Are there "ANTI-computer" positions at BG?
Bill Taylor
In chess, there are several well-known "anti-computer" positions;
cunningly constructed to take advantage of known (and currently
inoperable) blind spots that the sandbrains have.
Several times there have been remarks made here, about situation-types that
bots do badly on; e.g. trying to get a 2nd checker back behind the prime.
But does anyone have a definite "test position"?
That is, a position from which a bot will (with high probability) lose;
but from which a human will (ditto) win?
That would be waaay koool. Can anyone help?
-------------------------------------------------------------------------------
Bill Taylor W.Ta...@math.canterbury.ac.nz
-------------------------------------------------------------------------------
God does not play dice with the universe, he plays Go.
-------------------------------------------------------------------------------
_______
/\ o o\
/o \ o o\_______
< >------> o /|
\ o/ o /_____/o|
\/______/ |oo|
| o |o/
|_______|/
-------------------------------------------------------------------------------
John Goodwin
On 26 Jun 1998 07:35:14 GMT, mat...@math.canterbury.ac.nz (Bill
Taylor) wrote:
>In chess, there are several well-known "anti-computer" positions;
>cunningly constructed to take advantage of known (and currently
>inoperable) blind spots that the sandbrains have.
>
>
>Several times there have been remarks made here, about situation-types that
>bots do badly on; e.g. trying to get a 2nd checker back behind the prime.
>
>But does anyone have a definite "test position"?
>
>That is, a position from which a bot will (with high probability) lose;
>but from which a human will (ditto) win?
>
>That would be waaay koool. Can anyone help?
>
There are various known weaknesses in the robot players, but coming up
with a specific position that will cause it to lose is difficult (more
likely impossible), because of the random nature of the game*.
I suspect that the best you could do would be to start play from a
position in which the particular robot was weak.
If you did that, you should win a significantly greater percentage of
games than normal (assuming you do not share the same weakness as the
robot).
JG
* There is no position before disengagement that will guarantee that
one player will win.
tapewo...@my-dejanews.com
> In chess, there are several well-known "anti-computer" positions;
> cunningly constructed to take advantage of known (and currently
> inoperable) blind spots that the sandbrains have.
> Several times there have been remarks made here, about situation-types that
> bots do badly on; e.g. trying to get a 2nd checker back behind the prime.
> But does anyone have a definite "test position"?
> That is, a position from which a bot will (with high probability) lose;
> but from which a human will (ditto) win?
> That would be waaay koool. Can anyone help?
I've always been told that the computers tend not to play well against
well-timed massive backgames. I unfortunately have only anecdotal evidence of
this, but it's also the opinion of many who are much more qualified than I.
You might try building two- and three-point backgames against the computer.
The reasoning most people give that the computers don't play well against
backgames is that the neural networks learn early on not to get into backgame
positions, and thus don't get any experience defending themselves against
them.
CR
-----== Posted via Deja News, The Leader in Internet Discussion ==-----
http://www.dejanews.com/ Now offering spam-free web-based newsreading
Gary Wong
> >cunningly constructed to take advantage of known (and currently
> >inoperable) blind spots that the sandbrains have.
> >
> >Several times there have been remarks made here, about situation-types that
> >bots do badly on; e.g. trying to get a 2nd checker back behind the prime.
> >
> >But does anyone have a definite "test position"?
> >
> >That is, a position from which a bot will (with high probability) lose;
> >but from which a human will (ditto) win?
> >
> >That would be waaay koool. Can anyone help?
I would claim that there are essentially no positions that the best bots play
terribly badly -- there are certainly positions where the best humans outplay
the best bots by a small amount (typically relatively rare and highly
tactical positions, like recirculating a chequer to pick up a second blot
like you mention) but I wouldn't go as far as to say the bot would lose with
a high probability where the human would win.
