Re: Can Nosovs be beaten?
Posted:
Jun 22, 2010, 4:51 AM
I'm going to go out on a limb here and say that for the dozen or so good* black 1sts and handful of viable** black 2nds for each one of those 1sts, probably less than 5% have a so-called "perfect play" white response that is accepted as such by the majority of pente experts. I would even go far as to say that a substantial number of them are not actually "perfect" (namely, that somewhere along the line black could change a move to make a winning position), so what piecraft is saying has no application to the actual game as nosovs and other mere mortals play it. There are very few well-played openings where black is clearly beaten before the 5th move, and by then black has rejected dozens of viable alternatives.
*: L10, M10, N10, O10, L9, M9, N9, O9, M8, N8, O8, N7, O7 **: pretty much any move that defends against white's opening attack while giving opportunities for an initiative-stealing parry
It's no use going back to yesterday, because I was a different person then.
Re: Can Nosovs be beaten?
Posted:
Jun 22, 2010, 12:38 PM
Well this is interesting. It seems from this that there are probably no truly known perfect lines. And the ones that the experts think are perfect, may not actually be perfect. Alison's point that there is no way to prove this stuff is key here I think.
So, if there are no known perfect lines then obviously P1 winning is not at the discretion of the expert player.
Does this mean that the game is not really very close to being solve at all? Because some talk about shutting down the database suggests that many think it is close to being solved. What do we make of this? If there are no known (or hardly any) perfect lines, then how close can it really be to being a done deal?
Perhaps the database really only contains a small percentage of possible strong lines, but human beings are very poor at finding improvements and so progress is slow now even though there is still much to discover.
Maybe the game just seems to be solved (or nearly)to some people, but in reality we have only just begun to scratch the surface.
Posts:
2,220
Registered:
Mar 4, 2007
From:
San Francisco
Age:
45 Home page
Re: Can Nosovs be beaten?
Posted:
Jun 22, 2010, 2:18 PM
no,.. there are perfect play lines that are known.. to clarify how much we know would take a lot of explaining. im just not sure where to begin here, it seems like a can of worms if opened..
Scire hostis animum - Intelligere ludum - Nosce te ipsum - Prima moventur conciliat - Nolite errare
Posts:
542
Registered:
May 9, 2002
From:
Northeast USA
Re: Can Nosovs be beaten?
Posted:
Jun 22, 2010, 11:43 PM
To answer this whole debate -- to my knowledge no one has solved the game of Pente.
I'm using "solved" here the way you would think of a computer solving the game. Suppose you are playing against a Level 12 opponent (mm_ai12) which looks 12 frames ahead (6 moves each) and crunches through all reasonably possible moves, then chooses what it thinks is the "best" move based on its algorithms and calculations. If at some point, the computer finds a move that leads to a win 100% of the time in less than 6 moves, it has "solved" the game from that point forward and will not lose. Furthermore, the game goes from taking several minutes "thinking" about the move to a state where it is moving in less than one second, because it is no longer "thinking" -- it "knows" which moves are winning moves no matter what.
Now consider an assumption that P1 should win every game in 27 moves or less with perfect play (an old estimate that I remember hearing from Pente experts of a prior era). If this is true, then you should be able to set the computer to play at Level 54, and as soon as it plays 1. K10 it "knows" that it will win because it has just calculated through all possible moves in the game and can guarantee victory in every scenario. At this point the computer has "solved" the game. However, at current supercomputing powers it would take many many centuries to complete such a calculation.
No players have "solved" the game in this manner. But expert players understand that P1 should always win and are strong enough players to use memorized moves combined with important concepts to demonstrate this futility of the game a vast majority of the time. When they fail to do so, they are always able to identify the mistake that led to the loss. As player 2, they understand that they cannot win unless the opponent makes a mistake and therefore the game feels futile to play. In this abstract way, expert players feel that the game is "solved".
Re: Can Nosovs be beaten?
Posted:
Jun 23, 2010, 7:39 AM
@piecraft:
It's not that there are no or very few known perfect lines--the fact is, however, that in the opening, a smart p2 will steer away from them towards less familiar (but ideally not much weaker) lines where p1's ideal response has not been discovered and proven.
