Re: How fast can you cheat??
tc...@lsa.umich.edu
muratk <mu...@compuplus.net> wrote:
>So now, I am offering them and any of their ilks to construct
>8 winning matches out of 10, of 31 points against gnudung
>using the seeds 3101 through 3110 (that's what I had used).
This is an interesting challenge. Perhaps someone who knows gnubg better
can explain how exactly the dice are generated from the Mersenne Twister.
Given particular seeds, do we get the same sequence of dice rolls? Or
is it more complicated than that?
--
Tim Chow tchow-at-alum-dot-mit-dot-edu
The range of our projectiles---even ... the artillery---however great, will
never exceed four of those miles of which as many thousand separate us from
the center of the earth. ---Galileo, Dialogues Concerning Two New Sciences
Michael Petch
On 20/08/09 5:51 AM, in article
f0b7cbe9-1f7a-4b15...@a37g2000prf.googlegroups.com, "muratk"
<mu...@compuplus.net> wrote:
> Recently I reported that I won (as a many times over "again"
> event) 8 out of 10 matches of 31 points against gnudung...
>
[snip]
> So now, I am offering them and any of their ilks to construct
> 8 winning matches out of 10, of 31 points against gnudung
> using the seeds 3101 through 3110 (that's what I had used).
>
[snip]
Its not that I don't believe you. I believe its possible by potentially
forcing Gnubg into certain situations where it doesn't properly identify the
true equity of the position (ie a weakness in the neural net - and they
exist!).
Your use of cheating always make me smile, because I doubt you use it in the
same context as everyone else. I often believe you equate your ability to
beat the bot so many times as an indication it cheats (Maybe I am the only
one who has been confused by this for the past few 5 or so years).
So here is what I'll ask, and I haven't seen anyone ask it. Murat, can you
email me a copy of the SGF file of the 10 matches, I will analyze them, post
them to my website, and allow r.g.b to discuss it. The one thing I have
never seen are these actual matches. Often it is hard to decipher what you
are talking about.
Allowing everyone to reviewing the matches can benefit everyone.
Thanks,
Michael
Michael Petch
On 20/08/09 9:26 AM, in article
4a8d6b09$0$504$b45e...@senator-bedfellow.mit.edu, "tc...@lsa.umich.edu"
<tc...@lsa.umich.edu> wrote:
>
> This is an interesting challenge. Perhaps someone who knows gnubg better
> can explain how exactly the dice are generated from the Mersenne Twister.
> Given particular seeds, do we get the same sequence of dice rolls? Or
> is it more complicated than that?
The answer is YES but with with a caveat. If you use the same seed you will
get the same dice results everytime. There is nothing special in the way
this is handled.
The oddity is in how the seed gets set when you use start a new match
through the GUI. If you start a new match, the seed is randomly set before
the match is played. You can see the seed it chose by starting a match, then
going to Settings/Options/Dice and looking at the "seed". If you set the
seed through the GUI (in options), and start a new match and then go back
and look at the seed it will change. I think what the GUI lacks in start
match is the option to use the current seed (instead of the one currently
set). I believe it was done this way so that for a casual player, each match
was different. I'm CC'ing the mailing list with this post/feature request..
Anyway, in current releases if you wish to use the GUI and set your own seed
and get the same dice rolls for an entire match do this:
A) Click "New" icon.
B) Enable "Manual Dice" checkbox
C) set any other player options that you wish.
D) Click "Ok"
E) The dice selection screen will appear. Don't select a roll, just hit
"Close"
E) Go to Options/Settings/Dice tab and
a) Set the random number generator to Mersenne Twister (or which ever
seeded PRNG you like)
b) set the seed to some value (3101)
c) Click "OK"
[At this point the seed is set]
F) Click anywhere on the board and you should get a message along the lines
of "A new match" has been started.
G) Play the match
Now do this process again and from the start with the same seed. The rolls
generated should be the same.
(There is a shorter way to set the seed from the "command panel" using the
"set seed" command for power users).
Michael
Michael Petch
On a side note, I can dramatically improve my odds of defeating the bot over
my usual win percentage by doing something rather trivial.
Start a 31pt match (for example). Play your opening roll to slot your home
board (These will be large errors from the perspective of analysis of
course). On your next roll cube it (Will be a bad cube). Bot will take. Keep
the 24pt anchor on bots side - you will need it, because we are going to
play a backgame (I have success with this on 2ply/2ply).
