Enter a Computer Next Year.
linuxandpoker
next years WSOP 10K No Limit Hold'em event.
It appears that the strategy has become so formulaic that it
would be "relatively" easy to program an application that could
compete. I will not make the same claim for PL or limit Hold'em.
I can see it now, a computer only supersatellite where the winner
gets a seat at the "Big Dance".
What do you say RGPers, would you sign said petition?
Yours truly,
linuxandpoker
Seth Peck
think, on the other hand, that any good player would have enough
variability to fool a hand-analyzing / strategy program when the big hand
comes, and would know when not to bet big (as the program would either
fold automatically or try to come over the top, given the hand / pot
odds).
I don't think any tournament director would mind as long as the event fees
were paid, and I think many players would be up for the challenge (or
taking down possibly the biggest fish in the room).
Seth
Paul Phillips
Seth Peck <pe...@CS.Arizona.EDU> wrote:
>I think, on the other hand, that any good player would have enough
>variability to fool a hand-analyzing / strategy program when the big hand
>comes, and would know when not to bet big (as the program would either
>fold automatically or try to come over the top, given the hand / pot
>odds).
Huh? Why would the program "automatically" do anything? The presumption
I think is that the program is written by a capable poker player, not a
poor one. It should be 10x harder to read than any player since there
would be no physical cues.
It would never happen though. If the original suggestion was serious,
I don't think the suggestor thought through any of the logistics.
--
Paul Phillips | You deserve a longer letter than this, but it is my
Vivid | unhappy fate seldom to treat people so well as they
Empiricist | deserve. -- Jane Austen
pull his pi pal! |----------* http://www.improving.org/paulp/ *----------
Thomas R. Hummel
r00lbot!"
In article <74990c89.02052...@posting.google.com>,
linuxa...@yahoo.com says...
Topset
Don't give this another thought. Your pot limit technique makes a random
number generator look predictable. It would be nearly impossible to commit
your style to code....:)
I am still looking for a Mississippi Stud game.....
"Paul Phillips" <rg...@improving.org> wrote in message
news:ack9mq$e1i$1...@spoon.improving.org...
Seth Peck
> Huh? Why would the program "automatically" do anything?
My bad, I shoulda used a different word, or at least clarified...
Creating an effective, winning poker playing "program" (I use quotes
because in my mind a program involves more than just a strategy), the
coding involved would store important data on each player on each play of
the hand (how many chips were available, who was playing, how long they
hesitated before playing, pot odds, etc). This data would be important
because a program wouldn't be able to take advantage of facial/audible
tells, so the data would be used to create a "read" on the opponents.
Combine this with a chart of the odds of winning a hand against X players
with the two cards the computer player is holding (based on previous or
current game statistics), and the computer player would be able to
determine whether to fold, call, bet, raise or reraise.
Now, when I say "automatically", it means that, unlike a player who takes
his time to decide whether to stay in a hand or not, the computer player
would simply consult a table, run an algorithm and/or choose a play.
Since everyone would know it's a program making the play, it wouldn't
"take its time" unless the programmer wanted to institute a random wait
time before announcing its move.
In the situation that I was describing, i.e., heads-up against a rock with
a possible all-in, if the rock overbets to try to buy the blinds or draw
out an all-in raise, the computer is going to "automatically" fold, call
or raise (possibly all-in) given the little data it has about the player
and the odds of winning with the cards in front of him.
I still think it's an interesting idea, but players could totally take
advantage of the program, even with a modest amount of randomness in the
program's strategy.
Seth
Paul Phillips
Seth Peck <pe...@CS.Arizona.EDU> wrote:
>I still think it's an interesting idea, but players could totally take
>advantage of the program, even with a modest amount of randomness in the
>program's strategy.
But... how? You've asserted that but it sounds to me like players could
take advantage of whatever algorithm you have in mind. It's far from clear
in the general case. In fact, from here it's obviously not true. You're
making a pile of assumptions about it -- that it's using an exploitable
algorithm, about how far it goes in assuming past behavior dictates future
behavior, and basically about how well it can be predicted.
Add to this that the players have to worry about seven other players at
the table in addition to the robot. They can't be basing their whole
strategy on fooling the robot -- even if they knew what they needed to
do to exploit it, which they wouldn't.