As Chuck Bower mentions in another article, cube errors by the bots
tend to be much worse than chequer play errors. One contrived
position which I believe Jellyfish misplays springs to mind (bear in
mind that it is tremendously unlikely to arise in play and so I don't
know if you'd consider it a "test position") -- this is position #127
in Robertie's _Advanced Backgammon_ which he uses to illustrate the
Kauder Paradox:
+13-14-15-16-17-18-------19-20-21-22-23-24-+
| | | |
| | | |
| | | |
| | | |
| | | |
v| |BAR| | O on roll, 1 cube, money game.
| | | 6 6 | Jacoby rule applies.
| | | O O |
| X X X | | X X O O |
| X X X | | X X O O O |
| O X X X | | X X O O O |
+12-11-10--9--8--7--------6--5--4--3--2--1-+
It should be fairly clear that O wins a certain gammon if he rolls a 6 to
clear X's 5-prime, but is cubed out otherwise. However, Jellyfish does not
seem to evaluate X's prime so far from its own board sufficiently valuable
(I'm not sure whether it can walk it home from there) and drops the cube
here (in reality X has a beaver). This is probably about as bad as you
can get Jellyfish to play (X wins 0.083 cubeless equity by playing on as
a human would, or loses 1.000 by dropping like Jellyfish).
Note that I don't have a Windows machine myself and can't test this on
a recent version of Jellyfish -- apologies for bashing it if it has
learnt to play the position in the meantime :-)
J...@opticon.demon.co.uk (John Goodwin) replies:
> * There is no position before disengagement that will guarantee that
> one player will win.
Oh, really? I'm willing to guarantee that O will win (a backgammon) in this
pre-disengagement position -- want to prop me on it? ;-)
+13-14-15-16-17-18-------19-20-21-22-23-24-+
| X | | X X X X X X |
| | | X X X X X X |
| | | |
| | | |
| | | |
v| |BAR| | O on roll, 1 cube, money game.
| | | |
| | | |
| | | |
| | | |
| | | O X X |
+12-11-10--9--8--7--------6--5--4--3--2--1-+
Cheers,
Gary.
--
Gary Wong, Department of Computer Science, University of Arizona
ga...@cs.arizona.edu http://www.cs.arizona.edu/~gary/
Chuck Bower
In article <6n0v1t$t3r$1...@nnrp1.dejanews.com>,
<tapewo...@my-dejanews.com> wrote:
>
>I've always been told that the computers tend not to play well against
>well-timed massive backgames. I unfortunately have only anecdotal evidence of
>this, but it's also the opinion of many who are much more qualified than I.
>You might try building two- and three-point backgames against the computer.
>The reasoning most people give that the computers don't play well against
>backgames is that the neural networks learn early on not to get into backgame
>positions, and thus don't get any experience defending themselves against
>them.
There is an inherent problem here, even if it is true that robots
don't defend BG's well. First you must get into one! If you end up with
a single point "backgame" half the time while trying to steer into a true
(two point) backgame, you will probably end up a BIG loser against the best
bots (JF, Snowie, M-Loner,...).
In the past, even against the weak robots, one necessary
conditions was to steer the robot into a position where it thought it was
ahead (in equity) but you actually were. Then you (and the bot) cube at
every opportunity and you end up coming out ahead in the long run. You
win a majority of games when the cube is high (32, 64,...) and lose a
majority of all games, but then only when the cube is low.
If there were such a path to trick JF (for example) into a losing
situation (averaged over all games, obviously), I haven't heard about it.
David Montgomery is one of the most knowledgable players on both playing
backgames (and propositions) and also on the ability of the current robots.
Recently I asked him a similar question concerning JF and he responded in
the negative. But he's only human. ;)
Chuck
bo...@bigbang.astro.indiana.edu
c_ray on FIBS
News1
Gary Wong wrote in message ...
<SNIP>
> +13-14-15-16-17-18-------19-20-21-22-23-24-+
> | | | |
> | | | |
> | | | |
> | | | |
> | | | | O on roll,
> v| |BAR| | 1 cube, money game.
> | | | 6 6 | Jacoby rule
> | | | O O |
> | X X X | | X X O O |
> | X X X | | X X O O O |
> | O X X X | | X X O O O |
> +12-11-10--9--8--7--------6--5--4--3--2--1-+
>
>It should be fairly clear that O wins a certain gammon if he rolls a 6 to
>clear X's 5-prime, but is cubed out otherwise. However, Jellyfish does not
>seem to evaluate X's prime so far from its own board sufficiently valuable
>(I'm not sure whether it can walk it home from there) and drops the cube
>here (in reality X has a beaver). This is probably about as bad as you
>can get Jellyfish to play (X wins 0.083 cubeless equity by playing on as
>a human would, or loses 1.000 by dropping like Jellyfish).
>
>Note that I don't have a Windows machine myself and can't test this on
>a recent version of Jellyfish -- apologies for bashing it if it has
>learnt to play the position in the meantime :-)
>
Here is what snowie says...