Keepign your anchor roll, and make more blots available. The bot will
eventually recube early on. When the bot recubes, take and then on your next
roll recube. I often get 16 and 32 cubes off the bat, and I end up with
reasonable chances on the backgames.
Of course what the bot doesn't know is that its neural net doesn't evaluate
the equity of so many blots as being "good". The human knows better (One
that can play a good backgame) and you can take advantage of it. You'd be
surprised at the number of matches you can beat the bot with this way.
Sometimes the bot gets very good/lucky rolls at the beginning, and sometimes
you have to cut your losses and drop a bots recube.
Michael
tc...@lsa.umich.edu
Michael Petch <mpe...@capp-sysware.com> wrote:
>The answer is YES but with with a caveat. If you use the same seed you will
>get the same dice results everytime. There is nothing special in the way
>this is handled.
O.K., then my guess is that it shouldn't be too hard to win a 31-point match
in a single game, by studying the dice sequence carefully and then jacking
the cube up to 32 early on. I might try to do this this weekend.
Michael Petch
On 20/08/09 3:13 PM, in article
4a8dbc6c$0$504$b45e...@senator-bedfellow.mit.edu, "tc...@lsa.umich.edu"
<tc...@lsa.umich.edu> wrote:
> In article <C6B305C4.1EE5D%mpe...@capp-sysware.com>,
> Michael Petch <mpe...@capp-sysware.com> wrote:
>> The answer is YES but with with a caveat. If you use the same seed you will
>> get the same dice results everytime. There is nothing special in the way
>> this is handled.
>
> O.K., then my guess is that it shouldn't be too hard to win a 31-point match
> in a single game, by studying the dice sequence carefully and then jacking
> the cube up to 32 early on. I might try to do this this weekend.
Correct, we do not know what Murat did. He may be above board or he may not
be. My other post at least asked him to present the 10 matches, but you are
right, for all you know is he set the seed got all the rolls saved by
playing a match. Then he replayed the match with full knowledge of the
upcoming rolls.
Or he may have played to a weakness in the bot.
Having Murat play in a method that you have suggested (previous post))
where he has no knowledge of the seed or the rolls and for him to reproduce
his feat would at least prove/disprove that he had foreknowledge of the
rolls.
I'd start with the known fact. He has 10 matches. Lets see them. I am sure
we could probably tell if he was exploiting a neural net weakness. If he was
then we can assume its not a randomness issue. If it doesn't appear he is
exploiting a weakness in the net, then I think proving it with some
strategy with "remote" rolling from someone else who feeds him the rolls one
by one (With the ability for Murat to verify after).
Michael Petch
My original post was to bring up a feature request about overriding the seed
reset in Gnubg when you start a new match.
http://lists.gnu.org/archive/html/bug-gnubg/2009-08/msg00239.html
The message below was the response from a mailing list user, that some users
of rec.games.backgammon might have an interest in:
http://lists.gnu.org/archive/html/bug-gnubg/2009-08/msg00240.html
muratk
> Its not that I don't believe you. I believe its possible by
> potentially forcing Gnubg into certain situations where it
> doesn't properly identify the true equity of the position
> (ie a weakness in the neural net - and they exist!).
I'm glad some of you guys are finally willing to discuss this.
Actually, it's not the first time but nothing had come out of
it in the past (perhaps because of the closed-mind approach?)
First of all, I should clarify that I am not in any position
to claim that I could exploit "certain weaknesses" of gnubg,
simply because I don't analyze and categorize positions and
games as you people do. I don't know many of the terms that
you use nor do I care to learn. Even the simple ones that I
understand like backgame, race, etc. are probably more vague
to me than to you guys.
Basically I do "just play" and relate my observations here,
which is that I seem to win against gnubg beyond generalized
expectations based on my being unknown in tournament circles,
etc. And I don't blame others for being reluctant to believe
my claims because initially back in the days of jellyfish and
FIBS, I had exceeded my own expectations also.
Then, of course, comes the need to somehow explain this. And
since I have never been measured against other "rated humans"
to claim to be a "world class" player, the only other option
is to conclude that I do well against the bots because I can
exploit the predictability of their moves and dice roll.
Long gone are my nervous days against jellyfish. I think the
confidence you gain from doing well helps you become even a
better player. As I had written before, I am no longer even
consistent with opening moves. Maybe I get bored of routine
moves or try to create more challenge. I have to admit that
I am a BG addict. I play almost every day. Over the years,
I must have played literally thousands of matches against
jellyfish and gnubg.