Put another way: just what is it about what humans do when playing no limit
holdem that you think is so irreproducible in software? I think I'm pretty
good at the game and I believe I could write a program that's at least as
good and probably better than I am, if I were motivated to do so.
--
Paul Phillips | Glower to the peephole!
Apatheist |
Empiricist |
pp: i haul pills |----------* http://www.improving.org/paulp/ *----------
Robert Ladd
attributes were necessary to win, I think that a computer could be
programmed with not only the lightning quick math that would be necessary,
but could also be programmed with some artificial intelligence so as not to
always follow that math.
The "tells" part will work both ways though. No one will be able to pick up
on the computer's ear twitch when it has pocket Aces, but the computer
wouldn't be able to pick up on the table talk that sometimes might be a
giveaway.
My guess is the biggest problem with tournament poker is the changing
blinds, stack size relative to everyone elses, and varying your play as
these things change. The computer would do this flawlessly, and depending
on how good the information was from the expert tournament player that was
recruited to do the analysis, this would be a great advantage.
But, I think the biggest advantage the computer would have would be the
unemotional detachment that most humans don't have. It would simply look
for the best play, based on it's choices and execute that play. No second
guesses, no changing it's mind because of a vicious stare. And as time
went on and it stored away info on certain people's play it would get better
and better.
With the luck factor being undeniably very high, I'm wouldn't suggest that
it would win a high percentage of tournaments, but I think if a truly good
tournament player got together with a good analyst a program could be
designed that would win as much or more money than many of the top pros.
Robert Ladd
"Paul Phillips" <rg...@improving.org> wrote in message
news:ack9mq$e1i$1...@spoon.improving.org...
Andrew Bloch
> In article
<Pine.GSO.4.33.020523...@lectura.CS.Arizona.EDU>,
> Seth Peck <pe...@CS.Arizona.EDU> wrote:
> >I still think it's an interesting idea, but players could totally take
> >advantage of the program, even with a modest amount of randomness in the
> >program's strategy.
>
> But... how?
Maybe he's trying to say that players could collude. Of course, that could
happen to human players also.
Andy Bloch
A. Prock
[snip]
>but could also be programmed with some artificial intelligence so as not to
>always follow that math.
!?!
'cause artifical intelligencs is to math as butterflies
are to cars?
>The "tells" part will work both ways though. No one will be able to pick up
>on the computer's ear twitch when it has pocket Aces, but the computer
>wouldn't be able to pick up on the table talk that sometimes might be a
>giveaway.
In my experience, the biggest tell a player has is whether they
check/call/bet/raise. The rest is usually gravy.
- Andrew
Robert Ladd
"A. Prock" <jeffy...@yahoo.com> wrote in message
news:3cedd75b$0$24739$8026...@spool.cs.wisc.edu...
> According to Robert Ladd <rl...@cox.net>:
> 'cause artifical intelligencs is to math as butterflies
> are to cars?
The point wasn't comparing AI to math, but in using AI to ignore the math
when certain situations with certain players arise, or throwing in plays
that would normally be illogical, but because of gathered information would
be appropriate now.
Robert Ladd
Seth Peck
Actually, artificial intelligence, in its current form, involves an
incredible amount of discrete mathematics and symbolic logic. But
deciding which hand to play and which to fold, for a computer, is a simple
analysis of numbers. It's the human elements of the game that are harder
for a computer to analyze, and therefore either take time to compute or
are simply ignored.
Programs can be written that use methods of logic that we don't use. For
example, it is written in philosophical and mathematical literatures that
modus tollens ("If A, then B. Not B, therefore Not A") is not a method by
which humans naturally think.
Do you use heuristic search, set notation, idempotent laws, amortization,
probabilities when you play poker? Probably only the last one of these.
A computer program can use all of these while the cards are still being
dealt. You'd have to be a savant to be able to calculate as quickly as a
computer.
In addition, a program can remember more details about any given
situation than most humans. I could write a program that simply keeps
track of how long player A takes to decide whether to play or fold, and
in cases of playing, the average time waited on hands that were won, lost,
or taken down. Imagine tracking that information while playing against a
fish...you'd have a serious advantage if you knew that the average or
mode pause time for a winning hand was X seconds and the average or mode
pause time for a bluffing hand was Y seconds. Statistical data mining can
look for commonalities in play.
