AI vs GO

[betovernetcapper]

Have recently read a few books on game theory & while computers have been able to beat experts in chess, checkers & even poker, as of two years ago the game of Go was not in the wheelhouse of AI. That has just changed as Google's DeepMind has trashed the top 5 GO players in the world!

I'm far from an expert on such stuff, but this seems like a major leap forward.

[Robert Fischer]

That is so cool.


If I'm reading correctly, AlphaGo is using Monte Carlo Tree Search.

[Jess Hawsen Arown]

Originally Posted by betovernetcapper

Have recently read a few books on game theory & while computers have been able to beat experts in chess, checkers & even poker, as of two years ago the game of Go was not in the wheelhouse of AI. That has just changed as Google's DeepMind has trashed the top 5 GO players in the world!

I'm far from an expert on such stuff, but this seems like a major leap forward.

Computers cannot beat anything. They are ignorant machines--just like vacuum cleaners. It is their programmers that enable them to do amazing things.

Like other machines, all computers can do is turn on and turn off. A long time ago, clever humans realized that by aligning off and on to mean different things based on their sequence, the computer age began.

[_______]

Heads up limit hold'em is essentially solved in the sense that there is an algorithm that can beat the best human players. There is an undeniable "best practice" for this specific circumstance and any deviation from it will be a long term drag on a bankroll.

But there is no algo that can consistently beat multiple players in limit. Or beat a single opponent in no limit.

And it goes without saying, no one is even trying to beat any of the other game varieties that fall under the term "poker".

So including poker in the universe of "solved" games requires an asterisk.

[Delta Cone]

http://www.latimes.com/science/scien...307-story.html

Seems the computers are making progress in heads up no-limit.

[betovernetcapper]

Originally Posted by _______

Heads up limit hold'em is essentially solved in the sense that there is an algorithm that can beat the best human players. There is an undeniable "best practice" for this specific circumstance and any deviation from it will be a long term drag on a bankroll.

But there is no algo that can consistently beat multiple players in limit. Or beat a single opponent in no limit.

And it goes without saying, no one is even trying to beat any of the other game varieties that fall under the term "poker".

So including poker in the universe of "solved" games requires an asterisk.

A team from the University of Alberta's Computer Poker Research Group created a program called DeepStack that did just that in December, 2016. It's written in Science Magazine. There are other AI Poker programs that consistently can beat expert players. They are programed with the usual rules & pot odds & such, but with the programs playing millions of hands, they have "learned" how to bluff & employ strategies that seem counter intuitive. Google it.

[_______]

Originally Posted by Delta Cone

http://www.latimes.com/science/scien...307-story.html

Seems the computers are making progress in heads up no-limit.

News to me. Thanks for the link.

[thaskalos]

Will horse-betting be the next casualty of this AI onslaught?

[Dave Schwartz]

Originally Posted by thaskalos

Will horse-betting be the next casualty of this AI onslaught?

I sure hope so.

[betovernetcapper]

Originally Posted by Dave Schwartz

I sure hope so.

DITTO

[barahona44]

Originally Posted by thaskalos

Will horse-betting be the next casualty of this AI onslaught?

Before I answer that question, I have to download "R2-D2's Santa Anita Tip Sheet".

[DeltaLover]

Originally Posted by thaskalos

Will horse-betting be the next casualty of this AI onslaught?

This question can only be answered empirically rather and not by using logic and analytical thinking.

Given the nature of the game, he who will eventually solve this game will do his best to conceal his findings so it is possible (at least in theory) to never find out. Of course this “solution” can very well become obsolete very quickly as more people will discover and start using it.

More than this, it is also possible for the game to simply be unbeatable and impossible for a machine learning / AI approach to solve it. You can understand why something like this is possible if you think of any unbeatable game (like dice, baccarat for example); no matter how large data or how sophisticated algorithms you will use it is simply impossible to find a winning solution in a game like this . Something similar is quite possible to exist in racing, especially given the huge percentage of takeout which multiplies tremendously the required scaling that the crowd needs to be beaten.

Regarding my last statement about the effect of takeout, I am not sure that decreasing it, will necessary lead to better results. In contrary I tend to believe that lowering the takeout can very well make things much worse for the winning or break even bettor. Think about!

[porchy44]

Originally Posted by DeltaLover

In contrary I tend to believe that lowering the takeout can very well make things much worse for the winning or break even bettor. Think about!

I would think that statement fly's in the face of common sense for obvious reasons. I am not sure what you are getting at ?

[DeltaLover]

Originally Posted by porchy44

I would think that statement fly's in the face of common sense for obvious reasons. I am not sure what you are getting at ?

So, what do you think will happen to a break-even player of the game as we know it today, if takeout will be lowered significantly?

[betovernetcapper]

t

Originally Posted by DeltaLover

So, what do you think will happen to a break-even player of the game as we know it today, if takeout will be lowered significantly?

If the takeout were to be eliminated entirely, player who are breaking even would have an ROI of $1.17. Any reduction or rebate could do nothing but help the player.

The HKJC gives a rebate of 9% on losing bets to large betters. A number of betting syndicates strive to break even. When your betting a quarter of a million or so in a day that 9% is significant.