Research Article: Using Machine Learning to Increase Card Counting Performance

[Overkill]

The below link will allow you to download an article relating to using machine learning to enhance card counting performance. It's largely 'over my head,' but it references Humble, etc. Potential pitfalls include simulations with unrealistically player-favorable penetration as well as the use of "traces" (programming shortcuts in lieu of more involved simulations).

But here is the abstract followed by the link:

Abstract:
This paper investigates a new method to boost Blackjack betting returns. We show betting criteria identified by genetic algorithm significantly outperform standard game theoretic criteria on ten different professional counting systems.

https://www.researchgate.net/publica...tting_Criteria

[Norm]

I read through the abstract. Seems to be saying that most players don't play basic strategy well so he has a better method of playing blackjack than poorly played basic strategy or other attempts at AI play. Must be missing something.

[Overkill]

The author discusses new, evolved AI betting criteria (new, evolved AI PLAY criteria and new, evolved AI COUNTING criteria have apparently been discussed in other works). In this paper, Coleman says, "The new (betting) criteria has a nonlinear form similar to the version in Humble and Cooper (1980)."

The author applies this new criteria to 10 of the most popular counting systems...

...(cue quietly chirping crickets). ...I had thought this thread would be quite popular. But perhaps the material is tough to slodge through, and/or perhaps the author is 'missing the mark' due to, ironically, OVERsimplification in his approach.

[Norm]

Makes no sense to me. And my first program, 57 years ago, was an attempt at a heuristic checkers playing game. Games like checkers can benefit from machine learning. But your opponent in blackjack plays by fixed rules. Unless you are talking tournament play.

[Archvaldor]

The below link will allow you to download an article relating to using machine learning to enhance card counting performance. It's largely 'over my head,' but it references Humble, etc. Potential pitfalls include simulations with unrealistically player-favorable penetration as well as the use of "traces" (programming shortcuts in lieu of more involved simulations).

But here is the abstract followed by the link:

Abstract:
This paper investigates a new method to boost Blackjack betting returns. We show betting criteria identified by genetic algorithm significantly outperform standard game theoretic criteria on ten different professional counting systems.

https://www.researchgate.net/publica...tting_Criteria

Most card-counting systems capture very close to 100% of the returns from a hypothetically perfect betting system. There's no room for improvement.

AI is not going to help. AI will not help you mix cement. People know how to mix cement. There is no super clever AI way to mix cement. This is the same.

[Norm]

Agree except for the example. Actually, modern concrete is not as good as the way the Romans mixed it 2,000 years ago. They mixed in quicklime in a manner that made it self-healing when cracked. This is why some Roman buildings have survived so long as well as they have.

https://news.mit.edu/2023/roman-conc...ime-casts-0106

[UncleChoo]

While on the subject of AI, anyone look into that new chatGPT tool? https://openai.com/blog/chatgpt/
I am not qualified, but maybe one of the immortals on here can put it to the test by asking it questions about complicated Blackjack theory.
The hype around it seems a little over the top.

[Archvaldor]

While on the subject of AI, anyone look into that new chatGPT tool? https://openai.com/blog/chatgpt/
I am not qualified, but maybe one of the immortals on here can put it to the test by asking it questions about complicated Blackjack theory.
The hype around it seems a little over the top.

It is a little over the top, but it can do some extraordinary things. For example it wrote me some code for a simple blackjack game.

The problem with chatgtp is that it isn't really AI, it is actually a lot closer in theory to a discipline like meteorology. It is searching its archive of data for a comparable
answer to the question you are asking it. Its archive must come from the web (the only archive large enough), so enough people believe some
stupid nonsense online it will regurgitate that. Blackjack and gambling suffer disproportionately from people believing stupid nonsense online.

Additionally when it doesn't know something it lies. A lot. Very confidently. It is really sociopathic.

[Trixar]

It is a little over the top, but it can do some extraordinary things. For example it wrote me some code for a simple blackjack game.

The problem with chatgtp is that it isn't really AI, it is actually a lot closer in theory to a discipline like meteorology. It is searching its archive of data for a comparable
answer to the question you are asking it. Its archive must come from the web (the only archive large enough), so enough people believe some
stupid nonsense online it will regurgitate that. Blackjack and gambling suffer disproportionately from people believing stupid nonsense online.

Additionally when it doesn't know something it lies. A lot. Very confidently. It is really sociopathic.

flow of the cards.jpg

sorry, I can't resist

[Norm]

Interestingly, I asked and got a different answer. Also incorrect based on dealer bust rate nonsense.

[Trixar]

Full disclosure, I asked it to regenerate the answer. The first one was more appropriate, but still not great, probably similar to what you saw.

[Norm]

Yes, I tried some simple BJ questions a couple weeks ago, which it answered correctly. Haven't asked anything interesting.

[UncleChoo]

Yes, I tried some simple BJ questions a couple weeks ago, which it answered correctly. Haven't asked anything interesting.

I can't help but wonder if technology like this will ultimately end up dumbing down society as we end up depending on tools like this. Maybe not so much now, but more towards mid 21st century and beyond.