So, many times I'll say "let me play this differently this
time and see what happens". If it gets me in trouble and if
I can recover from it, maybe I get more satisfaction. When
I chose an inferior move because I think I foresee what is
coming and preempt a joker for gnubg, I enjoy it so much
that I do literally laugh out loud (and sometimes I even
start talking to the bot :))
Anyway, I wrote all these to give an idea of the settings
around the comments I make here in RGB.
> Your use of cheating always make me smile, because I
> doubt you use it in the same context as everyone else.
In this particular case, I thought I was being accused of
(at least potentially) cheating to fabricate a series of
matches in order to claim I could beat gnubg.
> So here is what I'll ask, and I haven't seen anyone ask
> it. Murat, can you email me a copy of the SGF file of the
> 10 matches, I will analyze them, post them to my website,
> and allow r.g.b to discuss it. The one thing I have never
> seen are these actual matches. Often it is hard to decipher
> what you are talking about.
Yes I can email them to you. In fact, I thought about making
them available on my own website but never did because I
didn't think anybody would bother to look at them (not that
I can blame anybody for that, because I myself find it very
boring to look at matches other people had played).
I also can and often do analyze them myself using gnubg but
what good does it do to know that gnubg thinks I make tons
of cube and checker errors?
The only think meaningful and potentially useful would be
to visually observe those matches and try to come up with
a possible explanation for why my seemingly inferior moves
and mishandling the cube produces good results for me.
When I try to explain why I would make a certain move, I
find it to be a very interesting process but even so, I
still believe that it is a waste of time, since the same
position will probably never reoccur or I won't be able
to recognize it anyway.
> Allowing everyone to reviewing the matches can benefit
> everyone.
Ok. Actually I just started playing 33-point matches to
see how similar the results will be. I will save them also.
Over the years, I went through cycles of saving hundreds
of sets of matches and then deleting them thinking that
not even myself would ever look at them again... :(
MK
muratk
> Keepign your anchor roll, and make more blots available. The bot
> will eventually recube early on. When the bot recubes, take and
> then on your next roll recube. I often get 16 and 32 cubes off
> the bat, and I end up with reasonable chances on the backgames.
I had tried stuff like this with jellyfish years ago and there
may still be some discussions in archives about "jacking up the
cube" but this not something that I am not trying to beat gnubg.
Yes, the cube sometimes goes to high too fast but most matches
seem normal length to me and some are quite long and tedious.
Of course, my perception of a "normal length" match may be very
different than other people's.
> Of course what the bot doesn't know is that its neural net
> doesn't evaluate the equity of so many blots as being "good".
> The human knows better (One that can play a good backgame)
This is music to my ears... :)) We just don't hear to often
that humans know better than bots anymore.
Playing BG against bots (and/or bot-like people) is like eating
bio-engineered thick skinned tomatoes without juice or taste. I
really feel bad for people who never had a chance to experience
cubeless, botless backgammon... :((
MK
tc...@lsa.umich.edu
muratk <mu...@compuplus.net> wrote:
>In this particular case, I thought I was being accused of
>(at least potentially) cheating to fabricate a series of
>matches in order to claim I could beat gnubg.
To be clear, the overall point I was making was that even a "showdown" of the
kind you proposed would not necessarily gain you the credibility you might
think it would, because without strong protocols in place, people could
easily "explain away" your achievement by accusing you of cheating.
As an analogy, I have often idly wondered what I would do if I woke up one
morning and suddenly found that I had mysteriously acquired the telekinetic
ability to move objects at a distance. The first thing I would do would be to
experiment with the ability, to see if I could reproduce it at will. Then I
would explore its limits---do I have to be able to see an object to move it?
If not, what is my maximum range? Are more massive objects harder to move?
How fine is my control---if I move an object, do other objects nearby also
move? And so on. At some point I would want to let other people know about
my ability. However, because so many people have fraudulently claimed such
an ability, I would proceed rather differently from most people. I would
try to contact a well-known magician like James Randi, or perhaps someone
like Persi Diaconis or Martin Gardner since I know Diaconis slightly and have
corresponded with Gardner. Having the ability validated by a well-known
skeptic right off the bat would save a lot of time.
>Yes I can email them to you.
Ah...excellent! I will be very interested to see these matches.
>I also can and often do analyze them myself using gnubg but
>what good does it do to know that gnubg thinks I make tons
>of cube and checker errors?
