AI Crushes Poker Pros

Correct me if I'm wrong, I'm no poker expert, but it would seem to have great implications for the future of the game especially when played online. I would guess that this supercomputer is not currently available to the public, but that some similar version might soon be.

https://www.washingtonpost.com/news/innovations/wp/2017/02/01/the-great-artificial-intelligence-gamble-that-finally-paid-off/?hpid=hp_no-name_hp-in-the-news%3Apage%2Fin-the-news&utm_term=.59f0fc40374d

Quote: lilredrooster

Correct me if I'm wrong, I'm no poker expert, but it would seem to have great implications for the future of the game especially when played online. I would guess that this supercomputer is not currently available to the public, but that some similar version might soon be.

I made a decent amount betting this, but I probably actually left a lot of money on the table. A lot of poker "pros" who haven't been online in awhile haven't seen the evolution of the game and so they thought the technology was years away. It was always moving toward a by-the-numbers solvable game, but it has become drastically more so relatively recently.

It's definitely not good for online play. Do fish really need an additional reason to question losing on an online site, lol?

What will be interesting will be seeing if it also affects live play in a noticeable way. If a bunch of online pros decide that online now is too risky/no longer profitable, they'll start playing live, and online pros>>>>>live pros. In addition, online pros have a stereotype of being introverted and unlikable, so if they come into the games and the fish play with them, the fish are more likely to seek out house games because they are not being entertained enough to justify their losses.

Quote: lilredrooster

Correct me if I'm wrong, I'm no poker expert, but it would seem to have great implications for the future of the game especially when played online. I would guess that this supercomputer is not currently available to the public, but that some similar version might soon be.

https://www.washingtonpost.com/news/innovations/wp/2017/02/01/the-great-artificial-intelligence-gamble-that-finally-paid-off/?hpid=hp_no-name_hp-in-the-news%3Apage%2Fin-the-news&utm_term=.59f0fc40374d

Not really. If you look at the format of the tournament this was literally nothing like a ring game or tournament, which would both involve a ton of different strategy. This is set up sole for heads up play and the style they were playing didn't involve any strategy for short stack, super stack, blind level changes, etc.

Also... Bots are programs. You can even program in some randomness by saying "change up your aggression level every X hands according to an RNG" but at the end of the day there are variables that go in to function to make hard coded decisions. Humans, given enough time, could figure out how to manipulate these inputs.

I've played against bots before, actually about 10-12 years ago on Bodog (now Bovada). They were really easy to spot because they acted so fast, and then they were really easy to play against because they did generically programmed things...

1) If you folded a lot on the first blind level or two they would register you as a 'tight' player. Then you could get away with betting whatever you wanted to later. Vice Versa if you bet a lot then tightened up later they would call you with weaker hands.

2) Whenever the bots had something like trips or better they would NEVER fold. So this lead to the players calling bot bets on just about any kind of draw then jamming it all in whenever you hit and usually getting called by the bot (especially if you set up #1 early to think you're a loose player).

3) Over betting the pot to offset the math was another easy way to steal pots. If the bot had a pair and put you on a pair and had X percent chance to win a pot Y size, then by simply over-betting the pot you'd unbalance the equation that would make it unmathematical to call.

These are just a few examples of an older AI, but the same principles hold true for the newer AI. Given enough time our human friends could have figured it's game. From the little I watched of the event I felt the pro's I watched played VERY VERY conservatively as opposed to most heads up players.

The only problem with that is online pro's are used to grinding in 4-8 games simultaneously. I'm not sure how that grind translates to single game live-play.

Quote: Skeptic

The only problem with that is online pro's are used to grinding in 4-8 games simultaneously. I'm not sure how that grind translates to single game live-play.

And they play games of different formats, and other games with different rules (such as Omaha, Stud, Razz, etc). Not only is online fairly safe so long as you apparently don't go play NL Hold'Em Heads Up tables... but even then if you apply the methods from above I'm sure over your "average" bot (not this science born one) that you'd do just fine.

How long before the heads-up tables online are just bot vs. bot? That might actually be fun to watch.

Not "Chinese Poker" but merely Poker in China.

BBC news reports:

Artificial Intelligence poker program wins nearly 300,000 dollars playing human experts in China for a week.

This was reported as an 'exhibition' rather than a tournament, but it still shows how the AI programs are improving. AI programs know that you can't bluff all the time but must bluff some of the time and perhaps the AI programs are getting better at selecting when to bluff.

Limit or no limit?