3-Ply
Equity: -0.313
0.0% 0.7% 50.7% 49.3% 31.8% 1.5%
Double decision
Good enough: 0%
Too good to double: 0%
Take decision Current Doubled Borderline
Equities: -0.313 -0.313 0.570
Proper cube action: No double, beaver
David Montgomery
In article <wtbtrfl...@brigantine.CS.Arizona.EDU> Gary Wong <ga...@cs.arizona.edu> writes:
>> On 26 Jun 1998 07:35:14 GMT, mat...@math.canterbury.ac.nz (Bill
>> Taylor) wrote:
>> >That is, a position from which a bot will (with high probability) lose;
>> >but from which a human will (ditto) win?
>
>terribly badly -- there are certainly positions where the best humans outplay
>the best bots by a small amount (typically relatively rare and highly
>tactical positions, like recirculating a chequer to pick up a second blot
>like you mention) but I wouldn't go as far as to say the bot would lose with
>a high probability where the human would win.
I would say that there are anti-computer positions. However, defining
them as positions that humans would win with high probability and that
the bot would lose with high probability seems a bit much. Also, it
doesn't take into account gammons and backgammons. I think it would
be better to define them as positions where bots give up some amount
of equity relative to good human play.
As Gary pointed out, the bots can be far off on their cube actions in
some positions, so you can get big errors that way. But it is also
true that even just looking at cubeless checker play, there are many
positions where the bots play terribly.
Gary showed this position:
> +13-14-15-16-17-18-------19-20-21-22-23-24-+
> | | | |
> | | | |
> | | | |
> | | | |
> | | | |
> v| |BAR| O O | X on roll, cubeless equity?
> | | | O O |
> | | | O O |
> | X X X | | X X O O |
> | X X X | | X X O O O |
> | O X X X | | X X O O O |
> +12-11-10--9--8--7--------6--5--4--3--2--1-+
O's home board
I've put X on roll and made the position cubeless. With good
human play, X is probably better than +.6 cubeless; with SW
at the helm, X is *negative* about the same amount. SW still wins
the position nearly half the time, so I wouldn't say that it meets Bill
Taylor's definition of an anti-computer positions, but giving
up ~1.25ppg cubeless, it meets mine.
BTW, this position isn't that contrived. If you play for a
backgame from the start against Snowie, or JF versions 1 and
2, it isn't that uncommon to end up with positions similar to
this. With the cube in play, you can jack it up to 512 and
come out ahead even though you lose gammons on most games.
> +13-14-15-16-17-18-------19-20-21-22-23-24-+
> | O O | | O X X |
> | | | |
> | | | |
> | | | |
> | | | |
> v| |BAR| | X on roll, cubeless equity?
> | | | |
> | | | |
> | | | |
> | | | O O O O O |
> | O O | | O O O O O X |
> +12-11-10--9--8--7--------6--5--4--3--2--1-+
O's home board
This is a position from the 4/97 Chicago Point.
With good human play, X is worth about .25 cubeless.
Rolling the position out with JF, you'll get between
.4 and .5, depending on the version and level you use.
(It looks like version 2 plays it better.) Whether
a .2 cubeless error qualifies as an anti-computer
position I'll leave up to the reader.
A minor point: the bots get these kinds of positions wrong because
they are strategic, not tactical. The bots don't have the right long
term plan. In tactical positions, the bots are extraordinarily strong.
David Montgomery
mo...@cs.umd.edu
monty on FIBS
Kit Woolsey
Chuck Bower (bo...@bigbang.astro.indiana.edu) wrote:
: In the past, even against the weak robots, one necessary
: conditions was to steer the robot into a position where it thought it was
: ahead (in equity) but you actually were. Then you (and the bot) cube at
: every opportunity and you end up coming out ahead in the long run. You
: win a majority of games when the cube is high (32, 64,...) and lose a
: majority of all games, but then only when the cube is low.
: If there were such a path to trick JF (for example) into a losing
: situation (averaged over all games, obviously), I haven't heard about it.
: David Montgomery is one of the most knowledgable players on both playing
: backgames (and propositions) and also on the ability of the current robots.
