The AI is treated like a player, and it's very separate code. IT's not able to get that info. Even if it had access, the way the code works it doesn't have a concept that would allow it to ponder the sequence of dice. It's just looks at any given state of the board, and the current roll, and through trial and error has figured out what moves will give it an advantage. Its not currently being trained, and hence could not (even if it had the data which it doesn't) be able to ponder a strategy of what might come up next with the dice, or further more deal with the unpredictably of when you pick your dice which would change the sequence.Death_Row wrote: Jonas... since there is a list "PRNG".... has there been any testing done to verify the AI isn't somehow utilizing (gotten access) those pre-defined numbers? There should be some way of exposing the list for testing purposes to find out how the numbers are being pulled off (by the bot) the list. Since no one knows what or how the AI/neural-net is doing things.
Also when the AI was trained it wasn't using OUR RNG, so that would rule out that it learned some obscure pattern if we had one in the roll calculations.
Its basically not a issue, there are just some aspect that make it an impossibility. The AI just plays in a way thats a little foreign to human players, appearing to make silly moves one time, and stellar moves at others. That perceived erratic play in a human opponent might make one ponder if they are just getting lucky or cheating, but i this case it's just thinking in a different way.
We may do that the next time we update the Backgammon client.Death_Row wrote: I still believe the implementation of the cups (giving control back to the players: single and server player) would help the perception of the game.