It can be very illuminating. If there is a systematic weakness in the bot
then careful analysis of such matches can potentially expose them.
Back in the days of Marion Tinsley, there was a famous game that he played
against the computer program Chinook, in which Chinook made a move and
Tinsley immediately said, "You'll regret that." For several moves after
that, Chinook could not see what was wrong with its position. Then suddenly,
it did the computer equivalent of an "uh-oh," and its evaluation started to
plummet. Tinsley had indeed precisely identified Chinook's losing move.
Similar phenomena occur today when strong computer chess programs play each
other. The loser usually doesn't see what's wrong (of course---or else it
wouldn't have played the losing move!), but eventually the light dawns.
Tracing back from the clearly lost position one move at a time can be very
illuminating.
I personally believe that the bots are very strong, and that there is nothing
rigged about the dice, but I also believe that many people put too much faith
in them. We're a long way from fully understanding backgammon, and we
shouldn't let the playing strength of the bots delude us into thinking that
they always know the right answer.
Andrew B.
> Back in the days of Marion Tinsley, there was a famous game that he played
> against the computer program Chinook, in which Chinook made a move and
> Tinsley immediately said, "You'll regret that." For several moves after
> that, Chinook could not see what was wrong with its position. Then suddenly,
> it did the computer equivalent of an "uh-oh," and its evaluation started to
> plummet. Tinsley had indeed precisely identified Chinook's losing move.
As given on P129 of http://www.msri.org/publications/books/Book29/files/schaeffer.pdf
.
What it doesn't say is whether Tinsley had analysed the game to the
end, or whether it was an instinctive evaluation on his part.
tc...@lsa.umich.edu
Andrew B. <bul...@gmail.com> wrote:
>What it doesn't say is whether Tinsley had analysed the game to the
>end, or whether it was an instinctive evaluation on his part.
This is not a sharp dichotomy. For a crude analogy, suppose in chess my
opponent blunders away his queen. I immediately know that he has a lost
position. I certainly haven't "analyzed the game to the end," but is it
merely an "instinctive evaluation"? That's a bit of a misleading term in
this context.
Top checkers players can analyze many ply further ahead than is possible
in (say) chess. Tinsley would undoubtedly have analyzed all of Chinook's
leading options already while pondering his previous move. Thus it is
fair to assume that he had already thoroughly prepared a winning plan
to counter Chinook's blunder. But is a carefully worked-out winning
plan an "analysis to the end" or an "instinctive evaluation"? Again,
this is a false dichotomy.
tc...@lsa.umich.edu
muratk <mu...@compuplus.net> wrote:
>On Aug 20, 3:13�pm, tc...@lsa.umich.edu wrote:
>> O.K., then my guess is that it shouldn't be too hard to
>> win a 31-point match in a single game, by studying the
>> dice sequence carefully and then jacking the cube up to
>> 32 early on. �I might try to do this this weekend.
>
You are right that I did not do it this past weekend. I did not forget, and
I try not to use the phrase "did not have time," because after all, everyone
has exactly the same amount of time every weekend---it's just what we choose
to spend that time on. It is true, however, that I decided not to spend the
time this past weekend working on that problem. I did spend a fair amount of
time this past weekend on backgammon, but on other projects, such as rolling
out my large backlog of positions that I want to analyze. Constructing that
position is still on my todo list and as I said, I have not forgotten about it.
I have a question for you. Did you send your matches to Michael Petch?
Michael Petch
On 28/08/09 8:17 AM, in article
4a97e705$0$514$b45e...@senator-bedfellow.mit.edu, "tc...@lsa.umich.edu"
Michael Petch
On 29/08/09 4:52 AM, in article
bfd4003e-d8b2-4b7d...@p10g2000prm.googlegroups.com, "muratk"
<mu...@compuplus.net> wrote:
> On Aug 28, 10:31�am, Michael Petch <mpe...@capp-sysware.com> wrote:
>
>> I haven't received anything yet.
>
>> <tc...@lsa.umich.edu> wrote:
>
>>> Did you send your matches to Michael Petch?
>
> After I made comments that they wouldn't "illuminate" much,
> I didn't know if you guys still wanted them.
>
> I just emailed Petch 2 zip files with 11 matches of 21 points
> and 10 matches of 33 points.
>
> Do what you want with them and tell us what they reveal to you?