It's no use going back to yesterday, because I was a different person then.
Re: Can Nosovs be beaten?
Posted:
Jun 24, 2010, 1:39 PM
ok karlw, I know where you are coming from there. I think we need to agree on terms then, otherwise this discussion is too hard to make sense of.
If we agree that P2 starts by default with a loss against a perfectly played P1, then this means that perfect lines must exist.
Let us assume that a 'line' is defined as a set of six consecutive moves by P1 and P2 starting with K10, and that starting from that position there are no responses that give P2 a win if P1 plays only the preset responses to anything P2 does from then on. This constitutes what I would call a perfect line.
When I talk about a perfect line, I am talking about all the variations under that line. This includes all possible responses for P2 on the next move and all the moves after that.
So if perfect lines exist, and they must if P2 is default loss to perfect play, then it comes back to whether a human being is capable of knowing such a line and all the variations under it, such that playing that line will (baring error - forgetting the right moves) always result in a win.
From what I am hearing, it seems that the experts think that not only does no-one know a perfect line they way I am defining it, but that perhaps no-one can know because this may be beyond human capability.
From what you are saying karlw, the P2 player will have these 6 moves (using this example) to 'steer' the game away from known perfect lines. But what if we reduced the number of opportunities P2 had to do this? What if we instead said that our line commences from the 5th move? Or the 3rd? Is this possible? I mean is it within human capability to know that much of the game tree of pente to thwart all misdirections by P2 from the 3rd move onwards? I don't think this is possible, do you?
Posts:
542
Registered:
May 9, 2002
From:
Northeast USA
Re: Can Nosovs be beaten?
Posted:
Jun 24, 2010, 7:45 PM
Piecraft, I think you are on the right track. The only flaw in your thinking is that you are setting a limit on how far along a "line" goes. For example, memorizing all possible moves for the first 6 moves for each player (which is impossible anyway) or memorizing moves starting after the 3rd move -- these are limitations that can be used against you and do not qualify as a perfect line.
In fact, this is precisely how computer opponents are beaten. Suppose you are playing against a computer opponent at Level 12 (looks ahead 6 moves). Assume that it selects an opening from its opening book that is not flawed. Then, if it looks ahead 6 moves and selects a move that it calculates to be a winning move no matter what happens in those next 6 moves, it could still be a losing move. An expert human player simply looks further out and finds a flaw in the computer's move that the computer could not see because it was not looking far enough into the future.
If you look at recent chess matches where a Grandmaster plays against a supercomputer, this becomes the major strategy. I remember following one in particular where the human, playing as Player 2, had solidified a neutral position by the midgame and it appeared that the game would end in a draw. Meanwhile, the human started methodically marching three of his pawns on one side down the board, seemingly harmless given the current position. The computer assumed that these were just moves to perserve the status quo of a drawn game and proceeded to play a couple of meaningless moves that maintained its current position. Over a dozen moves later it became apparent that the advancement of these pawns were creating an opening for an attack. When this buildup first started, the end result was too far out into the future for the computer to see and so it was disregarding these moves for at least a few moves before the danger drifted into its search horizon -- and by then it was too late.
We assume that some day machines will exist that will be so powerful that it can solve these games by brute force, looking so far into the future that the end of the game is seen in every scenario -- at that point a "perfect line" can be selected that the other player will not be able to thwart no matter what it tries. But if you ever take a look at the insane number of calculations it would take to do this it is clear that we are not even close to that point yet.
Humans, on the other hand, have the advantage of logical thought ... algorithms that can change on the fly given the situation. Therefore, an expert player is also looking deep into the future before making a move, but is doing so more efficiently. For example, instead of looking at all possible moves and then trying to see what would happen with all possible responses to every one of those moves and then how he would respond to every one of those responses to every one of the possible moves ... and so on, a human player might only look at what he considers to be the three or four "best" moves and then tries to determine a couple of "best" responses to those moves, and so on. Looking very narrow, but deep compared to how a computer might approach the problem.
