PLO cap bot would be perfect for degens. Basically a coinflip game.

I received an email back from Professor Eadington who has brought the machine to the attention of his class as well as some students who work for IGT as engineers and mathematicians and this is some of what they have said. He himself hasnt commented on it yet but im sure he will give his insight about it at some point too. I have a few more questions i am going to pose to him and have him pass along to the IGT kids so if you have anything you want to know leave a question. Might be nice to hear what the IGT employees response is if you have a particular question. The email i got from him is copied and pasted below.

FROM AN IGT ENGINEER

I've heard this a lot recently ... and have especially been keeping up with this forum.

The Artificial Intelligence embedded in the game does not learn ... but it is extremely complex. To the best of my knowledge, and if you had a computer by your side... and knew the exact learned behavior in the game... and made absolutely no mistake ... you might break even ... but a human opponent could not keep that up.

We have invited some of the best players to try before we released the game ... maybe some of your students could try to make a living off of the machine?

Sounds like a good discussion ...

FROM A MATHEMATICIAN


I don't see why they rush to say it is automatically beatable. For a game in which one player lacks the psychological element, is there not an actual perfect strategy?


A chess program uses a fixed, non-learning strategy, and it is not easily beatable. And just because you KNOW what your opponent will do, that still doesn't mean you can automatically win.

And while it is not impossible for IGT to screw up, it seems unlikely, and his "I'm winning, that's proof" does not assuage my doubts. At best, this sounds like an even game, which is subsidized by the large schools of fish.

I immediately imagined that the poker set strategy would also be a mixed strategy, apparently adjusted for each player decision within a single hand, but then forgotten. Such that if it came down to the perfect strategy being bluffing:calling:folding 83:22:38, a player, even having seen many many hands, would still be puzzled as to whether The Brain is going to bluff, call or fold, since he seems to have done all of them at different points. In other words, the perfect strategy is not a pure strategy (like "given hand strength rank 12, and player bx of 1check AND 2raise, decision=FOLD"), which might eventually be deducible, but a pseudo-adaptive mixed strategy that will at least keep the player wondering.


So it seems more likely that The Brain follows not a simple linear protocol, but a slightly recursive game theoretic model.


Seeing as it still is an etched-in-silicon computer program, the player could still get some insight into the opponent, but it would take so long for these wise guys to figure it out (without hacking into IGT's warehouse) that indeed all the fish--and wise guys--will more than make it worth IGT's while to have built the machine.

Does anyone know if these are being rolled out to other casinos? Or just Bellagio at this time?

[QUOTE=Poker Clif;22242516] You're missing the point. If the machine is beatable at all, you would have to be very, very good at HULHE to beat it. Anyone who could beat it could make a lot more money playing some other form of poker.

The machine can't be beatable in the long term. If it was then it wouldn't be out on the casino floor. If it is beatable in the long term then it will soon be gone.

This quote should be a good life lesson for you young fellas out there. Matter of fact, non emotional, serious pronouncements like this give the aura of being very likely to be true. But in this case he say two things that are both 100% certain to be wrong. The game must be "beatable" if it isn't cheating. And there are tens of thousands of beatable machines that are knowlingly put on casino floors.

To expand a bit on the last two posts, the "machine must be unbeatable otherwise the casino wouldn't use it" seems to be another commonly held misconception (although this one has nothing to do with game theory.)

The machine does NOT have to be unbeatable. It simply needs to beat the population of players it actually plays against (more accurately it needs to beat this population by enough to cover the opportunity cost of having it there.) In practice, this means that it only needs to be good enough that almost everyone who can beat it has something better to do with their time. If you need to be a very good player to beat the machine, you can make more money by going to a regular poker game and playing that, particularly because the machine only offers small stakes. Again, I don't know whether or not the machine is that good-only that it doesn't need to actually be unbeatable to make money.

blackjack is vastly more beatable than this thing possibly could be, enough so that there are professional black jack players, and it's not going anywhere

sportsbooks are also beatable

Well, that's one of the issues we're discussing here. I don't think it's crystal-clear quite yet.

However, there may be some people who are unable or unwilling to do so, or simply like the idea of taking money from the bot. It wouldn't take many of those sorts of people to tie up the machines.

Doesn't there exist the possibility of somebody becoming very good at beating this particular bot and bots in general? The strategy used to beat the bot wouldn't necessarily be a good practical strategy against a human so it is possible that a person that can beat the machine might not be able to beat a human for that much more money.

Any word of this machine being available at any online casinos?

fyp

Why the Singularity Isn't Going to Happen.

Have you taken a close look at the rules of blackjack games in Strip casinos lately?

Oh, an opinion-based blog! Well in the face of such overwhelming evidence, I'm revising my stance.

Touche.

Tens of thousands yeah? I know vegas is hurting but c'mon..lol

Off-topic, but that article was total crap.

If the writer is responding to others spouting crap, who cares. But she addresses none of the actual issues surrounding the Singularity.

You realize don't you that I was talking about poker machines where perfect play plus cashing in points is worth a few bucks an hour to experts.

No more points at the Bel for these games. The machines has a sign on them yesterday.

def interesting that they are working on NLHE machines

hmmm, sportsbooks are beatable because humans set the odds, whilst blackjack continues to be tweaked so its not beatable.

I don't think this is correct. PSPACE, EXPTIME, EXSPACE are all likely harder than NP complete problems. And there are also NP hard problems that are probably harder than NP complete. (And ones that already have been proved to be harder, like the halting problem, though obv that was proved undecidable.) Not an expert on this but that is my understanding. I'm also not sure if poker if HU limit is NP complete, I know some Nash equilibria are, but don't know the restrictions and haven't gone through the proofs. If you have a source handy for that please post it.

HU LHE is not NP complete. It is solvable using linear programming.

I might be confused but I thought set packing was NP complete and solvable with linear programming?

HULHE is just one question. One question can't be NP-complete. NP-complete problems are classes of questions with a number associated with them.

For example, the traveling salesman with n cities is NP-complete. So, if you write some bad ass computer program that, when given an arbitrary set of cities with distances between them, it is conjectured that this program's run-time will get much much higher as its run time increases and that this increase will be faster than the increase of any polynomial. For the traveling salesman problem, I believe it's conjectured that this function is 2^n plus some terms that are essentially insignificant.

However, if you give me a specific question that is a member of the set of traveling salesman questions like "What's the fastest route between the 100 largest cities on earth?", I can theoretically write you a program that gives the solution in 100 steps. First, I simply solve the problem the long way (which would take something like 10^30th steps ). Then, I write a computer program that prints the solution. (In fact, if you let me choose as my "character set" the set of routes between the cities, I can write a program that gives a solution in one step.)

Similarly, I can in theory write a program that "solves" HULHE in under 10^15 steps, which I guess would take my laptop a few days. All it needs to do is print out the frequency of plays in every situation. In practice, there's probably a solution that's a lot simpler than just a list of correct plays, so I'd be willing to bet that there exists a program that "solves" HULHE that my laptop could run in a day. Of course, the problem is that in order to write a program like that, you'd have to know the solution first . But, yeah, the term NP-Complete is meaningless in this context.