>
> To me, the only things they reveal is that gnudung is a rigged
> piece of shit and the people who created it must be sick scums ;)
>
> MK
Michael Petch
The data/matches that Murat sent can be accessed here (The SGF files are
available for download or you can view the html versions in your browser):
http://www.capp-sysware.com/analysis/murat/index.html
I have some views based on cube choices and relative strengths of the
players. The bot only knows how to do an analysis and provide feedback with
the assumption that the player will play like itself (perfect play). If the
weaker player has any basic understanding of assymetric playing strengths
(Human may know he isn't as good as the bot and change their own play
accordingly - especially cubes). Murat clearly plays an aggressive cube
against the Bot.
Jake Jacob's and Walter Trice's book "Can a Fish Taste Twice as Good?" has
very good discussion on this matter.
GnuBG has some assymetric MET's, I'll post later the matches analyzed with
one of these MET's. Might be curious to see how it may perceive the cube
choices.
I can not say that Murat didn't cheat (and I can't say he did either). He
has been kind enough to supply us the data. To remove all doubt regarding
cheating (And sustained high win rates), if something could be set up as Tim
suggested -it would help.
Thanks Murat for sending this initial data.
Michael
Michael Petch
On 29/08/09 7:38 PM, in article C6BF343B.1F3AA%mpe...@capp-sysware.com,
"Michael Petch" <mpe...@capp-sysware.com> wrote:
>
> The data/matches that Murat sent can be accessed here (The SGF files are
> available for download or you can view the html versions in your browser):
>
I meant to point out that the Grandmaster checker/Worldclass cube choice for
the analysis was done because those were the settings Murat used while
playing the bot.
The SGF files posted are not as Murat sent them. I redid an analysis for
each file for both players (Murat's files weren't completely analysed for
his own play).
muratk
> I have some views based on cube choices and relative strengths
> of the players. The bot only knows how to do an analysis and
> provide feedback with the assumption that the player will play
> like itself (perfect play).
I criticized this approach many times in the past and called it
"jacking oneself off"...
Not only that the result of playing against oneself 4 billion
times ends up far from "perfect play", but the bots and their
worshippers don't even follow their own advice/learning...
If they had, most games wouldn't go past the first few rolls,
because the stats already tell you what would happen after 4
billion tries...!
Of course, no human can live long enough to play that many
games, so they may be excused for being wishful thinkers to
hope that the stats will be lopsided and that the 4 million
games that they will play will "beat the statistics"... :)
But what about the bots...? They can exist forever and play
an endless number of games. So what don't the piece of shit
bots accept defeat at the first roll when the stats tell
them that they will lose in that position even after 4 billion
tries...?? Why...??? Can some cocksucker answer me this...???
> If the weaker player has any basic understanding of
> assymetric playing strengths (Human may know he isn't
> as good as the bot and change their own play accordingly
> - especially cubes).
Does this mean that you assume the human to be the weaker
player?
If yes, then the rest of what you can say on this will not
go beyond "world-class, phd-class, etc. bullshit"... :))
> Murat clearly plays an aggressive cube against the Bot.
I never liked the cube but once I submitted to the fact
that I had to live with it, I came to believe/practice
that you should use it, almost as another checker, not
just let it sit there.
Personally, I end up wondering way too often why gnudung
won't double and eventually end up saying "Ahaa, the scum
knew what was coming", because almost invariably I seem
to win after such "timid" cube handling by gnudung...
> GnuBG has some assymetric MET's, I'll post later the
> matches analyzed with one of these MET's. Might be
> curious to see how it may perceive the cube choices.
Even if I can't rise to you guys' level to be part of
such discussions, it will be very interesting to read
and try to understand.
> I can not say that Murat didn't cheat (and I can't say
> he did either).
How about gnudung..? Can you say it cheated or not..?? :)
> He has been kind enough to supply us the data. To remove
> all doubt regarding cheating (And sustained high win rates),
> if something could be set up as Tim suggested -it would help.
In the player settings, I see an entry for "socket". It sure
looks like it would be an easy thing to let me play against
gnudung over the Internet if somebody could handle the other
end.
That would be much more fun because I would get to make
instant comments... :)) Not as fun as playing against gnudung
in front of real people while looking at their faces, though.. :(
MK
muratk
> The data/matches that Murat sent can be accessed here
> (The SGF files are available for download or you can
> view the html versions in your browser):
> http://www.capp-sysware.com/analysis/murat/index.html
I was amazed and impressed by the colorful and detailed
graphic analysis apparently produced using gnudung, so
much that I scrolled through a good number of them.