So, how does a human determine "best" moves and "best" responses, etc? Mostly relying on experience and using a couple of methods. First, he might rely on relatively few memorized "common" positions and play a predetermined move. Once this option is no longer available, he must rely on his understanding of basic and advanced concepts of the game and find what "looks" like a good move based on his knowledge and ability to apply such concepts.
Computers don't seem to be able to do this for complex games since it would require an extremely complex algorithm that a human would have to create, which would be nearly impossible to do flawlessly. Instead, algorithms attempt to evaluate the strength of a resulting position after a move is played but can only do so by looking at all reasonably possible moves and then again and again however deep it will search.
Based on this, you can see that humans do not have the entire game memorized, but can become extremely skilled at recognizing winning moves as the game unfolds. Therefore, over a span of hundreds of games it makes sense that even the best human players will eventually make a mistake. The players that make such mistakes with the lowest frequency will end up with the highest rating and will be considered the best players.
Although the game has not been "solved", it still feels that way to many expert players trying to beat other expert players as Player 2. Even though all perfect lines are not known and memorized, it is assumed that an expert player will be able to stay on a perfect line in most games and that's when the futility of the game becomes apparent.
That's why you'll sometimes hear expert players talking about preparing for a match against a specific opponent. It becomes a game of trying to determine the weaknesses of that particular human opponent and maximizing your chances to exploit those weaknesses, and so on. You are just trying to increase the probability of your opponent making a mistake.
At that point, some expert players enjoy that aspect of competitive play, and others just generally lose interest.
Posts:
2,220
Registered:
Mar 4, 2007
From:
San Francisco
Age:
45 Home page
Re: Can Nosovs be beaten?
Posted:
Jun 24, 2010, 9:45 PM
nice post up2.
up2ng; >a human player might only look at what he considers to >be the three or four "best" moves and then tries to >determine a couple of "best" responses to those moves, >and so on. Looking very narrow, but deep compared to how >a computer might approach the problem.
yes, we can eliminate the width, allowing more energy of thought to look further in depth down the move tree.
also, a human is able to reverse engenier a position. i can throw a stone to the side in my mind as P2, and imagine what will white do if allowed, i look at the out come and in that after math position i ask my self perhaps, "if i had a extra stone here (wich i do) where would it be best placed". this is only one example of reverse engeneering, and i dont think computers can do this.
here is a example of this which nosovs pulled on me. he basically used the reverse thinking principle, and i dont think computers can do this yet.
and heres a little something about the game's phases.
when talking about the opening, a opening is all moves made before the first forcing 3 is made.
so the amount of moves in a opening can vary.
then there is the mid game, and then the end game.
but some games don't have a mid game..
why no mid game?
well, first, the end game should be defined. the end game is a series of forcing moves ending only by victory.
so the end game can be called a VCT. victory by continuos threats.
so.. back to the mid game. if the first forcing 3 played (thus removing you from the opening) and that player continues to place forcing moves all the way to victory that never give up initiative nor drifts into a static position even for a single move, then the mid game never existed in that game, not even for a moment.
games that do not have mid games where white wins, means that P2 played a bad opening? usually, but not always. perhaps the VCT was not intuitive. perhaps P2 had a ton of brilliant traps set if P1 went a more obvious looking way.
here is a example of where there was no mid game. but does this make blacks opening weak? was whites moves intuitive? white's 3rd 4th 5th 6th,.. white could had gone different ways, and fallen into traps?
Message was edited by: zoeyk at Jun 24, 2010 5:11 PM
Scire hostis animum - Intelligere ludum - Nosce te ipsum - Prima moventur conciliat - Nolite errare
Posts:
6
Registered:
Mar 10, 2009
From:
NC
Age:
30
Re: Can Nosovs be beaten?
Posted:
Jun 25, 2010, 9:13 AM
I just finished reading this epic forum discussion. Not only is my mind reeling with poorly thought out responses to most of the premises/conclusions stated, but my eyes hurt and I can barely see straight! With this in mind, and with full disclosure of my amateur status in the Pente world, I would like to offer a summation and rephrasing of some important ideas and conclusions.