That last thing I mentioned--well, that's one of many areas where players
can take advantage of the programming. If someone varies their play
enough, the average time won't mean anything and the statistics will be
uninteresting enough for the program to have to ignore it.
Add to this that while winning is an optimal situation, there's no stress
to the computer about how much money it will actually win or lose upon the
outcome of the tournament. A computer doesn't have to worry about getting
tired, drunk, insulted or going on a raging tilt.
Paul: I'm not sure by your argument if you're saying that a program can
or can't be made to be among the best players. I'm saying that almost any
amount of programming, no matter how well-designed, can be exploited by an
excellent player.
However, I would love to start a programming project to prove either one
of these theories true or false! (I think this has already been done,
but I can't remember who did it) Let me know if you have any ideas or
plans...I'm not doing anything right now.
Seth Peck
unemployed computer science graduate
Paul Phillips
Robert Ladd <rl...@cox.net> wrote:
>The point wasn't comparing AI to math, but in using AI to ignore the math
>when certain situations with certain players arise, or throwing in plays
>that would normally be illogical, but because of gathered information would
>be appropriate now.
I think what he's telling you is, that IS math. Math is more than
pot odds and winning percentages. That "gathered information" is data
and the subsequent analysis and "illogical play" is driven by math.
--
Paul Phillips | Where there's smoke, there's mirrors!
Future Perfect |
Empiricist |
slap pi uphill! |----------* http://www.improving.org/paulp/ *----------
Paul Phillips
Seth Peck <pe...@CS.Arizona.EDU> wrote:
>Paul: I'm not sure by your argument if you're saying that a program can
>or can't be made to be among the best players.
You're not sure? I said I could write one to play as well as I -- there
must be an implication in there somewhere. I'm saying it can.
>I'm saying that almost any amount of programming, no matter how
>well-designed, can be exploited by an excellent player.
Can any human, no matter how excellent, be exploited by an excellent
player? I'm still waiting for the rundown on what it is you think that
humans do that the computer cannot do. There is an obvious list of
stuff that the computer can do that the humans cannot do (or at least
cannot do nearly as well.)
>However, I would love to start a programming project to prove either one
>of these theories true or false! (I think this has already been done,
>but I can't remember who did it) Let me know if you have any ideas or
>plans...I'm not doing anything right now.
A number of people have done some work on poker playing bots. Darse
probably has a good one cooked up.
http://www.cs.ualberta.ca/~darse/Papers/AAAI98.html
If you need motivation, it should go without saying that a capable poker
playing bot on the online poker sites could provide steady income.
>Seth Peck
>unemployed computer science graduate
Hey, me too, unless you can call poker a job.
Octo the Genarian
one could make a perfect (and therefore better than any human) computer
player. Its weakness would be that it would be unable to pick-up on tells,
but on the other hand:
(1) Perfect randomization, where appropriate, would destroy any
predictability.
(2) If one fed it hand histories it could analyze its opponents behavior for
trends so subtle that no human would pick up on them.
After all, if one can make a computer play world-class chess, poker should
be a breeze (especially limit and pot limit, where very subtle mathematical
calculations might be beneficial.)
This is kind of scary! Maybe someone should look into this bot idea. Never
work again...
"Paul Phillips" <rg...@improving.org> wrote in message
news:ackqv0$kto$1...@spoon.improving.org...
Robert Ladd
Ahhh, I missed his question mark at the end of that sentence.
Still, what I meant was that artificial intelligence is an application of
"logic" that can produce two different answers for the exact same question
(I did mean pot odds, implied odds and winning percentages when I said
"math", as you surmised) based on the "learned" information gathered before
the question is asked.
I don't necessarily believe that "logic" is "math", although the two are
intertwined quite often and both usually fall under the same learning
disciplines.
What I was thinking when I wrote my first post on this subject was that AI
would allow the program to change how it would respond to pot odds and such
because of other outside "learned" information, such as how this specific
player bets, raises etc., even if the exact same situation arose again.
In other words, the program could learn to play one player one way in a
situation, and another player differently in the same situation, or play the
same player differently if the same situation arose twice based on it's
learned information. It could learn to vary it's bet sizes, calls and folds
so that the human players wouldn't be able to get a line on it's behavior,
while at the same time gathering more and more information on the human
opponents so that it increases it's ability to read situations better.
