Why do we THINK that bots cheat?
alan...@my-dejanews.com
Let's assume, contrary to our best instincts, that JF, Snowie, MonteCarlo and
the other bots on FIBS so often get the perfect roll at the perfect time
simply because of the laws of averages .... (I SAID it was contrary to our
best instincts, bear with me!) .... Then what is it about human nature that
makes us leap to the assumption that they do cheat? I would never make that
leap in playing against a human player. Or at least, I never have ... nor
have I ever felt that a human opponent was getting such good rolls that
cheating was a more plausible explanation than good luck. I have played many
more games against human players than against bots, so if I believe in the
law of averages, there should have been more instances of good rolling by
human opponents than by bots. Is it just that the bots make better use of
the rolls they get?
I can imagine this to be true in the case of the bots that are extremely
highly rated. Since I rarely get to play against humans in the 1900-2000
range, I can't say for sure that their rolls would not appear inhumanly good
if I did play them. Thus, the improbably good fortune of JF, Snowie and
yabs, and their ilk, may just be due to exquisite use of ordinary rolls.
But what about MonteCarlo? MC seems to get the ideal roll (you know the
drill: making an indirect shot off the bar against a 5-point board, rolling
boxcars just when needed to win a race or escape from a near-prime, etc.)
nearly as often as JF or yabs, yet otherwise plays at a level (and rating)
about as mediocre as my own. MC's luck seems much better evidence of bot
cheating than that of the higher ranking bots.
Of course, the high ranking bots could get their ranking by cheating too; but
my point is that, in Monte Carlo's case, the overabundance of lucky rolls is
not likely to be due to brilliant play in setting up opportunities.
What do y'all think?
alanback
-----== Posted via Deja News, The Leader in Internet Discussion ==-----
http://www.dejanews.com/ Now offering spam-free web-based newsreading
Tom Vavasour
in reply to alan...@my-dejanews.com
>drill: making an indirect shot off the bar against a 5-point board, rolling
>boxcars just when needed to win a race or escape from a near-prime, etc.)
>nearly as often as JF or yabs, yet otherwise plays at a level (and rating)
>about as mediocre as my own. MC's luck seems much better evidence of bot
>cheating than that of the higher ranking bots.
>
>Of course, the high ranking bots could get their ranking by cheating too; but
>my point is that, in Monte Carlo's case, the overabundance of lucky rolls is
>not likely to be due to brilliant play in setting up opportunities.
Just more of the same. I'm struggling along at around 1580 in ratings
on fibs. I thought it was a bad run but it's been going on long enough
that there must be a basic flaw in my playing technique. That said, I
seem to win about 75% of my matches against MonteCarlo.
I personally haven't noticed MC roll killer dice with me for ages. You
must be getting them all :)
...Tom
They can't break you if you don't have a spine
-Wally from Dilbert
Patti Beadles
In article <35730616...@newsch.es.co.nz>,
Tom Vavasour <tom...@ihug.co.nz> wrote:
>Just more of the same. I'm struggling along at around 1580 in ratings
>on fibs. I thought it was a bad run but it's been going on long enough
>that there must be a basic flaw in my playing technique. That said, I
>seem to win about 75% of my matches against MonteCarlo.
You might question what the difference is between you playing
MonteCarlo and playing live opponents.
One likely one is that you're probably not playing 1-pointers all the
time against live opponents, while that's all MC plays. If this is
indeed true, then you may be a good checker player but weak with the
cube.
-Patti
--
Patti Beadles | Not just your average
pat...@netcom.com/pat...@gammon.com | degenerate gambling adrenaline
http://www.gammon.com/ | junkie software geek leatherbyke
or just yell, "Hey, Patti!" | nethead biker.
MJR
> But what about MonteCarlo? MC seems to get the ideal roll (you know the
> drill: making an indirect shot off the bar against a 5-point board, rolling
> boxcars just when needed to win a race or escape from a near-prime, etc.)
> nearly as often as JF or yabs, yet otherwise plays at a level (and rating)
> about as mediocre as my own. MC's luck seems much better evidence of bot
> cheating than that of the higher ranking bots.
