Has 'Perfect' Poker Bot Cepheus Really Solved Poker? Not Even Close

cepheus
Researchers at U of A.

Computer scientists at the University of Alberta have just finished a ten-year long project to build poker software that can beat every human player on the planet.

The program is called Cepheus and it plays "perfect" Heads-Up Limit Hold'em poker.

Mainstream media outlets have jumped on the idea that poker has been "solved" and we've entered the age of the machines.

As poker players, we're scewed, right?

Well, not exactly. Not even, really, all that much.

What is Cepheus?

There have been a lot of attempts to create poker software equal to virtual chess players. Since former chess world champion Garri Kasparov lost to Deep Blue, we know that machines are ultimately the better players.

In poker, many have dreaded the moment someone would come up with similarly superior software. Is their nightmare now coming true?

deepblue
Is Cepheus poker's Deep Blue? Not really.

Project leader Mike Howling and his team have used adaptive software. They saved every single possible situation of a poker game – in a heads-up game, that’s 316.000.000.000.000.000 different situations (!) -  and then had the machine play against a copy of itself.

The goal: beat Heads-Up Limit Hold’em. 200 processors took 70 days of calculations and generated a data base of 11 terabytes ... and then poker was solved!

The machine went from making completely arbitrary decisions to playing reasonably and then, eventually, perfect. It's even capable of bluffing.

The reason why it's mimicking human play so well is because it’s using a random generator for certain situations.

As you know, in poker, there is often no single right answer. Instead, theory will tell you that in a specific situation you should call 70% of the times and fold in 30%, for example.

Cepheus doesn’t make a decision here; its action is determined by the random generator. That makes the software as unpredictable as a human player.

Of course, Cepheus doesn’t win all the time. It will get bad beats just as anyone else. But in the long run, nobody should be able to beat it.

In fact, nobody will.

Strengths and Weaknesses of Cepheus

David Sklansky
Sklansky: I'll win faster and bigger

Just like Deep Blue, Cepheus is way superior to the ordinary human mind when it comes to calculations.

Computer software can go through a myriad of situations at ridiculous speed. Did you see the number above? That’s 316 quadrillion!

However, that’s also where one of the software’s weaknesses lies. Humans can rule out the vast majority of moves as they are obviously wrong.

In chess, this is easier to picture. For example, you wouldn’t move your king early in the game unless it becomes absolutely necessary. So the software must go through every possible move before it decides.

Also, although Cepheus is unbeatable, it’s not the best player in the world. Sounds paradoxical? It’s not.

As poker guru David Sklansky said in a recent response to the unveiling of Cepheus: “If the computer is playing a bad player, it will win, but it won’t win as quickly as a human being playing a bad player.”

“I will destroy that beginner to a greater degree than this computer program will.”

In other words, while it’s an advantage for Cepheus to be unemotional and insensitive to tilt, it’s also lacking the killer instinct.

Yet, if Sklansky would play Cepheus, he would get beat by the machine. It just might take a while.

By the way, you can test the program yourself. The software is online and you can play Cepheus here.

Gregory Merson vs. Keith Lehr
Cepheus still needs someone in the other chair.

Is This the End of Poker?

No. And the explanation for this is surprisingly simple.

Even if a computer program can now play better than any human, you have to play it to lose.

How many guys do you know who play Heads-Up Limit Hold’em? Does anybody? I checked the lobby of market leader PokerStars.

There were 58 open heads-up LHE tables at $1/$2 or higher. At 57 of them, some shark (I guess) was waiting for a victim.

Only one table was actually active.

Also, Cepheus is useless as soon as there are more than two players at the table. The interaction is too much for the program.

So, Cepheus is not going to have a large effect on the poker world.

So What’s it Good for Then?

The most amazing thing about Cepheus is that it can beat every player’s strategy. In fact, it’s not even trying to find out what these strategies are. It doesn’t care.

PokerKitchen 5
Other applications for Cepheus maybe more intriguing.

This is going to be exploited in a lot of different ways. Cepheus is going to be used in different fields – medicine, security, the fight against terrorism.

In Alberta, they are already working on software that helps to develop individual plans to manage diabetes. Sounds far-fetched, but Cepheus’ is going to be of great help there because of its capability to make good recommendations.