While I'm at it, an before any PHD's stick their necks
out to make comments :)) let me say a few things that
crossed my mind (not in any logical order).
1) When gnudung has 4 men on its 1-point while I have 2
on my 1-point, it marks it as an *unlucky* roll if it
doesn't roll a double. When I have 2 men on my 2-point
and gnudung has 2 on its 1-point, it marks it as *lucky*
roll if I don't roll a 1. I just don't get what gnudung
means by *lucky/unlucky* I guess... :(
2) When I made a purposefully inferior move, I thought
that I did so because I could read the future dice but
looking at the analysis afterwards, I see that I didn't
benefited from such moves immediately to support that.
In fact, many times I didn't benefit at all and I paid
dearly for such moves.
3) What matters though is that I recovered from them
often enough to win. As funny as it may sound to you,
I came to believe that I am the stringer player and
because of that I feel insecure playing matches shorter
than 20-points, thinking that I would not have enough
time to recover from my mistakes (along with the idea
that longer matches favor the stronger player). On the
other hand, I don't like matches over 40-points because
they tend to look like basketball matches when the last
few minutes determine the outcome.
4) A few of the bad moves are due to mis-clicking while
trying to play too fast. But, hey, I even recovered from
most of them :))
5) When gnudung says my checker play is awful I take it
personally :( but not so with the cube play. I never
agreed that there was anything to exaggerate about the
so-called "cube skill"... When you take statistics over
4 billion games and don't even play according to your
own findings, all that "doubling window", etc. is pure
bullshit...! Based on 4 billion games, when you reach
51% winning position, you double. Very simple... :))
6) Disclaimer: almost all moves after breaking contact
is produced by clicking the "End Game" button. Maybe
some of my best rating come from those moves gnudung
makes for me.. :))) If you look though, after breaking
contact, I almost never win... :((( So, lesson for
myself: "Do all you can to keep contact"...!!
7) About early doubles in the game: I will almost always
tale them but also almost always wonder to myself "this
piece of shit bot must know something that I don't"...
And, yes, often it does... :)) You can see it too, no..?
8) The real question is: do you ever get suspicious about
gnudung's checker moves and try to guess the upcoming
dice rolls..? Do you do well in guessing..??
9) OK, enough of this baloney. One last comment: gnudung
more often rates me "Awful" but occasionally "Supernatural".
I thought once you were "Supernatural" you were always
"Supernatural"... :))))
MK
tc...@lsa.umich.edu
>and get the same dice rolls for an entire match do this:
>
>A) Click "New" icon.
>B) Enable "Manual Dice" checkbox
>C) set any other player options that you wish.
>D) Click "Ok"
>E) The dice selection screen will appear. Don't select a roll, just hit
>"Close"
>E) Go to Options/Settings/Dice tab and
> a) Set the random number generator to Mersenne Twister (or which ever
> seeded PRNG you like)
> b) set the seed to some value (3101)
> c) Click "OK"
>
>[At this point the seed is set]
>
>F) Click anywhere on the board and you should get a message along the lines
>of "A new match" has been started.
>G) Play the match
This worked almost as advertised, with one glitch. I set the match length to
31 games, but when the new match started, it reset to 7 games. I couldn't
figure out how to get a 31-game match.
Anyway, as a proof of concept, I carried out the plan I described previously,
with a 7-point match and with the Mersenne Twister seed 3101. That is, I
played out a match mindlessly just to see what the dice rolls would be, and
I wrote them all down. Then I started a new match with the same seed, and
played "for real," now with the crushing advantage of knowing exactly what
the rolls would be. In the first game, I doubled at my first opportunity.
I left a blot where gnubg would hit it loose because I knew I could double
hit back and launch a blitz. Blitzing is pretty easy when you know what
your opponent is going to roll. :-) I closed gnubg out, meandered my
checkers around and started bearing off. Bearing off aggressively is also
easy when you know what your opponent is going to roll, but since I had
closed out only two checkers, I couldn't quite manage a backgammon and
had to settle for a gammon. I was now leading 4-0 to 7.
In the next game, it was easy to get the cube up to 8 to play for the match.
I made the golden anchor and sent four of gnubg's checkers back, but couldn't
see how to prevent gnubg from anchoring on the 3-point. However, by hitting
the loose checkers I gained plenty of time to build a prime and choose an
opportune time to break anchor when I knew gnubg wouldn't be able to hit me.