>
> Of course, the high ranking bots could get their ranking by cheating too; but
> my point is that, in Monte Carlo's case, the overabundance of lucky rolls is
> not likely to be due to brilliant play in setting up opportunities.
>
>
> alanback
>
Hi,
I am the author of MonteCarlo, and I can assure you that it does not
cheat. (How can I make it cheat even if I wanted it to? Its just
another player as far as FIBS is concerned).
I think part of the reason it seems to win more than it should is
because it makes plays which are obviously wrong, particularly when
bearing off against contact. I would say that humans tend to become
good players along a different path than a neural net does. With just a
little experience, a human learns how to bear off safely. So when
playing aginst another human, if you notice them making "beginner"
mistakes in the bear off you may make some assumptions about their level
of play. Losing to a player who you think isn't as sharp as you is
usually attributed to their getting lucky.
These assumptions aren't valid against a computer player. In technical
positions, humans can select the best play even when the difference
amongst the candidate moves is very small. Neural nets tend to have
difficulty selecting THE best play in these types of positions. (But,
atleast the cost of the error is minimal. ) When a neural net makes a
mistake in these positions, its easy to assume that they are missing
some key understanding of the game, then make assumptions about their
skill, followed by the feeling that they got lucky when they win.
But, think about this: If the neural nets are making obvious mistakes
in some situations, then they must be making up for it by superior play
in other positions. The positional play of the neural net is usually
stronger than it seems. I think this is because its easy for humans to
think there's not much difference between 2 plays when really there is.
So a good positional play by a bot is likely to go unnoticed.
I'd say a good way to characterize a neural net is that its play is a
sequence of small errors, as opposed to human play, which is a sequence
of correct plays with large blunders scattered about. The difference
between a good human and a good bot is that the good human will rarely
make technical mistakes but a good bot commonly will. So a human is
better able to convince you of their mastery of the game.
---Matt Reklaitis
Patti Beadles
In article <357337FC...@ptd.net>, MJR <hac...@ptd.net> wrote:
>The difference
>between a good human and a good bot is that the good human will rarely
>make technical mistakes but a good bot commonly will.
And a bot will never blunder by overlooking a play, or make a bad
choice because it's tired or gets distracted by somebody walking by,
or...
Hank Youngerman
Every so often when I am playing a weaker player who objects to my
"good luck" in an unsportsmanlike way, I make the remark "Good players
make their own luck." And that's the point of good checker play.
Good players get more good rolls because they MAKE more good rolls.
That's how you win at backgammon. 'Bots are very good players, so
they make more good rolls. Shouldn't we expect them to come up now
and then?
John Goodwin
On Mon, 01 Jun 1998 14:24:43 GMT, alan...@my-dejanews.com wrote:
..asking why we think that the robots cheat.
***
Imagine that you decided to play a game of backgammon against a human.
They agreed, but said that they would tell you what each dice throw
was.
Not that they would throw the dice in your view, simply that they
would tell you what the throws were.
Imagine that they then proceeded to wipe the floor with you.
Time after time after time.
How many "lucky" dice would they get before you began to suspect they
were cheating?
Probably none, because you would not play a game on those terms.
The question should be why do we *not* think that the robots cheat.
There are three reasons (that I can think of):
1) Lack of motive
2) Reassurance of others who play the same robot (particularly
experts who are in a better position to see that each robot gets
the "luck" that it deserves).
3) Facilities incorporated in the program to ensure that it does not
cheat
One or more of these allow us to have fun and learn, and make the
robots a valuable contribution to the game.
JG
David Smyth
On Tue, 2 Jun 1998 01:41:56 GMT, pat...@netcom.com (Patti Beadles)
wrote:
>In article <357337FC...@ptd.net>, MJR <hac...@ptd.net> wrote:
>>The difference
>>between a good human and a good bot is that the good human will rarely
>>make technical mistakes but a good bot commonly will.
>
>And a bot will never blunder by overlooking a play, or make a bad
>choice because it's tired or gets distracted by somebody walking by,
>or...
