Mission146
Posted by Mission146
Feb 26, 2017

Apparently, the fact that our bodies grow finitely and then begin to deteriorate leading to eventual death, except our noses and ears which apparently grow forever (they actually don’t, it’s gravity sagging the cartilage) isn’t enough to satisfy us as to our own basic fallibility.  Instead, we have felt it necessary in many tasks to come up with machines that can beat us at, well, pretty much everything.  When it comes to games, we have come up with machines that cannot be beaten at checkers (perfect strategy) and that apparently cannot be beaten at chess.  

However, we have always had poker. Poker has always shared with chess the fact that it is a thinking man’s game, but it has always seemed like the type of game that is actually deeper, in many ways.  While there is a calculable (though incredibly high) number of possible chess games that could take place, given the combinations of cards and seeming randomness of the human mind, how could we possibly put a number of the amount of possible poker sessions that could theoretically take place?  I don’t know that the computers have done that, but at least so far, they have found a way to beat us.  

Poker has a psychological component that a computer simply shouldn’t be able to overcome, because as well as a computer can think, it just seems unnatural for it to be able to think more uniquely than a human.  Psychology is all about flaws, but in those flaws hide randomness, whereas computers should largely be about order and procedure...how could they do that and still manage to crack our Psychology?  

For one example of what I mean, let’s take a look at bluffing, shoving all-in with absolute garbage, just as one example.  You push all in, there’s no chance for a straight or flush on the board of any kind, you’ve just got Trips, Two Pairs and Pairs as the only possibilities.  One man sees something in the guy who is bluffing, a twitch in the eye, a slight curling of the lip, a false over-eagerness to push all of his chips into the center of the felt...he calls. The bluffer sighs inwardly as he turns his rags over whereas the guy who called tosses only a jack, the only card he actually needs to beat the bluffer, he called despite the king on the board.  

“That was a sick call, man, good hand,” the bluffer admits.  

“I didn’t know if I was right,” the caller says genuinely.  

How can a computer crack that?  

I understand how a computer can crack chess, you can’t, ‘Bluff,’ moving into a sure one move checkmate.  You either move into a sure checkmate or you don’t.  Furthermore, if you have a computer capable of analyzing the several thousand most likely combinations of moves all the way to the endgame, given the current position, you can’t bluff being in a worse spot than the computer.  For two master chess players, it is very difficult to play the board equivalent of, ‘Rope-A-Dope,’ that you will sometimes see between two amateurs of only limited skill.  (Trust me, I’ve watched enough good players and played enough games as an amateur of limited---very little, if I’m honest---skill to know!)  

In the sport of Chess, you can’t bluff having a superior position when your position is inferior, particularly not when your position is far inferior.  In poker, however, you can.  That’s where we would expect the man to have the advantage over the machine.  A man who is essentially, ‘Drawing dead,’ or already dead in the water after the River (fifth community card in Texas Hold ‘Em and some other games) may not always be able to pull out the actual win, but sometimes he can.  He can win even though he has lost.  Chess players don’t have that luxury, unless they are playing an inferior opponent, and if that is the case, the opponent isn’t all that likely to get to the point where he had, ‘The best of it,’ to begin with.  

Another major difference between Poker and Chess is that Poker is won in the trenches, where it is largely fought.  When you have two players of vastly different skill levels in chess, it’s kind of like one country armed with nukes when the best the other country has is a fleet of tanks that would have already been decommissioned by the first country.  If I were to play someone such as Magnus Carlsen in a game of chess, I lose.  That’s all there is to it.  However, I could theoretically sit down and beat Phil Ivey heads up in No Limit Texas Hold ‘Em.  I’m not saying I would be likely to, in fact, he’d probably demolish me, I’m just saying I could.  

Inferior players often get to win the occasional battle in poker, when it comes to an individual session, and even moreso, an individual hand.  Imagine a superior player ends up with two clubs in his hole cards (make it J-10) and the flop comes Ac-4c-8c and I have an Ad and a Kc: If that player, through skill, psychology and subterfuge manages to get me to push all in with my pair of Aces against the Flush he already has, then he has a 68.38% chance of taking all my money!  The opponent has the superior hand and he makes (in most cases) the superior play, granted, if I am really low-stacked in that situation it could theoretically be the best play for me to make even if I know he has the flush, but usually, he outplayed me.  

