New AI contest: suggestions?

[Sang-drax]

Darryl, are you saying that your strategy is unbeatable? Either you're wrong, or you're going to be very rich playing no limit Hold'em tournaments.

Hold'em would be nice, but I agree with PJ that it'll probably yield too few entries. Everone (and I mean everyone who has ever touched a C++ compiler) could write a program playing Black Jack.

Is this page a good page for Black Jack rules?
http://www.pipbet.com/rules/black-jack-rules.html
I've never heard of surrendering before.

[PJYelton]

Yeah, I've heard of surrendering but I've never been to a casino that allowed it.

The most common rules are:
Double Down on any including splits
Multiple Splits allowed
Dealer stops on all 17's, including soft 17's
No surrenders
Insurance allowed

I would use these rules, although an interesting option would be to play multiple sessions with the rules, betting limits, and deck quantities changing.

[MadCow257]

if no one (no bot?) will call then there's the winning strategy

With 8+ people on a table, someone is going to be confident enough to call you within a round or two and you'll get hit major.

EDIT
These are the standard rules I think: http://www.blackjackinfo.com/blackjack-rules.php
Adding variations would be interesting though.

[Darryl]

With 8+ people on a table, someone is going to be confident enough to call you within a round or two and you'll get hit major.

EDIT
These are the standard rules I think: http://www.blackjackinfo.com/blackjack-rules.php
Adding variations would be interesting though.

With 8 people, yes!, with 8 bots, I don't know... where is that confidence going to come from?

[PJYelton]

Isn't the whole point to create an AI that is capable of making such decisions? I'd love to have my bot go against another bot that just made bets every hand.

[Darryl]

Darryl, are you saying that your strategy is unbeatable? Either you're wrong, or you're going to be very rich playing no limit Hold'em tournaments.

Hold'em would be nice, but I agree with PJ that it'll probably yield too few entries. Everone (and I mean everyone who has ever touched a C++ compiler) could write a program playing Black Jack.

Is this page a good page for Black Jack rules?
http://www.pipbet.com/rules/black-jack-rules.html
I've never heard of surrendering before.

I do not have a unbeatable strategy against people so I won't be getting rich... However, I think I do have an unbeatable strategy against computers, though, whether I could successfully recreate that in an AI may be questionable and of course there is always a bit of luck involved.

Why do you think it would yield to few entries? They wouldn't be writing the whole game, they would just derive their AI player from some base player class provided. That way they only concentrate on the AI. I have a poker framework I created that can be used to easily create different variations of poker. I would be more than willing to provide it for this contest. It has 5-card draw and stud poker, but it wouldn't be too difficult to derive a hold'em variation for it.

[Darryl]

Isn't the whole point to create an AI that is capable of making such decisions? I'd love to have my bot go against another bot that just made bets every hand.

I was being overly simple when I stated that betting high every hand to force everyone to fold was the winning strategy, but instead I would suggest it more of a general strategy. I think unless a lot of detail went into the AI, it would be too mechanical/predictable... against other bots, that's ok because their AI wouldn't be much better, but against a human for example, they would get blown away, because the human would catch on to what they need to force the bot to fold often after driving up the pot.

[PJYelton]

Sorry, I do agree that a decent human player can usually beat a number of computer players, I thought you were implying that it would be easy to create an AI to accomplish this. However usually the way to do it isn't betting high on every hand but instead to play tight since everytime I've seen a bot play they tend to be very loose and willing to call with just about anything.

I think there will be few entries because it would be a difficult AI to write for and in all honesty few people here would put much effort into something that will take more than a day to write. The connect four contest we had was much simpler than poker and even that we only got four entries. Plus I think it would be a pain to write a decent AI since there are so many variables involved in making a decent decision, and that doesn't include whatever AI you have set up to analyze the other players. I envision huge amounts of if statements one after the other that look like this:

if(IHaveTopPair && NoFlushAvailable && (OnlyThreeCardsToStraight || 
  FourCardsToStraightButMissingCardUnlikelyInHisHand) 
  && (NoPairOnBoard || PairButUnlikelyHeHoldsTrips) &&   HeChecked && 
  HesTightPlayer && IHaveMoreThanTwiceHisMoney)
     bet(myMoney/2);
else if .....

Of course you could simplify it to only a couple of scenarios, but then your AI would be extremely predictable. I just don't see a lot of people putting much effort into this one, and a hold'em tournament with four or less people just wouldn't be much fun.

[joshdick]

Speaking of simplicity, how about programming an AI for the Prisoners' Dilemma?

[Sang-drax]

Speaking of simplicity, how about programming an AI for the Prisoners' Dilemma?

Heh

int getDescision()
{
  return SQUEAL;
}

Why do you think it would yield to few entries?

Because the connect four tournament had four participants. If enough people wants Hold'em, then it'll be Hold'em. Perhaps we should create a poll in a while.

[Darryl]

Over at CPP-Home, we had a Clue AI contest, which required a pretty involved AI and we have quite a decent showing. I don't think it's the difficulty of programming that holds people back but the lack of interest or challenge. To me doing a min-max connect four AI would be quite boring and if implemented correctly end-up being futile! (though I am not sure, just guessing that correct play always leads to a draw?)

However I don't think AI poker is that interesting either, this whole discussion started by me commenting that Poker is no fun when you don't play for money, because people aren't afraid to call your bluff if they have nothing to lose. Which then followed they'd be afraid to lose the tournament, but I still say that wouldn't be enough of a deterrent as was stated, AI poker usually plays loose.

[Sang-drax]

I don't think it's the difficulty of programming that holds people back but the lack of interest or challenge.

No, I think it's the lack of time.
Out of the persons who joined the connect four contest, only about 50% actually submitted anything.

though I am not sure, just guessing that correct play always leads to a draw?)

Actually, it has been proven that the person making the first move can always win.

[PJYelton]

Heh

int getDescision()
{
  return SQUEAL;
}

LOL

Actually, it has been proven that the person making the first move can always win.

True, although I don't think an AI has been written that can accomplish this yet but I could be wrong.

I agree with Sang-drax, people love a challenge, just not one that takes more than a few hours

[Darryl]

No, I think it's the lack of time.
Out of the persons who joined the connect four contest, only about 50% actually submitted anything.

I disagree, while time may be a factor for some, there are lots of programmers with plenty of time. Look at the popularity of POTM contest (which are quite involved) or mathschallenge.net - eulur's challenge, if programmers can find time for those then surely they would for an "appealing" contest here.

Also a bit of advertising wouldn't hurt... go to other programming forums like cpp-home, cppworld, etc. and announce the contest.

[pianorain]

there are lots of programmers with plenty of time

With jobs?

I agree with Sang-drax. If I had time, I'd certainly participate in more contests. It's more interesting than doing other people's homework.