>
And a bot knows no fear. It can be battered by the dice and will
never flinch. Unlike all but the most steely nerved human players...
Patti Beadles
Interesting but flawed analogy, in the case of online bots.
For bots who play on FIBS and other servers, a better analogy is that
you agree to play with someone, but a third party, someone entirely
unconcerned with the outcome of the game, throws the dice and tells
you what the rolls are.
It now becomes a question of whether you trust the person throwing the
dice. And in the case of a backgammon server, since he has no
incentive whatsoever to do anything other than give honest rolls, I
would.
-Patti
--
Patti Beadles | Not just your average purple-haired
Vince Mounts
Hank Youngerman wrote in message <357365e2....@news.mindspring.com>...
This same basic response has been seen a number of times in this thread and
I question its validity in some aspects. Don't get me wrong. I entirely
agree with the statement that good players have more good numbers. If that
weren't the case then there would be no such thing as good and bad
players,,, just lucky and unlucky players. What I question is wether this
explaination goes to the heart of what the posters were talking about in the
first place. For example, I play Jellyfish with manual dice now-a-days
becuase my game suffered from my distrust (however unfounded i know it to
be) of its self-generated dice. The reason is not becuase it got good rolls
when it was clear that lots of rolls are were good. I am no pro that is for
sure(holding at around 1670 on fibs for a while now) but what bothered me
was the number of miracle rolls. And again though I am not a pro I am good
enough to spot a miracle when i see it. Multiple high doubles in a pure
race. Hitting indirect shots with perhaps a 1 in 18 chance in very critical
situations. Getting in from the bar just at the right time. Hitting all the
critical 1 in 3 chance shots. Rolling one of three doubles(or less) from the
bar to get 2 men in from it just in time to save getting gammoned etc etc.
All things that are unlikely. Despite the fact that the bot played well to
that point in the game,,, the bot is in trouble and it rolls its miracles a
little too often(or so it seems to some).
I think these are the rolls that people are complaining about becuase with
even a minimum amout of skill a player can recognise when there are lots of
rolls that will be good for their opponent. The thing to realize of course
is that ,,, IT HAPPENS!!! The main problem here though is that becuase the
bot itself is generating the rolls it is human nature to question them. As I
stated on this group before in another post, that is why I only play
Jellyfish with manual dice. I am mediocre as it is and often times a real
steamer when bad luck comes my way. I don't need any questions lurking about
the dice to make my play even poorer than it already is.
-----
Vince Mounts (a.k.a einniv)
E-Mail: vmo...@mindspring.com
Home Page URL: http://vmounts.home.mindspring.com
alan...@my-dejanews.com
Thanks to all for your thoughtful and thought-provoking responses. I was
especially impressed that Matt met the question on its own terms and wasn't
the least bit defensive. Hank's bit of banter reminded me of my own favorite
response (when opponents complain I am excessively lucky) that "I often feel
that way myself against a better player."
Matt, in my relatively rare moments of extreme paranoia (and based on bits
and pieces of circumstantial evidence gathered during 3 years on FIBS) I
believe that the 'bots log on to FIBS through a different port which gives
them access to special cheat commands. However, I don't believe that often,
and most of the time (between matches at least) I just grumble about my luck
against EVERYBODY ;-) ...
Thanks again.
Alan
John Goodwin
On Tue, 2 Jun 1998 15:23:39 GMT, pat...@netcom.com (Patti Beadles)
wrote:
>Interesting but flawed analogy, in the case of online bots.
He asked why we think that the robots cheat, not why we think the two
party server/player cheats.
I'll agree that he mentioned robots playing on FIBS, but then we must
either assume that he knows that the server throws the dice, in which
case he either thinks the server and robot are cheating in collusion,
or the question makes no sense, or we assume that he thinks that for
some reason the server lets the robot throw.
In each case, my points stand, because all he sees is some entity, or
composite entity that controls both his and his opponent's dice.
>For bots who play on FIBS and other servers, a better analogy is that
>you agree to play with someone, but a third party, someone entirely
>unconcerned with the outcome of the game, throws the dice and tells
>you what the rolls are.