The next card that comes out is a 3c, he has no potential for a straight flush.  He was the better player, he had the better hand, he made the better play.  Thing is, I outdrew him, and you can’t teach that. It either happens or it doesn’t.

It is for precisely that reason that an inferior poker player can win an individual session against a player way beyond his/her skill level just because of plain, good-old-fashioned dumb luck.  (Before any of my fellow brothers-in-math say anything, no, I’m not advocating for the existence of, ‘Luck,’ yes, I understand if that situation happens 10,000 times, the player with the worst of it wins 3,162 times.)  

Anything can happen, until now…

We brought four of our best, maybe not our four absolute best, but undoubtedly four of our best poker players to represent humanity in its battle against artificial supremacy...but supremacy nonetheless.  These four men fought valiantly, but in the end were proven to be nothing more than men, men of fallible flesh and bone with fallible minds and fallible thoughts. We tried, we failed.  

Fighting out of the Blue Corner, a dispassionate artificially intelligent program with absolutely no pride (or anything else) riding on the outcome, we have The Challenger, Liberatus.  Liberatus is a computer program designed by Carnegie Mellon University with the cold and unfeeling objective of continuously improving itself at poker through the use of data and developing its own strategy to defeat even the best of human competitors.  

Fortunately, for humanity, this is a handicap match.

Out of the Red Corner, we have Dong Kim, Jason Les, Jimmy Chou and Daniel McAulay.  

Dong Kim is known as one of the world’s foremost heads-up specialists, with a focus on Heads Up No Limit Texas Hold ‘Em.  The man known as, “Donger Kim,” is mainly known as one of the best online Heads Up players and has only limited exposure in the live realm, only true poker enthusiasts (or those familiar with this Rivers event) will have heard of him.  

Jason Les is also a Heads Up No Limit Hold ‘Em specialist and a name that more casual fans might be familiar with.  He finished 40th in the 2016 NLHE World Series of Poker Main Event, though his biggest cash came in December of 2014 which saw him placing fifth of fifty-five in the $100,000 NLHE Alpha8 tournament....good enough for a $431,640 payday.  That tournament was held at the Bellagio.  Les has finished with a six-figure payday in several other tournaments and has amassed a total of just over 1.5 million dollars in lifetime tournament wins.  

Jimmy Chou is also a Heads-Up NLHE specialist considered one of the best in the game whether that game be sit-and-go or a cash game.  Though the Rivers site would have him as, ‘Jimmy Chou,’ he is listed as, ‘Jimmy Zhou,’ on the Hendon Poker Mob Database and came in first place in the APPT 2015 Asia Championship of Poker, held in Macau.  That was good for over three-quarters of a million (after conversion to USD) and he defended that title respectably the next year finishing in fourth for a payday of just over a quarter-million.  (Again, USD)  

Finally, Daniel McAulay has a game tailored exclusively to specialize in Heads-Up NLHE whose main claim to fame is the cash game in which he specializes and has, reportedly, never refused a challenge.  Though he placed 7th in the 2016 $1,500 NLHE Shootout, his tournament history is somewhat shorter than the others as he really focuses on the heads-up cash game.  

These fine competitors who represent certainly among the best of what humanity has to offer against a machine were absolutely demolished.  In mathematical terms, the proper phrasing of the utter thrashing that these players felt is, ‘Statistically significant.”

Libratus needed to beat the humans (who won by an almost statistically significant margin against the AI less than two years ago) by $7.70 per hand for the result to be considered, ‘Statistically Significant.”  To put that in perspective, the machine was not playing 5,000 hands against the humans, but it needed to win that much per hand over 120,000 hands in order for the win to be considered statistically significant.  