And yes, I believe that if you were the person dumping your knowledge into
the design of the "logic" of a poker playing program, then the computer
should be a better poker player than you are, just because it can process
and hold onto many more pieces of information and balance them much faster
than you have the time or ability to do during the heat of a tournament.
Robert Ladd
"Paul Phillips" <rg...@improving.org> wrote in message
news:ackq6h$kl0$1...@spoon.improving.org...
Mr. V
the WSOP with a computer.
Mike, is that true?
This pot is raised.
Mr. V
"linuxandpoker" <linuxa...@yahoo.com> schrieb im Newsbeitrag
news:74990c89.02052...@posting.google.com...
Seth Peck
humans do that the computer cannot do.
Aside from interpretting tells?
A lot of what humans do is based on visual cues, whereas a program would
have a "state" of the game...i.e. data on what is going on mathematically
in the game, time statistics, etc. Can a computer program interpret a sad
expression on a person's face? A happy expression? let alone a blank
stare?
I think the big one is not so much what a computer program cannot do as
what it should be able to. For example, every once in a while even the
best players play a shit hand, like 2-9o. Would a computer program do
this without a distinct advantage?
I'll have to think some more...and try to differentiate between a good
program and a realistic program. A good program would win--a realistic
program would probably just irritate Phil Hellmuth.
Seth Peck
(has no good anagrams for his name)
AJohn808
T. Pascal
with their long development and programming cycles have reached an
above-average status, but I think they're topping out. Chess is
downright straight forward compared to Poker. And current poker
programs plain suck and I have never seen one I couldn't beat after a
while (I'm not even very good, quite frankly). I have never been able
to beat a moderately good chess program.
I do have another problem I didn't see raised -- what are the
logistics of entering the hands into the computer, raising, folding?
I don't think it will work.
Slightly off topic: At the satellites this year, a conversation
started on how blind people would be able to play poker. I think it's
related to the logistics problem. How do you solve that?
Treefrog
> This subject is intriguing. The more I think about it, the more I think that
> one could make a perfect (and therefore better than any human) computer
> player. Its weakness would be that it would be unable to pick-up on tells,
> but on the other hand:
This weakness is fairly minor, in most cases. It is that select few
cases where it is incredibly important though that might hurt the
"bot" in a NL tourney. In a low/mid limit ring game the ability to
read "tells" would have almost zero impact on the "bots" bottom line
(IMO).
>
> (1) Perfect randomization, where appropriate, would destroy any
> predictability.
>
> (2) If one fed it hand histories it could analyze its opponents behavior for
> trends so subtle that no human would pick up on them.
>
This is a horrendously difficult process to model this into a program
though. To make a simplistic example: What if the "bot" has played
20 hands with you total. It has seen you bluff raise twice. You make
a big raise at it. Random processor comes up with the fact that you
bluff 10% of the time and decides this is the 1 in 10 time it will
play back. "Bot" pushes all-in against you holding the nuts. "Bot"
says IGHN.
So you obviously need a more complex algorithm. But each decison
factor you add to the tree makes the program more complex and fragile.
You introduce more decisions and variable, you introduce more
potential for a devestating error in logic. I don't mean to imply it
cannot be done, but it is a very difficult process. For a limit game
these few error will not be devestating, but we are talking about a NL
tourney here.
> After all, if one can make a computer play world-class chess, poker should
> be a breeze (especially limit and pot limit, where very subtle mathematical
> calculations might be beneficial.)
>
No, no, no. Bad comparison...this comparison drives me crazy!
Chess = game of COMPLETE information. All the pieces and are right in
front of you
Poker = game of IMCOMPLETE information. Many cards araer hidden or
yet to be delt.
Computers EXCEL at situations with COMPLETE information. Computers,
to this point, are sucky in most IMCOMPLETE informational situations.
The technology is rapidly improving though. Could a "bot" win the
WSOP? Surely, but what the % likelyhood is would be a subject of
conjecture we will not settle.
> This is kind of scary! Maybe someone should look into this bot idea. Never
> work again...
>
The technology already exists to create a bot that would mop up most
low-limit ring games I believe. Consistently cashing in a NL tourney
is a bit off still I think. I doubt Paul could really write a bot
that would play as well as him in NL but limit poker would be a breeze
in comparison.