1. Nosovos can obviously be beaten in both skill and score. However, it would take an exceedingly rare uber-badass of the umpteenth degree to pull off the former, and at least an exceptionally dedicated master to accomplish the later. The golden child of Pente surely exists or will exist, and I'm sure Nosovs will enjoy the long-term challenge, if and when it occurs. This assumes of course that Nosovs is not himself the golden child! Perhaps a generational battle of the Golden Children is thus in order.
2. The database is a tool. It can only hurt or help the game to the degree that people use or misuse it. As a teaching tool it can be one of the most valuable commodities in the Pente world. As a crutch it can be a terminal instrument of creativity?s demise.
3. This game, like any game, has a huge psychological component. To pretend it?s simply a math problem conceals some of the finer points of real-time human Pente. Nosov?s seems to understand this more than most, and it?s what sets him apart from the average ?master?.
4. As far as the epistemology of P1?s Perfect Play is concerned, I think it's best and safest to be skeptical. The word perfect is a heavy word, and should be lifted with cautious humility at worst, or never outside of a theoretical sense at best.
What we can know (and what the practical definition of Perfect Play could/should be) are exceptionally strong lines that have historically held up to 99%-100% of past P2 challenge variations. What makes the game non-futile is the art in finding the novel ways to throw off the strongest or most perfect of known P1 winning lines. I agree with the concept of P1 Perfect Play, and trust Zoey, Karl, and others when they state that patterns have been discovered that are thus far undefeated. However, I must maintain my position, that Perfect Play is unKnowable in the big ?P? sense of the world Perfect, and the big ?K? sense of the word Know. This is a subtle but important point for the future growth of the game.
In the science of Pente, each game is a new experiment. To say that a line has never been beaten is far different than saying it will never or can never be beaten. It only takes one brilliantly placed P2 stone to set up the potential to turn on end a well accepted and long established ?perfect play line?. Then it is simply a matter of finding the newest ?perfect? P1 response, and so on ad infinitum. I think this is where the subtle distinction between frustrating and futile can be found. Again, I think there is such a thing as Perfect Play. I just don?t think it?s a good idea to pretend you know you are walking with perfect steps, and prematurely dismiss a witty P2?s ability to perceive a weakness P1, or any past P1 has yet to notice.
While it may never be the case that All variations can be worked out such that there is a Known dance routine in need of mere clean execution to annihilate any hope of P2?s chances for success, it?s attainment is sufficiently outside the limits of real-time human play to make any hope of ?solving? the game a non-issue. Is the game extra difficult for master vs. master as P2? Sure. Does that make it less fun? I don?t think so. Master level players should delight in master level challenges.
Furthermore, at some future time, I?m sure a computer of sufficiently monstrous computational stature can work out every possible Pente game and then any average player can view the ?Manual of P1?s Perfect Play in All Cases?. However, just making moves out of a handbook is not really ?playing? Pente.
Lastly, I want to comment on an idea that was brought up along with this issue. Master players never choose to lose. They may experiment with risky maneuvers they project to end in victory, and hope these tactics are sufficiently strong and near enough to Perfect Play to assure victory, but that is far different than ?choosing? to lose by deviating from a history of undefeated P1 lines. As each new master P2 attempts to uniquely unseat (perhaps in vain) a hitherto undefeated P1 line, they will force the master P1 to deviate in some way from well trod patterns. This should not be equated to ?choosing? to lose in my opinion.
Thanks to the community for such an in-depth and educational discussion! I seriously doubt all this typing has improved my ability to properly place stones, but all the reading has surely increased my appreciation for both Pente and the Pente.Org players.
PS. Prof. Tate and Prof Zoey, please excuse the 130 words I omitted which would have made this a perfect 1000 word essay.
Posts:
542
Registered:
May 9, 2002
From:
Northeast USA
Re: Can Nosovs be beaten?
Posted:
Jun 25, 2010, 6:50 PM
peter80, that was a nice summary.
The only part I would disagree with slightly is this:
"However, I must maintain my position, that Perfect Play is unKnowable in the big ?P? sense of the world Perfect, and the big ?K? sense of the word Know."