Unfortunately, the humans could not even come close to providing that level of challenge for Libratus who steamrolled them in the early days of the twenty day challenge and never looked back.  The machine won $1,766,250 more than its human counterparts which was good for about $14.70 per hand, almost double the result that would have been statistically significant.  Dong Kim was the best of the four we had to offer followed closely by McAulay.  

While the event was followed closely by the poker faithful and those keen on artificial intelligence alike, there can be no question that those who are keen on the AI enjoyed it significantly more as it represented an almost undoubtable victory.  If those four men aren’t the absolute best of what humanity has in heads up poker, they are certainly close, and Libratus dominated by such a wide margin of the bare minimum to be statistically meaningful that it should be pretty safe to say no human can beat it.  

Granted, that may not spell the end of poker as we know it, but it spells the end of human dominance in poker.  If a computer program could be programmed to work quickly enough, after being fed data, to spit out an optimal play and even have a fraction of Liberatus’ ability, that program will dominate most human players pretty decisively.  It is almost beyond doubt that several people are at work on such a program right now.  

While I don’t have that much faith in the general honesty of humanity, hopefully nobody turns such a program lose on actual humans...though I’d be surprised if that doesn’t ever happen.  

Maybe it would be fun to have some sort of, ‘Meta-Poker,’ in which individual people attempt to program a poker-learning piece of software and then they have each of their respective programs compete against each other for cash.  You could have different poker tiers, human poker, and AI poker.  Humans can continue to play humans and AI can continue to play AI.  You could even have sites (or existing sites) that house AI vs. AI tables!

Or not.  More likely people will develop AI to beat other people, instead.  It seems that there would be much less risk that way.  Besides, while one piece of AI would generally be demolished by a much superior piece of AI, it stands to reason that a not terribly good (comparably) piece of AI would wipe the floor with an average human.  Basically printing money against any human being foolish enough to go against it.  

At that point, the poker sites would possibly have to begin taking countermeasures, so this opens a whole new can of worms!  The sites, in order to ensure fairness, might have to actually develop (or acquire the rights) to some seriously powerful poker AI to analyze the play of humans in order to ensure that the humans aren’t playing impossibly well.  

What a mess!

Either way, we’ll always be able to have fun and at least a fighting chance against one another with a physical table and physical chips.  Just keep an eye on the guy that seems to be checking his phone a little too often.  

Comments

BralasLT
BralasLT Feb 26, 2017

The speed at which AI is learning is quite incredible, it's actually a bit scary, because it could easily escape our control. Many have heard about the Deep Blue, unbeatable in chess. But DQN is also already well-known - it can learn to play different video games from scratch. It takes only a few hours to completely masters the games (it starts as a newborn, literally not knowing what it needs to do to play.)

But there's a limit for now and mostly because we don't know pretty much anything about the consciousness. That part is impossible to program. And dangerous part actually is that if we manage to create AI with a general intelligence how it is going to function without conscience. We will be lucky if it just beat us in poker. :)

onenickelmiracle
onenickelmiracle Feb 26, 2017

It's possible our own intelligence is artificial.Depends mainly if the universe is artificial.

Mission146
Mission146 Feb 27, 2017

BralasLT,

That's pretty incredible about the video game mastery, how long does it usually take it to best humans in those games for which world records are kept? Can it usually do it in under a day?

Yeah, the ultimate AI may just decide we are inefficient creatures and the planet would be better off without us! We'll be in trouble when we can develop AI that can uniquely program its own AI, I should imagine.

BralasLT
BralasLT Mar 08, 2017

Mission146,

It is incredible. Those systems start with only raw pixels and score as inputs, and then they are able to continually adapt its behavior, and develop strategies, without any human intervention. AI was tested on 49 different game scenarios, and it suppressed professional human reference every single time. There's a video on youtube "Mnih Supplementary Data video 2 R3" showing how AI learns to play a game (old Atari game).

There are some conflicted data online, who can master the game faster, but I guess it's interesting that when AI masters a game, it's better than any human.

Please login or register in order to leave a comment


Sorry, an internal error has occurred. The error has been logged and will be reviewed.

In the meantime, please try your request again if you haven't already.

If the problem occurs more than once, rest assured that it will be investigated and fixed soon.