At the University of California, scientists are building “a system for deploying air marshals on flights and coastguard patrols at ports.

In the same way that Cepheus found a way of not being beaten at poker, the USC project is aimed at devising policies that cannot be exploited by any adversaries," writes The Independent.

So, Cepheus is indeed a milestone with regards to game theory software, but it’s not going to destroy anything in poker.

Keep on playing. Just stay away from the $1/$2 LHE HU tables, because me and my new friend Ceph might be sitting there. Waiting.

Please fill the required fields correctly!

Error saving comment!

You need to wait 3 minutes before posting another comment.

Daniel 2017-01-10 07:13:12

I'm sure I'd have to play for even longer, but in close to 400 hands I played against Cepheus, I basically quadrupled my play money. The patterns seem really predictable. Maybe in the long long run it would win--I can accept that data without first hand evidence, I suppose--but it's hardly a brutal opponent.

Carl 2016-06-03 10:16:41

I have read that Cepheus has one flaw and that it can't adapt and that is sort of everything in poker. This was posted in TheGuardian website. I think to create a great poker bot you don't need to be a computer scientist but more a good poker player. I was able to create a poker bot that most of the time wins and it also 'adapts' unlike Cepheus and I was able to do this by using a simple tool like abz poker bot. My point is that it's all about being a good poker player and not a scientist.

Eric 2015-11-08 02:33:38

I was just losing to this bot after maybe a dozen or so hands, but I think I got the hang of it and started over. I just beat this program after 100 hands. The problem is it doesn't ever seem to bluff, and is very predictable. So when you have good hand, you lead it into thinking it has you beat and it may just keep betting things up, even though you've probably won. And if the bot shows weakness, you just bluff and it typically folds. It makes me feel like I'm really good at poker, but I know I'm not. I think this bot is just bad and has in no way "solved" poker. Also, while you can read the bot because it's very predictable, as far as I can tell it can never "read" you or adjust its style of play based on your style of play.

Silvio Peters 2015-09-25 11:49:04

I played 1 time against him and got $540 up. He is not a good player.

Jachryl 2015-04-23 16:29:14

Just remembered this program was still around and I took a run at it. First time playing it I won.

Bogdan 2015-03-10 11:46:22

I join the list. Just won about $300. Why is this such a good player? Very logical, very mathematical, but why unbeatable?? Over what? 1.000.000 hands? Probably. At the rate of the hands are dealt, yes, you can't compete with a machine :) Otherwise hats off for the programming. Hopefully it will be useful in other domains, because I see no future in this field.

Matt 2015-02-02 01:45:58

I've beaten this program three times in a row now. The first time wasn't by much - the spread was only $80 (I was up $40 and Cepheus was down 40. Each game is 100 hands). The second game the spread was $1320 and the third game it was $500. Needless to say, I wasn't impressed by its play, and I would say that for a decent player it is actually pretty easy to figure out. For the record, I'm a decent poker player, but not a professional and actually don't consider myself that good at heads-up poker (or limit poker, for that matter). I think the folks at the University of Alberta need to reevaluate the claims they are making for this program.

Carbon Cycle 2015-01-26 15:58:35

Cepheus is not unbeatable, as I have already beaten it twice out of five 100 hand matches. It may beat most on-line challengers because the web interface is horrible and that beats down a human's emotions. Worse, the queue to play is 50 users long, but only two can play at once. It used to be 4 players, but I noticed the server kept crashing and you would get summarily tossed out of your match. So, you end up waiting as long as two hours to get into a match. There is no indication of where you are in the queue, so you can easily get distracted and miss your 1-minute window to start playing, hours after you have finally gotten into the queue. It might help to pass the time, if you could see how the current players were doing. But no, this is all about Cepheus. Still, if you can hang in there it is not a perfect poker player, and you can learn how to beat it fairly rapidly. Cepheus still has to get cards to win, and that remains a 50-50 proposition. Don't be intimidated by its betting strategy. Often, it doesn't have good cards, but keeps betting into them. You can bluff it and slow play it effectively, lose small and win big, just like playing a real person.

Best Poker Sites - Editor`s Pick