Where?

http://monkerware.com/solver.html

Identification technology is the only way to save online poker, thats is, making sure it's a person clicking the buttons.

It's possible, there just needs to be a shift away from poker sites relying on bot rake.

It's not enough to verify that a human is clicking the buttons. You also need to verify that the human isn't being told which buttons to click. Seems very difficult to do. I think online poker is ultimately doomed.

chess complexity vs poker complexity?

Brown and Sandholm state in their paper that hidden info games (such as poker) are far more complex.

"Hidden information makes a game far more complex for a number of reasons. Rather than simply search for an optimal sequence of actions, an AI for imperfect-information games must determine how to balance actions appropriately, so that the opponent never finds out too much about the private information the AI has. For example, bluffing is a necessary feature in any competitive poker strategy, but bluffing all the time would be a bad strategy. In other words, the value of an action depends on the probability it is played."

Online poker is rip. You're a fool if you think it can be saved. This has all already been played out in the backgammon world and the results are in, nobody plays for real money online.

These AI *******s are doing everything in their power to destroy my livelihood and the livelihood of other poker/strategy game players.

The poker players who willingly participate in AI matches are poker Uncle Toms.

As wolfy said this has already happened in backgammon. And in chess. There's essentially zero online gambling on these games.

These AI guys are the enemy. As I've said before, they could quite easily invent a game in which the rules could be as simple or complex as they'd need it to be. However, this will never happen, because fewer people would be intere$ted in that, (even though it would have the same, if not better, scientific value) and said AI comp sci. traitors to humanity will make less money.

Checkers got solved in 2006-7. Googling hasn't been a great help in determining how much the popularity of that game has decreased, but it seems eminently logical that some new people who might want to learn that game see that's it's solved and say, "oh, there's nowhere else to go, nevermind".

Congratulations to Professor Douchebag and the other guy for being footsoldiers in the war to kill all strategy games.

May as well ask here: In what sense are the Monkersolver guys claiming to provide 'solutions' to multiplayer spots?

To the doomsayers: Is it really so unreasonable to talk about banning those who play 'too perfect'? Yes I'm aware that mixed strategies muddy the water; but on the other hand, just like the strategy-enacters, the rule-enforcers have a large sample over which to collate evidence (which sounds difficult but should be automate-able for the most part). I would think a sort of reverse-engineering approach would be best, i.e. first identify players who never make a 0%-freq-gto play. Then flag up those who make <5% plays for closer scrutiny, etc. Point being that even with a moderate amount of built-in deliberate imperfections, it may be harder than some are assuming to hide in plain sight.

That's just not how reality works. The world doesn't owe you a living. You are like the buggywhip manufacturer screaming about how Ford is destroying your livelyhood.

Progress marches on, you can't stop it. Its your responsibility to adapt.

I guarantee that if online chess were played for significant money there would be people developing hybrid engines that play more human like.

Well said.

Yeah, except for that minute difference that combustible engine vehicles greatly improved people's lives, when compared to the horse and buggy, whereas these AI *******s destroying our games (when they DO NOT NEED TO in order to make advances) improved nobody's life compared to what it was before.

Also, you know, once these guys kill poker, there will be nothing to adapt to, because, um, nobody will play. I mean, unless by "adapt", you mean, quit poker in order to get some corporate soul-sucking type job. Yay, can't wait to work until death so some fatcat can become marginally more rich.

It was actually extremely poorly said, not in the least because his entire post is an old hoary cliche that's been trotted out for decades, but, whatever.

"The world doesn't owe you a living". Yeah, I can understand that while simultaneously thinking that AI developers and the like are *******s for ACTIVELY WORKING TO DESTROY POKER.

AI developers are like starving wolves and poker players who work with them are like sheep that are pouring worcestershire sauce all over their bodies.

Solving poker is not the ends, it is a means for developing better AI with real-world applications with the potential for improving your life.

Looking forward.

In 50 to 70 years it is very possible AI will do everything better than people.

Expect to see a NY Times bestseller written by AI well before then.

the images all seem offline. I tried the "trick" from a few pages earlier trying to remove .jpg, but it don't work. Can somebody please upload? Otherwise it's impossible to follow the discussion of hands.

they are all still up, quote the post and cut the gyazo link (w/o the .jpg)

It's a nice story and I agree with this, but NSF (main funding agency) does not. Would be cool to see where all the research could lead if academia weren't such a clown show.

