[Computer-go] ManyFaces swindled of victory?

terry mcintyre terrymcintyre at yahoo.com
Sat Nov 13 08:33:18 PST 2010


Don, as you say, humans are very good at discerning patterns - and the game of 
Go is all about patterns. 

Now, in some cases, the pattern-matching ability can lead humans astray, but in 
other cases, it's a done deal. 

For example, we hashed over the concept of nakade a while back. These are 
patterns which strong humans recognize at a glance. Groups with certain shapes 
are mathematically, provably, totally, dead beyond hope, assuming proper play. 
At that time, many programs were weak in that area. Now, strong programs usually 
do not fall into such simple traps.

In addition to "dead beyond hope" and "certainly alive", strong humans also 
recognize "can live with ko" and "seki" shapes - again, beyond a shadow of a 
doubt, as mathematically certain as the sunrise in the morning. 

The analogy with the stock market misses the mark because the stock market has 
many millions of independent actors ( the human beings who make buy and sell 
decisions ) who act upon many billions of facts, most of which are inaccessible 
to the punditocracy who try to make sense of the markets. (would-be pundits are 
also often handicapped by inferior models, but that's another tale.)

Go is a game of complete information played by exactly two players, if we ignore 
rengo or phantom variants; when the position is simplified enough ( something 
which strong players actively seek ), the result is mathematically provable - 
and well within the limits of what humans can do with their pattern recognition 
facilities. 

The topic here is that of large semeai; many games have shown that programs are 
vulnerable to misjudging the outcome; Darren Cook has written up a page 
describing the problem with random playouts and complex semeai which depend upon 
precise move ordering. In many cases, semeai are won or lost by a single play, 
and one must play A, B, C, D precisely in order, in response to any one of Z, Y, 
W, X, etc. A large class of problems depend on "if my group A loses a liberty, I 
must take a liberty from group B in such-and-such order." Strong humans read out 
these problems and play them correctly. Strong programs have been observed to 
fail.

Many well-known joseki create a local situation with such "win by 1" capturing 
races. Usually, the person who loses the local race gains something - 
"influence" or "thickness" in compensation; strong professional players consider 
these sequences to lead to equal results for both players. If a player takes 
that compensation, converts it to cash ( territory ), and also manages to 
swindle a program out of winning the local semeai, the player can easily win; 
it's like being able to write off a $500,000 mortgage and keep the house, 
because the lender made a mistake in the paperwork.

It used to be fairly easy to set up a ladder and a ladder-breaker, and programs 
would still play out the ladder as if the ladder breaker were not present - a 
huge misconception. Strong programs don't seem to fall into that trap anymore - 
but they do fall for semeai which are conceptually similar, in a mathematically 
provable sense.

While I present a single game to illustrate my case, I generalize from many 
games. It's still merely a black box analysis, however; I leave it to the 
programmers to "open the box" and discover how the internals map to the 
observable externals. 

There is a difference between Go and Chess. In Chess, only one thing matters: 
put the other guy in checkmate, and you win, even if all you have on the board 
is one king and one pawn, and the other player has a dozen pieces. In Go, once 
you achieve a territorial advantage, you need only keep what is yours. Going 
back a year or two, programs were not very good at keeping what was theirs; they 
played odd yose moves which yielded up territory without gaining anything in 
terms of improving their winrate. Semeai are the midgame equivalent - moves 
which are mathematically constrained in ways which significantly alter the real 
status of the game, as opposed to the hypothetical winrate of any algorithm 
which does not understand those constraints. 
Terry McIntyre <terrymcintyre at yahoo.com>


Unix/Linux Systems Administration
Taking time to do it right saves having to do it twice.




________________________________
From: Don Dailey <dailey.don at gmail.com>
To: computer-go at dvandva.org
Sent: Sat, November 13, 2010 10:14:03 AM
Subject: Re: [Computer-go] ManyFaces swindled of victory?




On Sat, Nov 13, 2010 at 9:08 AM, terry mcintyre <terrymcintyre at yahoo.com> wrote:

Don, I'm enough of a go player to know that, in such a position, a 2d human 
would trounce the kyu player, if given the middle game position to start with - 
but I'd appreciate the views of a dan-level player.  MFGO in such positions 
often fails to take care of its groups - even though it is far stronger than I.

I'm not saying you are wrong,  I'm just trying to keep everyone honest.   It's 
just that the hair on the back of my neck sticks up when I see an idea start to 
become gospel and I hope that as we figure all of this out we are not making 
incorrect assumptions that will set us back.     

Here is what bothers me about this:

You saw a position and said, "in such a position" a 2d human would trounce the 
kyu player.  Not only are you making a judgement about a specific position 
(which may or may not be true),  but you are generalizing it based on what you 
subjectively think "such a position" means.     You have used at least 2 
subjective opinions, linked them together and use this to draw a conclusion.   
(Presumably this is just a tentative conclusion,  a hypothesis, which I think is 
the right way to think about such things.)    

Don't forget that as humans we have very strong pattern recognition facilities. 
  We can see animals and all kinds of various objects in cloud formations and 
studies have shown that we tend to find patterns in random things.   This often 
tends to make us draw conclusions very quickly based on little evidence or just 
plain gut feelings as witnessed by the large number of criminal cases that were 
incorrectly judged.      This is both a strength and a weakness that we have but 
it is the source of massive amounts of incorrect "common wisdom."

