Poker has been solved

[Doobs]

Chris Hall was asked to play against and write about the new 'unbeatable' Limit Hold'em poker bot Cepheus.

http://www.theguardian.com/sport/2015/jan/08/online-poker-bot-cepheus-flaw

Thin "I'm better than a computer" brag article imo.

More an article on Chris Hall highlighting variance over small sample sizes and a lack of understanding of GTO.  The bot shouldn't adapt to his play.

The optimal game theory line is only optimal if every decision by both sides is the optimal one.

Obviously it could still provide such an advantage that it would still win over the long term, but I suspect that isn't the case. To make an unbeatable program it would have to be able to adapt. It's a limitation of game theory that it only provides a model for framework rather than a clear solution

If it is a GTO bot, it would make a move that would be unbeatable if the opponent played perfectly.  Any time the opponent diverges, the bot gets an edge.  There may be ways to get bigger edges vs specific opponents, but that isn't the purpose of the bot.

If the aim is to play GTO, it absolutely shouldn't adapt.

FWIW there may be several lines that are GTO and isn't just one solution.

For perfect GTO yes it shouldn't adapt, but the opponent diverging from GTO doesn't necessarily give you an edge. It's likely to and that's likely to provide a win - but it isn't always going to be the case.

To be unbeatable then a more thorough understanding of non-optimal moves is needed

If the bot is playing GTO, it is unbeatable in the long run.  If it is beatable, it isn't playing GTO.  If you believe anything de is true, then you don't understand GTO theory. 

[Jon MW]

It's been a while since I went to a Game Theory lecture - but I don't think it does exactly what you think it does

[doubleup]

Jon, the similar machine in Vegas even lets you see its folded cards as it doesn't give a f what you do.

[Ironside]

ok what happens if a bot is put up against an identical bot the only winner is the house?

[teddybloat]

By definition the bot is maximally exploiting the strategy that is maximally exploiting its own strategy.

This is the fabled equilibrium.  

So yes, it would break even against itself as it does not adapt. It doesnt need to adapt as any strategy that deviates from the corresponding equilibrium strategy loses increasing ev the further away from the equilibrium it moves. And since you can only do as well as break-even when playing perfect poker v the bot, it does not need to adapt to win v non-perfect strategies.

[teddybloat]

One quirk of this solving is that even though a winning hu-limit reg could not do as well  v the bot as the bot could v itself, a winning reg would have MORE of an edge v a hu-limit fish than the bot would v a hu-limit fish.

[KingPush]

Never heard of that guardian writer before but I don't think he understands what he's talking about. Isn't gto not a max exploit but just an equilibrium against its own strategy? And so any deviation from that perfect strategy means the bot wins in the long run.

Pretty sure there are no limit bots as well which also adjust and form new equilibriums when playing new players or bots.

[teddybloat]

GTO maximally exploits the corresponding GTO strat, so it is in equilibrium as neither strat has incentive to move from the equilibrium. (As they are maximally exploiting each other if one deviates it must, by definition, lose ev in doing so)

No-limit, certainly deep stacked will be a much harder beast to tame as there are  much bigger game trees with blocker-bets, underbets, overbets etc etc. Might not be enough computer processing power on the planet to map that out across the millions of possible board run-outs.

[teddybloat]

Just read the guardian article and it is lol bad.

Playing GTO guarentees you a minimum ev long-term. If you play perfectly v GTO you will limit it to that ev (break even). If you dont play perfectly against it, its ev will be above its mimimum expectation (you wont break even)

In the article he states he stopped bluffing. Well even if he was betting a perfectly balanced range the  program is calling the exact % of its range to make him indifferent to bluffing / checking anyway. He wont be betting a balanced range, and indeed looked like he was bluffing too much and giving the bot's bluff-catchers crazy ev. He then unbalanced the other way, but it doesnt matter to the bot:

If we simplify that situation and say the writer loses when he has air and wins when he has value.

If he stops bluffing then the program's bluff catchers lose ev, and it will now be bluff catching FAR too often v the writers new strategy. but it will gain all that ev back as the writer now checks back all his air and can only win a pot when he has a value hand, which wont be often enough to make up for the losses..

The only way to beat the bot is to bet in a manner that makes the bot indifferent between calling and folding. That is to be balanced perfecty, that is to be playing GTO

[tikay]

Never heard of that guardian writer before but I don't think he understands what he's talking about. Isn't gto not a max exploit but just an equilibrium against its own strategy? And so any deviation from that perfect strategy means the bot wins in the long run.

Pretty sure there are no limit bots as well which also adjust and form new equilibriums when playing new players or bots.

He is not a "Guardian writer" as such, he is in the poker media & occassionally gets stuff published in the mainstream media as a freelance contributor. He writes very well.

His blonde alias is "NoFlopsHomer", & he has over 20,000 posts to his name on blonde.

He was one of the core of blonde "Live Updaters" for several years.

As a player, & I know he won't mind me saying this, but he is an "enthusiast/recreational".

All credit to him for getting poker into mainstream media.

As an unrelated aside, his first "Live" cash was 8 or 9 years ago, here....


http://pokerdb.thehendonmob.com/event.php?a=r&n=17222

I could probably tell a tale about all 9 of those Finalists, 7 of whom are blondes, 3 of whom are still active on blonde.

[Simon Galloway]

I'm led to believe the HU bots have been removed from Vegas recently... make of that what you will.  In the early days of them being introduced you could rack up decent benefits playing them and breaking somewhere near even.  Diamond in a day was possible on Harrahs properties and was good for liberal room comps and f&b in the Diamond lounges.  Eventually they removed the benefits and there became little point playing it.

[teddybloat]

Opinons on online poker rooms  having marked bot tables?

So bot-hobbyists could put their bots on a table and regular players could play them for real $$$ knowing they are playing a bot?

Or peeps could pit their bots against each other...

[KingPush]

Opinons on online poker rooms  having marked bot tables?

So bot-hobbyists could put their bots on a table and regular players could play them for real $$$ knowing they are playing a bot?

Or peeps could pit their bots against each other...

There would be no point, I doubt many people would make a profit against them.

[teddybloat]

not many profit from sitting ike haxton's lobbies either but he makes a living, ditto the ones in vegas.

HU no limit bots would be good but not unbeatable, and once you work their strat out they could bee very profitable to play against.

i think they'd get action not just off other bot-makers but people wanting to battle bots take the unlimited action when it comes to hitting rake requirements etc

not sure i'd like it persoanally as it would reduce liquidity in other areas and the riggies would have a field day, but i could see why a poker room might want to promote it. would be a very unique feature. each bot having its own table etc.