Bearing off safely with several of gnubg's checkers hanging out in my home
board did require looking ahead several rolls; this was probably the trickiest
part of the match. But ultimately it wasn't too hard to do, and I overkilled
by winning another gammon.
Anyway, I'm now pretty confident that if I put the time in, I could quite
easily run a scam of the kind I described earlier, where I would boast of my
ability to beat gnubg in a match played over the web. I would need to get
the random seed, of course, but there are plenty of unsuspecting victims
out there that would give the seed to me for free. It might be too slow
to get the dice rolls by playing a match with gnubg on the side, but since
the gnubg source code is freely available, I could easily write a little
program that would instantly produce the dice rolls for me the moment I
got the seed, without having to go through the gnubg interface. Right now,
my ability to study the upcoming rolls and plan my strategy is poor, and so
I would be playing suspiciously slowly, or playing suspiciously "wild" moves
that make no sense unless you're cheating. However, I'm sure that with some
training I could get better at it.
I emphasize that I'm not accusing Murat of cheating in the above manner---
I'm just pointing out that it could be done, and hence that a match played
under uncontrolled conditions would not prove anything about one's ability
to beat gnubg under fair circumstances.
I sent the .sgf file of my match to Michael Petch, who perhaps can post it
alongside Murat's matches.
Michael Petch
On 30/08/09 8:52 PM, in article
4a9b3ae9$0$513$b45e...@senator-bedfellow.mit.edu, "tc...@lsa.umich.edu"
<tc...@lsa.umich.edu> wrote:
>
> This worked almost as advertised, with one glitch. I set the match length to
> 31 games, but when the new match started, it reset to 7 games. I couldn't
> figure out how to get a 31-game match.
>
Funny you are right. When you click on the board it reverts to 7 points. I
found that the value 7 comes from the default setting in
Setting/Options/"Match Tab" and default match length (C
There may be an easier way to do all this. IN fact I know Murat is able to
do it properly - maybe he knows of an easier way I am unaware of. I had
Murat send me another SGF file as a test to see if he knew how to set the
seed properly before a match, and I can confirm the output in the SGF file
matched the same rolls I saw in another match he sent for the same seed.
Michael Petch
On 30/08/09 10:27 PM, in article C6C0AD5F.1F3ED%mpe...@capp-sysware.com,
"Michael Petch" <mpe...@capp-sysware.com> wrote:
> maybe he knows of an easier way I am unaware of.
There is an easier way (I eluded to it in another post) involving using the
command line within the GUI.
Click on "view" menu/"Panels>" and put a click by "Command". On the right at
the bottom you should now have a Command box with a help button next to it.
First set the random number generator to Mersenne Twister (or one of the
other PRNG's), and then to set a seed and play a match all you type is this
in the command box:
set seed 3101
new match 31
And that's it. You should be prompted that a new match is starting and that
should do it! Set the seed value to whatever you wish, and the value 31
above can be replaced by the match length you wish. Not using the
traditional GUI interface can be beneficial.
Michael
Michael Petch
On 30/08/09 10:37 PM, in article C6C0AFAC.1F3F2%mpe...@capp-sysware.com,
"Michael Petch" <mpe...@capp-sysware.com> wrote:
> eluded
Alluded ;-)
tc...@lsa.umich.edu
muratk <mu...@compuplus.net> wrote:
>But the same feature would also make it very easy to
>fabricate matches. Right now long matches played with
>known seed values have some credibility because one
>can verify the rolls to be the same and fabricating
>long matches without being able to take back moves
>would take too much time and effort.
While I produced only a 7-point match, I think you're grossly overestimating
the difficulty of fabricating a long match. Once you have control of a game,
it's not hard to maintain that control if you have foreknowledge of the dice.
It's usually not necessary to have deep insight into gnubg's checker-play
strategy. For example, it's easy to predict gnubg's checker play when it
has two checkers on the bar. :-) Similarly, in many other positions it's
obvious what gnubg will do, and when it's not obvious it may not matter
anyway.
Once in a while you won't be able to gain control and you may lose, but that
will just make the match look more "authentic." The huge advantage of
knowing the dice is sure to make itself felt in the long run, so in a sense
the longer the match is the more confident you can be that you will come out
on top, without any need to backtrack.
tc...@lsa.umich.edu
muratk <mu...@compuplus.net> wrote:
>I would be curious to know how long it took you to "fabricate"
>your 7-point match...?