How many times have programmers here observed something wrong with the way their 
program plays,  made a judgement about WHY the program is playing badly, and 
then implemented a fix which clearly addresses the problem they think they are 
seeing - and lo and behold the program is now weaker?     It was worth a try but 
it's way too easy for such a chain of reasoning to be wrong.   

If you really want to see my point in action,  watch the stock market and the 
way each days activity is explained at the end of the day.    They will say 
something like,  "The nasdaq today was down based on fears of recession."     
The stock market is incredibly complex,  and yet the analysts ALWAYS seem to 
know the exact explanation for why it went up or down some slight amount.     
And that explanation is presented as a statement of fact,  the reason why.  

Sorry I'm being so picky about this but it drives me crazy, it's one of my 
peeves.   You didn't do anything wrong :-)

Don

  







 

>
>Dave Fotland himself has observed similar results; there are many game records 
>to back up this contention. A while back, I posted about a player who has a very 
>high winrate against MFGO - it looked like his account did nothing else but play 
>computer programs. 
>
>
>Strong humans organize and prioritize things differently than (current) strong 
>programs.
>
>
>If you can keep track of things at  blitz speed, and set up complicated fights 
>with large unsettled groups, and patiently encroach on a program's eyespace ( 
>even DDK players know that a dragon, however large, still requires two eyes to 
>live ), you can often give a program a good drubbing.
> 
>Earlier, we had a post regarding a shogi program which combines the votes of 
>four other programs. 
>
>
>Given that processor cores are now abundant, and that UCT programs have trouble 
>using all that horsepower, I wonder if some cores should look at the board in an 
>entirely different way and, every now and then, generate "Danger ,Will Robinson" 
>signals which trump the usual winrate algorithms - or at least tweak them toward 
>safety moves. 
>
>
>Is there a way to combine such disparate information into a coherent picture of 
>the board? 
>
>Terry McIntyre <terrymcintyre at yahoo.com>
>
>
>Unix/Linux Systems  Administration
>Taking time to do it right saves having to do it twice.
>
>
>
>
>
________________________________
From: Don Dailey <dailey.don at gmail.com>
>To: computer-go at dvandva.org
>Sent: Sat, November 13, 2010 8:18:19 AM
>Subject: Re: [Computer-go] ManyFaces swindled of victory?
>
>
>
>
>
>On Fri, Nov 12, 2010 at 9:24 PM, terry mcintyre <terrymcintyre at yahoo.com> 
wrote:
>
>This is a curious game - a 3 stone game betwixt ManyFaces and a 2 kyu player. 
>>
>>
>>It looks to my eye as if MFGO was well ahead in the middle game, yet managed to 
>>snatch defeat from the jaws of victory. 
>>
>>
>>Black has many unsettled groups; usually a stronger player causes much grief in 
>>situations like this, but black managed to turn around and swallow up not one 
>>but several white groups. 
>>
>>
>>Most interesting. I think this confirms an earlier thread - when there are 
>>several semeai on the board, MCT-based strategies get into difficulties. 

Are you sure of this?    I think it is human nature to try to make sense of 
things and create general rules and principles to explain what we think we may 
be observing.   It's also human nature to often be wrong about it.     The first 
time an idea is skillfully presented it tends to carry extra (sometimes 
undeserved) weight because everyone else interprets what they see in terms of 
that and it gets repeated over and over till it becomes the de-facto sound bite 
that propagates itself.  

Maybe it's completely true and there is no dispute or it's obvious - 
unfortunately I'm not a go player so I'm in no position to add my 2 cents but I 
would be very cautious about making  general statements about the relative 
strengths and weaknesses of humans compared to computers.      

I know from experience with chess for many decades that peoples perception of 
the computers strengths and weaknesses were constantly misinterpreted,  very 
often based on a just a game or two someone played.     People were just trying 
to make sense of it, but they got it wrong a lot.    In one case in the 80's my 
program played a master and lost tactically and believe it or not the master 
felt the program was strong positionally but weak tactically (because he beat it 
with a tactical shot) even though we now understand it's completely the 
opposite.    The truth of the matter was that the computer was simply a lot 
weaker than this particular player so it was inferior all the way around.  

If your supposition is true, then it's a good thing to know because perhaps it 
can be addressed.    Or it might end up being one of those things that 
diminishes in importance over time as computers get generally stronger and 
stronger.    In chess, computers have always been weaker in true positional 
understanding (even to this day) but we have added so many layers of strength 
that this weakness is only a relative thing.   It's weaker than other parts of 
it's game,  but not so weak that anyone can easily take advantage of it.   Most 
computer games resemble Grandmaster games in the quality of positional play.



 

>
>I think strong human players tend to have a better grasp of "this particular 
>fight is settled, no need to study it further until a  liberty is removed (in 
>which case respond now!); let's focus on this other fight instead . . ."
> 
>
>
>
>_______________________________________________
>Computer-go mailing list
>Computer-go at dvandva.org
>http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
>
>
>
>_______________________________________________
>Computer-go mailing list
>Computer-go at dvandva.org
>http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
>



      
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://computer-go.org/pipermail/computer-go/attachments/20101113/744fc16a/attachment.html>


More information about the Computer-go mailing list