Treefrog
JonCooke
I think the computer would potentially be a big cash-game winner, but
although it could be programmed to perform better than average in
tournaments, I doubt it would ever be top-flight.
The great tournament players put themsleves out on a limb more than is
mathematically sound, but survive it becasue they can read when they
are being "rebluffed".
The computer could be programmed to be +EV for each decision, which
would be ideal playing cash, but couldn't multiply it's stack by 600
more than once in 200, say, as required in the big one.
A. Prock
>intertwined quite often and both usually fall under the same learning
>disciplines.
Logic is the purest form of math. When you break it all
down, math is just the logical game of:
axiom/theorem/proof
where the rules of proof are dictated by whatever brand
of logic you happen to be using.
It is certainly true that logic isn't calculus, but then
complex analysis isn't non-linear optimization either.
But they are all math.
- Andrew
Robert Ladd
Andrew,
Logic is the application of choices depending on the use of available
information. It doesn't necessarily have to be about mathematical
calculations. Of course computers use boolean logic, which is mathematics,
to take one logical path or another.
if player_count greater than OneMillionFiveLimit then
first_place_prize = 2000000
else
first_place_prize = 1500000
endif
if player_made_negative_comment_on_RGP then
remove_and_86
else
fleece_and_tolerate
endif
One is based on a mathematical bound being crossed, the other about a
questionable practice based on an emotional decision. True, both use
boolean logic to make their choice, and they have mathematical decision
making at their root, but the choice itself isn't always between two
mathematical answers.
Anyway, this discussion came up because I said:
>>"I think that a computer could be programmed with not only the lightning
quick math that would be necessary,
>>but could also be programmed with some artificial intelligence so as not
to always follow that math."
and you replied with:
>!?!
>'cause artifical intelligencs is to math as butterflies
>are to cars?
What I meant by "math" in this context was the pot odds and such that govern
a poker decision. AI could be used to evaluate a situation and ignore valid
pot odds because of other information it had gathered. I wasn't implying
that AI didn't use math as it's logical base.
Robert Ladd
"A. Prock" <jeffy...@yahoo.com> wrote in message
news:3cee6313$0$24737$8026...@spool.cs.wisc.edu...
Paul Phillips
Seth Peck <pe...@CS.Arizona.EDU> wrote:
>A lot of what humans do is based on visual cues, whereas a program would
>have a "state" of the game...i.e. data on what is going on mathematically
>in the game, time statistics, etc. Can a computer program interpret a sad
>expression on a person's face? A happy expression? let alone a blank
>stare?
Most people (myself included) will tell you that this stuff is of
incredibly small importance compared to everything else. As someone
else in this thread said, the biggest tell of all is whether they
check/call/bet/raise. And the inability to pick anything up is
counterbalanced by the inability to give anything away. Actually you
still can pick up how long it takes them to act, which over enough
time is probably worth a ton. The computer can always act instantly
or close enough.
>I think the big one is not so much what a computer program cannot do as
>what it should be able to. For example, every once in a while even the
>best players play a shit hand, like 2-9o. Would a computer program do
>this without a distinct advantage?
Obviously if there is some reason to do it, then yes, and no if no.
Since I know how effective it can be to go into "raise every hand until
someone plays back" mode at certain stages of long no limit tournaments,
I suspect that'd be in the computer's playbook too. But, in general, of
course the computer can do this -- the computer should be better at
playing 92o at the right frequency than the human is.
>I'll have to think some more...and try to differentiate between a good
>program and a realistic program. A good program would win--a realistic
>program would probably just irritate Phil Hellmuth.
I would quite enjoy having a computer bluff-raise phil and show him, so
we could see that tirade.
"These machines think they can run over me every hand... where do these
machines come from... mumble mumble"
>Seth Peck
>(has no good anagrams for his name)
Sure you do, you have a superhero anagram: you're the speck! SPOOOON!
Paul Phillips
they're all "Lost in Space" robots or something. As if computers can
only make short term logical decisions but can't take the long view like
a human can. IT'S SOFTWARE, PEOPLE! WE'RE IN CONTROL!
In article <53ebc436.02052...@posting.google.com>,
JonCooke <jco...@pmsi-consulting.com> wrote:
>The great tournament players put themsleves out on a limb more than is
>mathematically sound, but survive it becasue they can read when they
>are being "rebluffed".
