You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!

AI Plays Poker, Design Chips?

Arthur Hanson

Active member
If AI can master the very complex game of poker against the very top champions, how long before it conquers complex tasks like designing chips or just about anything else. If not fully designing chips it should be able to leverage human designers dramatically. It could be in the future humans don't design almost anything, but curate a program that designs almost anything in a particular field. AI/ML is about to break out in the next year or so with the proliferation of advanced memory, processors and programs. Dickerson the CEO of AMAT and considered one of the top 30 corporate executives has stated we on the cusp of AI coming into being and this poker program is to the point it even masters intangibles and bluffing even better the best of the world champion poker players. Below is a reference article and comments, thoughts and what companies are in the best situation to tackle and prosper from this revolution would be appreciated. I feel AMAT, MU and TSM (full disclosure, I own all three and have for many years) are just a few of the companies set to prosper and several of the EDA companies.

 

Daniel Payne

Moderator
Arthur, I listened to a radio episode talking about that poker-playing AI system this week, but remember that AI today is totally domain-specific, meaning that there is NO general purpose AI out there. Yes, AI and specifically Machine Learning is used in many places of EDA tools, just click the AI menu on SemiWiki to see which vendors and tools are adding ML, https://semiwiki.com/category/artificial-intelligence/
 
I agree with you Arthur. As a poker player and semiconductor professional I knew it was only a matter of time before AI dominated game play. The advantage AI has is that it can quickly evaluate hands and not make emotional decisions. AI can also evaluate all known players through previous games. The players cannot analyze AI bots since they have not played before. At some point in time professional players can understand and game the AI bot but the AI bot can game them back even faster. This really is another step towards AI dominance. Many thousands of companies are developing domain specific AI chips. It really will be tsunami like conditions, absolutely.
 

Arthur Hanson

Active member
Dan, from a conference call I listened to I had the understanding that the new AI gaming programs didn't need records of the previous players play. The reference I had was to chess, but they stated they only gave the AI the rules only and no game records what so ever and the AI developed its own strategies. This was on a Micron call and said they used a GAN type AI and it developed its strategy in a few hours, three I believe, but I'm not absolutely sure on that since I do about fifty hours of various types of research a week. I'm not a poker player, but play very high stakes in trading, nothing less than a thousand share block. Any feed back on this would be appreciated, especially the role of advanced memory and AI/ML.
 
This was an article that I read in Nature Briefing (email update). This article had the 'high' points but did not delve into specific poker players used in the study or how it was performed (or even about the algorithm). The article references a much deeper article and names the players. Here's the link: Brown, N. & Sandholm, T. Science https://doi.org/10.1126/science.aay2400 (2019)
 

Arthur Hanson

Active member
ou Arthur. As a poker player and semiconductor professional I knew it was only a matter of time before AI dominated game play. The advantage AI has is that it can quickly evaluate hands and not make emotional decisions. AI can also evaluate all known players through previous games. The players cannot analyze AI bots since they have not played before. At some point in time professional players can understand and game the AI bot but the AI bot can game them back even faster. This really is another step towards AI dominance. Many thousands of companies are developing domain specific AI chips. It really will be tsunami like conditions, absolutely.
[/QUOTE]

Dan, do you think it's possible MU has developed a way using a GAN AI system that uses the speed of 3dXpoint technologies in each adversarial
 

Arthur Hanson

Active member
Dan, from what I understand about GAN AI is that the two adversarial networks go point counter point so fast they develop and learn literally every strategy on their own and would be able to pick the very best strategy on every play what ever the game is or task for that matter. Any thoughts or comments on this would be appreciated. I have been working out a strategy for dealing with this in trading for companies like Renaisance I have no doubt have been using AI or at least ML for a while. Any thoughts, additions or comments on this would be appreciated.
 
Last edited:

Randy Smith

Moderator
So I researched this a bit more and found out that one of the limitations of the game played with the AI was that all stack sizes had to be equal. They did not teach the AI how to adapt to changing stack depths. As a poker player, this is a huge issue. This AI could not compete in a normal tournament or cash game. It could not deal with players not all having the same amount of chips to start every hand. This is easily spotted in the graphic leading the article as well. What this really means is that for one specific situation that would occur only on the first hand of a tournament, the AI would have an edge. This is not the monumental result the author was trying to portray.
 

Arthur Hanson

Active member
I have read about other AI programs playing against professional poker players in the past year and they said they didn't stand a chance, but I don't know all the specifics.
 

Arthur Hanson

Active member
So I researched this a bit more and found out that one of the limitations of the game played with the AI was that all stack sizes had to be equal. They did not teach the AI how to adapt to changing stack depths. As a poker player, this is a huge issue. This AI could not compete in a normal tournament or cash game. It could not deal with players not all having the same amount of chips to start every hand. This is easily spotted in the graphic leading the article as well. What this really means is that for one specific situation that would occur only on the first hand of a tournament, the AI would have an edge. This is not the monumental result the author was trying to portray.
From everything I've read the poker AI programs were able to deal with varying bets. I have read several articles were top pro players said they didn't stand a chance no matter what they did. I'm sure they greatly varied their bets as just one of the many strategies they tried to use to defeat the AI, since this is a very standard part of all poker strategies and would be duly noted if it didn't work. My mother was pro card counter in 21, but quickly got banned from the tables when the pit boss saw she varied her bet according to the odds. My father, an engineer bought her the Thorpe book and she spent about a year memorizing the tables. Any counter in 21 varies his bet radically based on the changing odds. If this wasn't done, high end poker players would have made prominent note of the flaw. I could never see them not prominently not noting such a serious flaw in such an experiment.

 
Top