: Recently I asked him a similar question concerning JF and he responded in
: the negative. But he's only human. ;)
In the earlier versions of Jellyfish this could be done. The idea was to
play a deep backgame, get all of your men sent back (JF was generally
cooperative here), hit one checker, and build a prime in the outer boards
(nothing in your inner board). Jellyfish thought that it was winning
this position significantly, and would double. You could redouble, and
get the cube as high as you wanted. Naturally a lot of the time this
wouldn't work out and you would lose 4 or 6 points, but when it did work
you would win a ton. For this reason, it was dangerous to back Jellyfish
in a money game against a player who knew this trick.
I'm pretty sure the latest version of Jellyfish evaluates this type of
position better and doesn't fall for this particular trap. There may be
other such types of positions. When Malcolm Davis backed Jellyfish in
the money match vs. Mike Senkiewicz and Nack Ballard last summer he took
the precaution of setting the "caution" switch on -- this gave him some
protection against JF going crazy on big cubes. It probably wasn't
necessary, and it the actual games nothing like that every happened.
Kit
Jacques Torrione
Well, take this position : each player has one checker on his 6 point. The
equity of the player on roll is 0.625 (he wins 81.3%). All bots evaluate
this position as a double/pass. Bots doesn't see that the opponent has a
very efficient recube and that he must take, although he has only 18.8%
winning chances. So human will loose less than bots here, by taking.
You can find others position like that : each player has 2 checkers on his 2
point for example. Correct cube action is double/take and bots pass.
So jellyfish ans snowie wrongly pass these positions. But if you make a
cubeful rollout with snowie, it see that it's a take! So bots are weak here
to evaluate the position, but with a rollout it play correctly.
Jacques Torrione
Rew Francis
>Well, take this position : each player has one checker on his 6 point. The
>equity of the player on roll is 0.625 (he wins 81.3%). All bots evaluate
>this position as a double/pass. Bots doesn't see that the opponent has a
>very efficient recube and that he must take, although he has only 18.8%
>winning chances.
A small point really but I don't think this is exactly right - it is
OK to drop this position.
Suppose the cube is on 1, player 1 doubles and player 2 takes. 75% of
the time player 1 wins on the next roll (all numbers except 1-1, 2-1,
3-1, 4-1 and 3-2). Of the remaining 25% player 2 will always redouble
to 4. Player 1 can either drop losing 2 points or take. If they take
then 75% of the time they will lose 4 points on the next roll, 25% of
the time they win 4 points - average loss 2 points - so it is equally
correct for player 1 to take or drop the redouble. So player 2's
equity in the games where they accept the initial double and player 1
misses is 2. So player 2 gets equity of 0.75*(-2) + 0.25*(2) = -1 by
taking the initial cube. So it is equally correct for player 2 to take
or pass the initial double. (Of course this applies only to money play
- varying match scores will change things a lot).
Rew
Jacques Torrione
Hello,
Oops, you're right. I never calculate this because .. I always thought
(why???) it was a take and you loose something by passing. In fact not..
it's the same to take or pass .
Thanks
Jacques Torrione
Jacques Torrione
>>You can find others position like that : each player has 2 checkers on his
2
>>point for example. Correct cube action is double/take and bots pass.
>
>don't get the opponent off. Of those, 26/36 do get you off = 20%)
>isn't that a drop for money (takepoint=25%)? If the opponent doesn't
>get off, it's then certainly a double/take for you.
>
>The previous articles have disappeared from my machine so I can't see
>if we're talking about a specific match score or not.
Hi,
no this one is a take and you loose something by passing.
Let's calculate this (for money game).
When you take you loose immediately 26/36 *2 points.
Then 10/36 you redouble to 4 (and of course opponent must take because it's
a last roll and he wins 10/36 = 27% ). So you redouble to 4 and wins 10/36 *
26/36 * 4 points. When you miss you loose 4 points so 10/36 * 10/36 * 4
points.
In short you loose 26/36*2 + 10/36*10/36*4 = 1.75 points and you win
10/36*26/36*4 = 0.80 points. So by taking you loose 1.75-0.8 = 0.95 points.
This is less than the point you loose by passing so it's a take.
Something more. You win only 10/36*26/36 = 20.06% and this is less than the
25% you speak about. But in money game 25% is the dead take point (in a last
roll when you can't recube). This is not a last roll so your take point is
less than 25%. In fact your recube is very efficient, because you give a
dead cube which the opponent must take with 27.7 % winning chances which is
only 2.7% higher than the dead take point. That's why it's a take with only
20.06 & winning chances.
As I said, bots wrongly pass this position, but snowie takes after a cubeful
rollout.