...
"To say that a line has never been beaten is far different than saying it will never or can never be beaten. It only takes one brilliantly placed P2 stone to set up the potential to turn on end a well accepted and long established ?perfect play line?. Then it is simply a matter of finding the newest ?perfect? P1 response, and so on ad infinitum. I think this is where the subtle distinction between frustrating and futile can be found. Again, I think there is such a thing as Perfect Play. I just don?t think it?s a good idea to pretend you know you are walking with perfect steps, and prematurely dismiss a witty P2?s ability to perceive a weakness P1, or any past P1 has yet to notice."
I think there is some confusion about the term "Line", at least for the purposes of this discussion about perfect lines. It would make sense that a line would be a single path of play, such as looking up a single game in the database and just scrolling through those moves to the end of the game. When we speak of perfect lines, we are talking about something that looks more like a tree -- moves that are winning moves no matter which response P2 will choose next. Depending on the P2 response, the line will go off in a different direction, but it is still guaranteed victory for white. Not only that -- but there can actually be MORE than one perfect move for P1 in a given situation. Meaning, there is more than one option for white available that guarentees victory at that moment in the game. As one far-fetched example, there are times when T1, far away from the action, is a perfect move for white -- although subjectively not the "best" move, it still maintains perfect play if P1 is guaranteed victory based on all possible P2 responses.
In addition, perfect play is not always unknowable. If P2 does a poor job of increasing the complexity of the game, then perfect moves for P1 are easy to identify. There are many cases where you can look a game up in the database, analyze it, and conclude that P1 made no mistakes -- meaning that P2 could not win. Now, if P2 had played a different response somewhere along the way that increased the complexity, the previous move for P1 would still be a winning move, but the NEXT move for white might be more difficult and perhaps white might have made a mistake deeper into the game if things played out differently. But, the moves that were actually played were indeed winning moves.
In this sense, the game is in fact futile, and not just frustrating.
I do agree that there are many moves that experts might agree are winning moves, when in fact they are not. An undiscovered novelty for P2 will prove such moves to be losing moves in the future.
Re: Can Nosovs be beaten?
Posted:
Jun 29, 2010, 4:51 PM
As pie is away at the moment, I thought I'd mention a couple of things on his behalf.
To up2ng. Piecraft is a computer programmer and knows a lot about chess programming having written a couple of these and in fact took the one I wrote in 3rd year and improved upon it quit a bit. I think he took my advice and tried to prompt responses from the experts without elaborating on his background to ensure that clearly written responses that the lay-person could understand would be provided. I think this was achieved. I know he understands very well about computer V human thinking (as do I) and the limitations of computers with search horizons and evaluation functions and so on. Having said that, your writing was as usual crisp, concise and easy reading. Thanks Dean.
Peter80. I think you misquote Piecraft in saying that he says experts choose to lose. He did not say that. He said they choose to risk losing (if they had a choice between that and playing perfect lines). This is a big difference.
For me this recent line of discussion has been the most illuminating part of the whole thread. Its not that new concepts are being tabled, but that a greater insight into the minds of the experts is being attained. It seems that the experts do not all have quite the same definition of perfect lines / perfect play as each other, although there seems to be general agreement overall.
Lastly, zoey. I know that piecraft understands the fact that some games don't have a middle game. In fact I spent about 3 hours over the board with him trying to find 'middle-gameless' lines and to determine under what conditions they occurred. This was a very interesting study.
Re: Can Nosovs be beaten?
Posted:
Jul 7, 2010, 9:51 AM
ok. Back from Holidays.
Thanks Ali for your post.
I don't really agree though that the experts have the same view/understanding of what each of them mean by perfect lines. There is a tendancy to use this term as if they all mean the same thing, but I think that behind this are very different understandings of what this really implies regarding solved game trees for all paths regardless of search horizons. I don't necessarily think that these concepts are not understood as such, but that when they say 'perfect line' they may mean to different extents.
and yeah up2ng, you write well dude.
And zoeyk, thanks for the comment about no-mid-game games. Alison and I came up with some good examples. I will have to share these with you at some point.
