I'm surprised by this. Why would their technology still be this primitive?
Do they have someone doing this manual "key'ing in" full time? Or on an as-needed basis (they'll call up to have someone begin counting down a shoe thus requiring the AP to continue to play so they can begin tracking/counting in live time)? Isn't the video all digital now anyway (so they can just replay the last 30 minutes on "table 3")?
This technology combined with RFID technology in chips and they have all the real time reports they need. Catching counters is just the tip of the iceberg. It Can be used to evaluate every player for comp evaluations. This can also provide the casino with its exact house edge via basic strategy mistakes, and most importantly they will be able to verify the tray down to the dollar to see if there is a cheating/stealing dealer and or mis pays. It will act like an Inventory control system/Point of sale system does for retail. Management can simply balance the tray at the end of the shift to verify if its over or short.
I have seen this technology in action at one of local casino. Floor people rarely come to table when dealers yelled out loud "Check play or black action" as they used to be. Dealers confirmed that under the fell there are sensors that let them know how much each player bet and how much you walk away with. Scary isn't it?
Isn't that video basically just using OCR with a fancy overlay? But yes, the technology from 2009 to today has come leaps and bounds.
Even with perfect counting, you can still lose money at Blackjack. Good luck trying to explain RoR and variance to those who want to buy it off Amazon...
Regarding it being AI once you have it playing hands of blackjack...not really? It really boils down to one process of counting the cards, one process of making the hit/stand based on basic strategy or deviations, and one process adjusting the bet levels.
Even if you had it make decisions based on hand composition and Effect of Removal, I'm not sure that would be AI, as you'd probably be having it use a table for those decisions.
Now, if you can make it into a neural network with the only goal to "win as much money as possible", with the stipulation that it can't choose to not play, and train it over hundreds of thousands of hours, and have each generation track it's results, that might be getting close to AI. It should, in theory, teach itself how to count cards (Bet higher when the count is positive without bankrupting itself), and maybe learn the Effect of Removal. You could than take that model and plug it into what you have now. Of course, this isn't cheap. You'd need to rent a bunch of AWS servers to train it, unless you happen to have a few hundred Quadros or Titans laying around
Let me know if you want to talk more about it! Tensorflow might be a good start, and Tensorflow 2.0 is coming out soon. There's also a TensorflowJS, in case you like JavaScript for some reason. There's also Tensorflow compatible boards, if you want to make it a lot smaller and more portable. ( https://www.sparkfun.com/categories/tags/tensorflow ) You could than add some LEDs, to indicate hit/stand/split/double, and display the AI's desired bet level with 8 LEDs in binary. Use Nixie Tubes for added coolness. Or I guess buy a LCD display and be lame. ;P
Edit: To summarize, the hard part of this is not the computer vision aspect, that is mostly a solved problem. The true AI part comes in with having it make informed decisions, which I would posit that programming it to count cards is not, but having it teach itself to count cards would be. As a bonus, maybe it will come up with an even better counting system than what we use today!
Last edited by nighterfighter; 06-25-2019 at 07:15 PM.
Now here is AI being used by the dark side. On second thought it might be quite enjoyable.
A terrifying app for making any woman appear naked was killed off by its creator.
https://www.cnn.com/2019/06/27/tech/...-ai/index.html
Email: [email protected]
Fun fact about DeepFakes...
There is now AI to detect Deep Fakes. That is to say, there is AI to detect AI. It's not the best yet, but it's definitely interesting to watch the arms race.
There's also AI to detect cyber intrusions against certain endpoints on your networks, AI to identify faces at border control checkpoints, and AI is being developed for satellites to assist with tasks from health and safety, mission management, and mission operations.
AI will definitely be a lot bigger in the future. Not sure how blackjack AI will fit into that.
Bookmarks