Jacques Torrione
Robert-Jan Veldhuizen
position.
>> +13-14-15-16-17-18-------19-20-21-22-23-24-+
>> | | | |
>> | | | |
>> | | | |
>> | | | |
>> | | | |
>> v| |BAR| O O | X on roll, cubeless equity?
>> | | | O O |
>> | | | O O |
>> | X X X | | X X O O |
>> | X X X | | X X O O O |
>> | O X X X | | X X O O O |
>> +12-11-10--9--8--7--------6--5--4--3--2--1-+
O's home board
Being a human myself, I wonder what move X should make if he rolls his
horror: 66. Is it 9/15(3) plus some other checker (which one?), or is
this spreading the checkers too much making it too hard to contain O if
he fails to roll a 6?
--
Zorba/Robert-Jan
David Montgomery
I would play 5/11 6/12 7/13 8/14. I'm not sure this is best, but
I'm pretty sure it's better than anything beginning 9/15(3).
The idea is to make points in order.
You really want to avoid putting checkers far out ahead of your
prime. Directly in front of a 6 prime is perfect, 2 in front quite
okay, especially if 1 in front is also slotted -- beyond that it
just takes too long to move the rest of your checkers to catch
up to them.
bshe...@hasbro.com
mo...@cs.umd.edu (David Montgomery) wrote:
> In article <1039.7624...@xs4all.nl> Robert-Jan Veldhuizen
<veld...@xs4all.nl> writes:
> >Skipping through some old rgb articles I found this "anti-computer"
> >position.
> >
> |>> +13-14-15-16-17-18-------19-20-21-22-23-24-+
> |>> | | | |
> |>> | | | |
> |>> | | | |
> |>> | | | |
> |>> | | | |
> |>> v| |BAR| O O | X on roll, cubeless equity?
> |>> | | | O O |
> |>> | | | O O |
> |>> | X X X | | X X O O |
> |>> | X X X | | X X O O O |
> |>> | O X X X | | X X O O O |
> |>> +12-11-10--9--8--7--------6--5--4--3--2--1-+
> > O's home board
> >
> >Being a human myself, I wonder what move X should make if he rolls his
> >horror: 66. Is it 9/15(3) plus some other checker (which one?), or is
> >this spreading the checkers too much making it too hard to contain O if
> >he fails to roll a 6?
>
> I would play 5/11 6/12 7/13 8/14. I'm not sure this is best, but
> I'm pretty sure it's better than anything beginning 9/15(3).
9/15(3) is an error because it allows O to escape with 1-5.
Monty's play is better, and may be best. One alternative is 5/17 6/18, with
the idea of using the builders atop the prime to roll it forward, but I tend
to agree with Monty that it is better to keep the builders closer to the edge
of the prime with 5/11 6/12 7/13 8/14.
Brian
-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
Robert-Jan Veldhuizen
DM> In article <1039.7624...@xs4all.nl> Robert-Jan Veldhuizen
DM> <veld...@xs4all.nl> writes:
>>Skipping through some old rgb articles I found this "anti-computer"
>>position.
>>
DM>|>> +13-14-15-16-17-18-------19-20-21-22-23-24-+
DM>|>> | | | |
DM>|>> | | | |
DM>|>> | | | |
DM>|>> | | | |
DM>|>> | | | |
DM>|>> v| |BAR| O O | X on roll, cubeless
DM>|>> equity?
DM>|>> | | | O O |
DM>|>> | | | O O |
DM>|>> | X X X | | X X O O |
DM>|>> | X X X | | X X O O O |
DM>|>> | O X X X | | X X O O O |
DM>|>> +12-11-10--9--8--7--------6--5--4--3--2--1-+
>> O's home board
>>
>>Being a human myself, I wonder what move X should make if he rolls his
>>horror: 66. Is it 9/15(3) plus some other checker (which one?), or is
>>this spreading the checkers too much making it too hard to contain O if
>>he fails to roll a 6?
DM> I would play 5/11 6/12 7/13 8/14. I'm not sure this is best, but
DM> I'm pretty sure it's better than anything beginning 9/15(3).
Sorry I blundered here...I thought X had to give up at least one point
of his prime anyway. I guess I confused this with rolling forward a
six-prime and rolling 4 4 or something which usually means breaking a
point, but in this case it's of course not necessary at all! Must've
been sleeping or something :-) Thanks for the comments anyway.
--
Zorba/Robert-Jan