.

I gave criteria for how we can argue for likeliness.

I also challenged you to offer a counter example by using this criteria to set up a position that is not a draw using the EGTBs we currently have.

You can’t even calculate likelihood. At best you can give a guess. None of us have any idea if on move 3, 4, 5, 12, etc. there isn’t some extraordinarily long mating net that no tools can see at the present moment.

USArmyParatrooper

Mar 21, 2018

0

#3735

godsofhell1235 wrote:

Ok, in #4944 I thought you were implying that 1.16 was better than 0.08 when determining whether a game is perfect or not. I guess I misunderstood you.

So you agree with me that if the eval never changes from draw, that it's a perfect game?

I would never use the word “perfect“ in that manner. But I agree that plenty of games have been played that were assessed to be near 0 (by current engines) from start to finish.

Just for me, that’s really butchering the word “perfect,” especially in a thread that is discussing solved chess.

DiogenesDue

Mar 21, 2018

0

#3736

vickalan wrote:

btickler wrote:

...According to whom?...

Read the forum. Unlike here, where anyone can post anything they want with no provision for expert review, the computer science forum is subject to peer-review.

As we have already established long ago, Stack Exchange is not a peer-reviewed science journal...it's a website for neophyte developers to get their questions answered, usually by other neophyte developers. The people posting there have no more information or expertise on the subject than you or I or anyone else on this thread does.

Let me guess, you probably think that all information on Linked In is vetted by "professional businessmen", right? That Quora is "peer reviewed"? Do you even understand what these terms you toss around mean? Sometimes it's like you've been hiding that English is a second language for you or something...

If you think that thread holds better, peer-reviewewd info, please...go right ahead and point out a single thing they have said that is more advanced or worthwhile than the (also useless) discussion here. The fact is, you know there isn't any, or you would have already quoted it here long ago in your diatribes. This is just another forum thread you Googled up in your failed attempts to prove/disprove something...anything...on this thread. The alternative is that you posted the question yourself anonymously a couple years back in a futile attempt to have others do your arguing here without you having to think or research yourself. Either way, it's pathetic.

edilio134

Mar 21, 2018

0

#3737

--------> None of us have any idea if on move 3, 4, 5, 12, etc. there isn’t some extraordinarily long mating net that no tools can see at the present moment.

None of us have any idea about multiverse life after Death angels and phantoms

ponz111

Mar 21, 2018

0

#3738

godsofhell1235 wrote:

USArmyParatrooper wrote:

godsofhell1235 wrote:

s23bog wrote:

So, no, you do not assume the games are perfect after passing a test or two. In fact, after passing all tests known to man, we cannot assume the game is perfect.

Again, I think you're giving undue weight to what it means for a game to be perfect under our rather lenient definition.

Ok maybe we can't say it's very likely to be perfect, but we can make rational arguments, using real criteria, that argue for a certain likelihood (certainly above impossible) that it's perfect.

After that we just observe that millions/over a billion games have been played, and then we claim that it's statistically likely (and very much not impossible) that a perfect game has been played.

Using any tools at your disposal, any at all. Can you tell me if in the kings gambit, after 2. e4 there are any lines where black wins by force?

I've actually been looking at KG lines somewhat seriously.

Very annoyingly it seems white can always equalize.

(But even more annoying from a practical perspective, white can purposefully not equalize, and get very dangerous initiatives in completely chaotic positions.)

---

Anyway I don't know why we're using an opening that's both non-symmetrical and creates serious potential threats (moving the f pawn is not equivalent to moving the c pawn due to the king and queen making the board asymmetrical along the vertical axis). i.e. the open h4-e1 diagonal.

try this variation for Black:

godsofhell1235

Mar 21, 2018

0

#3739

ponz111 wrote:

godsofhell1235 wrote:

USArmyParatrooper wrote:

godsofhell1235 wrote:

s23bog wrote:

So, no, you do not assume the games are perfect after passing a test or two. In fact, after passing all tests known to man, we cannot assume the game is perfect.

Again, I think you're giving undue weight to what it means for a game to be perfect under our rather lenient definition.

Ok maybe we can't say it's very likely to be perfect, but we can make rational arguments, using real criteria, that argue for a certain likelihood (certainly above impossible) that it's perfect.

After that we just observe that millions/over a billion games have been played, and then we claim that it's statistically likely (and very much not impossible) that a perfect game has been played.

Using any tools at your disposal, any at all. Can you tell me if in the kings gambit, after 2. e4 there are any lines where black wins by force?

I've actually been looking at KG lines somewhat seriously.

Very annoyingly it seems white can always equalize.

(But even more annoying from a practical perspective, white can purposefully not equalize, and get very dangerous initiatives in completely chaotic positions.)

---

Anyway I don't know why we're using an opening that's both non-symmetrical and creates serious potential threats (moving the f pawn is not equivalent to moving the c pawn due to the king and queen making the board asymmetrical along the vertical axis). i.e. the open h4-e1 diagonal.

try this variation for Black:

Thanks, I recently had a guy suggest that to me.

Although he strangly claimed that after 4.e5 black should play 4...Ne4. I think he was mistaken.

edilio134

Mar 21, 2018

0

#3740

e5 is a blunder