That's not an analogy, that's a description of what's happening!
(Allowing that in the electronic domain you cannot tell the difference
between someone and something.)
>It now becomes a question of whether you trust the person throwing the
>dice. And in the case of a backgammon server, since he has no
>incentive whatsoever to do anything other than give honest rolls, I
>would.
That was the first of my reasons why we do not, generally, think that
the robots (or composite entities) cheat.
JG
Rodrigo Andrade
I never played MC, but, when I play on FBS and I complain, or make comments
about a cheater, 9 out of 10 replies are some variation of the famous "Play
Monte Carlo then we'll talk..."
Rodrigo
Rodrigo Andrade
but what bothered me
: was the number of miracle rolls. And again though I am not a pro I am
good
: enough to spot a miracle when i see it. Multiple high doubles in a pure
: race. Hitting indirect shots with perhaps a 1 in 18 chance in very
critical
: situations. Getting in from the bar just at the right time. Hitting all
the
: critical 1 in 3 chance shots. Rolling one of three doubles(or less) from
the
: bar to get 2 men in from it just in time to save getting gammoned etc
etc.
Amen... I wonder how can anyone take that software seriously and still pay
as much as $250 for it... If you don't believe in miracles, play Jellyfish
and you will.
Rodrigo
John Goodwin
On 2 Jun 1998 23:25:30 GMT, "Rodrigo Andrade"
<candrade@_R_E_M_O_V_E_wt.net> wrote:
snip
>If you don't believe in miracles, play Jellyfish
>and you will.
Despite all that has been written on this NG, you still seem quite
convinced that JF gets "miracle rolls".
This is a (very) poorly disguised way of saying that you believe it is
cheating.
I recently posted an article entitled
"Satisfying yourself that Jellyfish does not cheat"
For your purposes, you might like to pretend that the title was:
"How to prove that Jellyfish does cheat"
The article gives step by step instructions that would enable you to
show quite conclusively that JF was cheating (if it ever did).
You would have concrete evidence that you could post to the group.
You would also make quite a name for yourself as the person who
finally managed to prove to all us gullible fools that we had been
taken for a ride.
So all you have to do is read the article:
Message-ID: <356bead7....@news.demon.co.uk>
then wait for a really blatant example of JF cheating, and follow the
simple instructions.
Then post your evidence to this NG, and wait for the plaudits and
laurels to which you would most certainly be entitled.
JG
Vince Mounts
John Goodwin wrote in message <35750589....@news.demon.co.uk>...
>On 2 Jun 1998 23:25:30 GMT, "Rodrigo Andrade"
><candrade@_R_E_M_O_V_E_wt.net> wrote:
>
>snip
>
>>If you don't believe in miracles, play Jellyfish
>>and you will.
>
>Despite all that has been written on this NG, you still seem quite
>convinced that JF gets "miracle rolls".
>
>This is a (very) poorly disguised way of saying that you believe it is
>cheating.
Not sure who this directed at but I wasn't using that wording to imply that
it actually was cheating, just that during play such rolls are easy to spot
and that since it bothered me so it affected by play so I roll manually now.
John Goodwin
It wasn't directed at you.
JG
Bill Taylor
"Vince Mounts" <vmo...@mindspring.com> writes:
|> The main problem here though is that becuase the
|> bot itself is generating the rolls it is human nature to question them. As I
|> stated on this group before in another post, that is why I only play
|> Jellyfish with manual dice.
AT LAST! Sense.
As this option is available, I can't undesrstand why *anyone* who has
suspicions just doesn't do this, then all is sweetness and light.
Hence (and otherwise!) these endless debates are so STUPID...
-------------------------------------------------------------------------------
Bill Taylor W.Ta...@math.canterbury.ac.nz
-------------------------------------------------------------------------------
God does not play dice with the universe - god *is* the dice.
-------------------------------------------------------------------------------
_______
/\ o o\
/o \ o o\_______
< >------> o /|
\ o/ o /_____/o|
\/______/ |oo|
| o |o/
|_______|/
-------------------------------------------------------------------------------