I think solving poker is quite specialized. It's hard to see how doing it better/faster improves anything besides poker. Sometimes it's hard to see how inventions carry out to different areas but here I am not very optimistic.

Anything close to solving multiway pots is far far away. You can get some approximations with some guarantees (like hands which have pure strategies in equilibirum are pure in the approximation) but those solutions might still be very exploitable (sometimes even more than 1bb/hand).

Sounds like whack-a-mole.

You can buy out current public solver developers for a few million which is already more than any site would be willing to pay. If they did though there will be more people coming into the public space as it would become empty and ripe for making significant money.

It's quite easy to cheat at chess in a way that increases your performance by say 100-200ELO points and not be banned (I am talking online). The main problem with chess though is that there is very little to none serious play online. People have fun and it's a training ground. Serious competition is played over the board. Chess for money won't happen online because of computers.

In chess you can measure how good you play (assuming the computers are perfect which is an assumption good enough for assessing human play) but in poker you can't. There is inifinitely many strategies very close to equilibrium. You can play various styles, use bet sizings etc. In addition to that you need thousands of hands (at least) to even start assessing similarities and frequencies while in chess it's sometimes obvious the computer is playing after one game.

Sandholm claims a lot of things. Traditional way of measuring game complexity is the number of states possible but this really doesn't work for poker. For example a game of "show me the real number which is the closest to 0" has infinitely many possible moves (you can name any real number) but is very easy to play.
In poker it's a bit similar, number of states doesn't matter, not to mention that it's measured incorrectly by about any paper I've seen (they overestimate by the factor of a million by counting all possible hand vs hand scenarios instead of range vs range scenarios). Poker is simpler than chess to play well for computers. This is obvious for anyone who attempted both poker and chess programming.

Yeah, cars took away horse riders, accounting software took away many accounting jobs. Machines took away a lot of dull farming jobs etc. It's how the civilization rolls. Some people with skills no longer useful are going to have hard time.

.

Same can be said about any online poker tool: HUDs, trackers, seating scripts etc.
People develop them because there will be buyers and there will be buyers because it gives them an edge (often temporary ones but if you are the quickest person jumping to new tools you will always be ahead). It's an arm race. If the endgame is that computers can provide as much value for degen gamblers playing vs them as other humans can then well, it's progress, we can have more people working on other useful things.

When an AI like AlphaGo plays poker against itself multiways, the nuances of multiway will get a lot closer to solved.

My estimation is that the first player to act has N^n number of ranges to consider, where N=number of players behind.

This probably true for the way that commercial solvers are focusing on things (they are very poker-specific bc the end goal obviously is to produce more valuable poker analysis for users). But it's pretty clear that some of the more fundamental ideas behind the approaches that were developed in academia can apply more broadly (at least in theory, it is not that clear yet whether they will be useful). My research has been cited by researchers for a lot of different problems (e.g., cybersecurity, trading, auctions, robot planning, disease management).

As concrete examples, the counterfactual regret minimization (CFR) which was one of of the main algorithms for approximating equilibrium by poker agents, has recently been applied to diabetes management and security.

https://webdocs.cs.ualberta.ca/~bowl...2nips-kofn.pdf

https://drive.google.com/file/d/0BxK...JTOFdTcFU/view

If the game gets "solved", does it get solved once and for all or does it only get solved for at the current understanding of the game? Are there more levels beyond solved?

Also, even the best AI could surely only solve based against a relatively fixed or slowly adapting opponents strategy? Does this put creative players at an advantage?

Lastly, the AI surely can't solve the "luck" element of poker, so would we expect to see changes in strategy against AI where risk is reduced for the positions where luck plays a small part and increased on the more risky spots?

It's all very confusing/interesting/scary/fascinating....

if its solved then its solved once and for all (unless the rules of the game change)

an AI that has solved poker will not lose in the long run to any counter-strategy. In fact, it will win more vs "creative" strategies rather than more standard close-to-gto strategies.

Variance (i.e. luck) has no effect on strategy choice assuming sufficient bankroll requirements. Trying to reduce variance usually ends up in reducing your EV so the best play is to reduce the stakes instead to be within the bankroll bounds.