I'm not sure...maybe half an hour? There were a couple of false starts
because it took me a while to learn how to set the seed properly. Then
the first time I didn't write down enough dice rolls so I had to restart.
I also got distracted at one point trying to see if I could win a backgammon
instead of a gammon, and restarted unnecessarily.
>And how "authentic" it looks...??
I'm sure it doesn't look authentic. It was my first attempt. As I said
before, with practice I'm sure one can do better. Perhaps Michael Petch can
post my match somewhere.
>Why don't you fabricate 10 matches of 7-points and 10 matches
>of 31-points, "authentic" looking of course, and tell us which
>was easier...?
Answering your question literally ("Why don't I...?") the answer is that the
exercise was interesting the first time around, but it's not so interesting
now. It would just be tedious. Also, to answer your question from another
article, the purpose of my doing this is to illustrate that a skillful cheat
could easily carry this out. I'm not claiming to be a particularly skillful
cheat myself---like anything, getting good at cheating takes practice. The
point is that a properly motivated person could easily get good enough at
fabricating plausible matches to do it in real time. A professional magician
performs much more impressive feats live, in front of spectators, all the
time.
>Unlike you, I am offering to duplicate my success against the
>gnudung piece of shit live, in front of spectators, if someone
>can arrange it to play against me over a network...!
It seems you're so conditioned to arguing against people that you're failing
to see how my observations are helping you design your experiment in a way
that will convince more skeptics.
>What do you say to that, "Moses"...??? :)))
I say that you're free to do whatever you want, but don't be surprised if
people don't accept your success at face value, because the alternative
explanation (that you're a skillful cheat) cannot be ruled out *if* the
random seed is sent to you in the clear.
It did occur to me that there might be a way to plug this loophole. As I
explained previously, if someone sends you an encrypted random seed, then
that would allow post-mortem verification that a fixed seed was used. Now,
the Mersenne Twister algorithm becomes predictable after a while, but this
loophole can again be plugged if we simply agree to refresh the seed after
a certain period of time. We'd have to check the details of the generator
to be certain, but the number of dice rolls in a 31-point match is probably
small enough that refreshing the seed after each match is good enough.
Wikipedia says that 624 observations of MY19937 is enough to recover the
seed, but I think this assumes that you see the full 32-bit word each time,
so it probably translates into more than 624 dice rolls.
I have another question for you. Do you claim that your advantage over gnubg
disappears if some other random-number generator is used? There are several
different options in gnubg for the type of generator. Or do you claim to be
able to beat gnubg using a variety of generators?
Michael Petch
On 01/09/09 9:11 AM, in article
4a9d39aa$0$514$b45e...@senator-bedfellow.mit.edu, "tc...@lsa.umich.edu"
<tc...@lsa.umich.edu> wrote:
> I'm sure it doesn't look authentic. It was my first attempt. As I said
> before, with practice I'm sure one can do better. Perhaps Michael Petch can
> post my match somewhere.
I can do that later today.
Btw, one other method of cheating is to simply use the same seed and then
play some of the early rolls differently. Once the match is altered
significantly, even the same dice can produce wildly different outcomes.
Then you can just play a bunch of matches with same seed and choose the ones
you won, and post what you wish people to see.
I am not saying that happened here, but its also another way without
recording dice from the same seed, to produce differing outcomes.
tc...@lsa.umich.edu
muratk <mu...@compuplus.net> wrote:
>Yup, one or the other... For the thousandth time, "I DON'T
>NEED TO KNOW THE SEED"...!!!
In that case, if you're happy with the opponent choosing a different seed
for each match, and not sending them until after the matches are over,
then your conditions sound reasonable. Anyone want to take Murat up on
his offer? I don't have the technology to do it myself.
I didn't see an answer to my question about using dice generated by some
method other than the Mersenne Twister. Does your offer stand for other
dice generation methods too?
muratk
> I didn't see an answer to my question about using dice
> generated by some method other than the Mersenne Twister.
> Does your offer stand for other dice generation methods too?
No. I have never played enough with manual dice or even
with other built-in dice generators to have anything
more than an impression that I didn't do as well.
Besides, it wouldn't be fun with manual dice :)
BTW guys, thanks for participating in this thread which
lasted way more than I expected. In fact I just noticed
that especially at the beginning I posted several times
without any swearwords. I really must have gotten into
the subject to forget my funny swearing... :))
Fortunately it doesn't take me long to realize nothing
is worth taking so seriously... ;)
MK