All together now: what makes you think the computer can't do the same?
If you think the great players are all thriving on physical tells in these
spots I absolutely guarantee you that you're wrong. Computers have
essentially the same information available and can make the same kinds
of decisions.
--
Paul Phillips | Some people ask me, "Hey, Alex! You are crazy! Why should
Future Perfect | we build teleportation? Its so impossible!"
Empiricist | -- Alex Chiu <www.alexchiu.com>
A. Prock
>
>Logic is the application of choices depending on the use of available
>information. It doesn't necessarily have to be about mathematical
>calculations. Of course computers use boolean logic, which is mathematics,
>to take one logical path or another.
I'm so confused here. Are you saying that logic is math or
not?
If you want to compromise, I'll cede that mathematical logic
is a rigorous subfield of mathematcis, but that the general
study of logic has classically been studied in the domain of
philosophy.
I'm not quite sure what this has to do with AI though. But,
I'll also cede that AI isn't a subfield of mathematics if
that makes you happy.
Here's what Russel/Norvig has to say about mathematics w.r.t.
artificial intelligence:
"Mathematicians provided the tools to manipulate statements of
logical certainty as well as uncertain, probabilistic statements.
They also set the groundwork for reasoning about algorithms."
- Andrew
Roger McDow
>they're all "Lost in Space" robots or something. As if computers can
>only make short term logical decisions but can't take the long view like
>a human can. IT'S SOFTWARE, PEOPLE! WE'RE IN CONTROL!
>
Paul, as usual, is right on the mark about this subject. Just think if
you could write software, and it can be done, which incorporates all of the
qualities of the best players. Nobody has even mentioned that a computer is
immune to the 3 most harmful things that jeopardize a great players stack:
Fatigue, Tilt, and (S.L.O.P) Sudden lack of patience. I am sure that someone
could add these things into the program if they feel that they are needed to
win tournaments.
Roger
Dave
that will play "optimally". I mean "optimally" in the game-theory
sense where the best counter strategy is the strategy itself (although
there may be other strategies that are equally good).
A simple example of this is rock-paper-scissors (RPS). The optimal
strategy is to pick each action exactly 1/3 the time. It doesn't
matter how clever your opponent is, he cannot have any advantage over
you.
In a one-on-one setting this program would be king. The best that an
opponent could hope to do is play as well as the program.
The problem with "optimal" strategy is that it doesn't necessarily
exploit flaws in opponent strategy. In the RPS example a program
playing "optimal" strategy will merely tie (on average) a person who
always picks rock while a clever person will soundly trounce a person
who always picks rock.
Designing a program to play cleverly isn't quite as easy as designing
a program to play "optimally". Every year some group holds a RPS
tournament where programs play millions of rounds of RPG against each
other. About half of the contestants are entries and the other half
are ... suckers. Like they might pick rock half the time or they might
pick the thing that the opponent previously picked. The trick is to
come up with a program that will take advantage of the suckers but
can't be taken advantage of. The best programs have been ones that do
a good job of recognizing a sucker but when they detect they are being
taken advantage of they revert to "optimal" strategy.
I believe that we are a long way from creating a poker program that
can take advantage of suckers while not being taken advantage of. The
problem itself is more of an art than a science. And while a program
has tremendous number-crunching advantage a human will be able to pick
up on other things better than a program. If a "sucker" has a quirk
(tell) that the program can't profile (like he chirps when he gets
good cards) then the program will be at a disadvantage compared to the
other people playing against that sucker.
Dave
Lightning Roy
A lot of thank you type statements going out tonight...Hell, I'm starting
to agree with some of you people...
---Doug (Or, maybe you all are agreeing with me...Finally)
In article <aclv1j$232$2...@spoon.improving.org>, Paul Phillips
<rg...@improving.org> wrote:
> All together now: what makes you think the computer can't do the same?
> If you think the great players are all thriving on physical tells in these
> spots I absolutely guarantee you that you're wrong. Computers have
> essentially the same information available and can make the same kinds
> of decisions.
>
"We cannot all be made like me with lots of true-blue blending..."
(Lightning Roy Harper)
-----------------
macp...@pbsilink.com
macp...@inreach.com
Keith Ellul
poker. It's not clear that it does. In fact, my intuition tells me
that it probably doesn't.
Keith