I think you need a specific narrow AI like what I proposed earlier for playing poker. Hyperon would then recognize that what the user wants is a prediction from the poker AI, and it would make that connection, potentially passing the relevant information to the narrow AI for processing. Please @bengoertzel correct me if I am wrong.
Yes I think one pluribus like on hyperon could be very nice ! But poker is a game of money. And I think we should focus on giving solutions to providers because the solutions for players are way too obscure if you dont have any experience before and if you follow all computer advices in real time you are just cheating
I was more suggesting about an AI who could extrapolate on why the computer do this thing rather than another one. What if I do the other thing ? where do my loss in EV come from ?
@Roro Models that are capable of explaining themselves is in no way a solved problem. For sure an interesting area of research, but at the moment if the AI could ‘explain why’ it does something, that would amount to > “because trial and error with my specific configuration led me to the conclusion that this play is optimal” or maybe “The gradients told me to do it”
@Roro What if instead, it acted as a training partner or as its own entity? You could set something up giving players the option to play against it. Perhaps the money they receive for winning is greater than the pool (some advantage multiplier for the players). They use AGI to play against it, winner takes the AGI.
If it were a training partner, perhaps you play against it, and after you select an action, the AI’s suggestion is provided, (or the probability distribution over the action that the AI would use in your position is revealed)
That would be nice for sure, in the poker world there is some apps where you can train with already solved solutions as you suggest. But sometimes the proposition of what you should play is very obscure and could change between two situations who could look very similar for humans, especially if they don’t know what they should look at.
I think it is important for the training partner to give you something you could use in practice against humans - likes what parameters are importants ==> if put data in spreadsheets of multiple solved hands you can speculate on why the computer does what he does. Isnt possible for an AI to do it too ? and merge the two AI together to create something good haha (even with an other one trained with hands played by humans to have a better adjustement over frequencies when suggesting something to the end user)
Partypoker launch one “bot” to advise players - I have no clue if it is some sort of IA or just something comparing what your statistics are with some spreadsheets but there is for sure a market in “poker education”.