Tl;DR Can pox be dominated gameplay wise by AI (Basicly undefeatable) Bonus: Can AI dominate BG building? So computers really changed the game of chess changing and countering openings with moves humans couldn't realisticly discover without a life time of dedication. I've often wondered if theoretically pox could be dominated by a computer, I mean we 've all seen tiny play, in his best it's almost robotic (other players like devils,nid and few others too, but their moves have personality and flavour; don't feel robotic to play against) Technically pox feels like a straight forward enough game tactics wise, so why not, but on the other hand there is an idea where the computer would have to estimate chance of enemy deploys etc;and pox sure as hell isn't 64 squares. What do you guys think tags for people I feel might be able to answer: @Sokolov @BurnPyro @kalasle @MrBadguy EDIT: can AI also dominate BG building?
Well tiny is a chess champion so it should feel like his moves are more "mechanical". I do not think an AI could dominate Pox as long as there are hidden runes in hand that can have a strong impact on the game. AI should have too many things to consider and shouldnt be able to elaborate a strategy good enough. In chess all possibilities are on the table. In poker-like games, the number of hidden cards to be considered is a lot smaller so the guess on chances its easier. I dont think in pox it could be possible to do it. Even in Go they only recently go an AI good enough to beat high rank players, and still its a game with no hidden cards (but a huge number of variations).
I think it's possible but not worth putting the programming time into. The game is too complex and you'd have to program behaviors that account for everything a player is thinking about which is not present on the board itself. So I conclude that it's technically possible to create AI that consistently beats more people but it's unfeasible in practice.
pretty much chess has a board and a default state, pox has a default map and everything after that is random, only based on what faction is played. You'd have to teach it meta bgs, player recognition to search for paterns in plays and deckbuilding etc, on top of that factor in reveal rate and nora spending
letting the program see the cards the opponent has available could probably halve the work. it's a lesser cheat, i've played games where the AI has more resources depending on dificulty.
Yeah, pretty easily. That is, it would require either a lot of programming, or a lot of running the AI through games to train it, but conceptually it's not that hard. Pox is the kind of thing you can get an AI to do. Probably? This one is a bit more complicated, and the kind of AI training you would have to do for it would be even more intensive than training it to play the game, but yeah, I think you could make an AI that would be good at building BGs. P.S. Don't mythologize Tiny. If he seems like a robot, that is because he has a solid strategic approach and consistent execution, but he is also totally un-inventive and never adapts his style to the BG he is playing. I think he has plateaued in a notably sub-optimal way.
Trust me, if there is anyone that doesn't overshoot tiny it's me, as someone who has played him an uncountable amount of times. He definitely isn't what he used to be in terms of seeing the best move ( a lot of OP BG crutching in his late days), but I still stand by what I said as a matter of respect to the times he played flawlessly top level. Which doesn't seem amazing to some, but consider a game where you made no mistakes or off-moves against an opponent who really knew what he was doing (not strong BG, I'm talking he himself would rarely blunder/not make the best move) Granted this isn't really often seen anymore, I dare say if I logged in and watched every game for the next month, I don't think I'd see a single game where I felt one side played flawlessly against substantial resistance ( even with tiny playing, he doesn't pull it off or have the opponents to do so anymore)
Tiny can adapt. He simply doesn't have to do so much anyways and doesn't have as much free time as of late what with being married and probably having a kid by now. I remember him tweaking his BGs or counter-picking top players at several points in time. One thing Tiny seems to consistently do is to take his time and think turns through when he can, which I believe leads to him making better plays than most. A lot of scrubs such as myself mostly make moves on a whim without thinking 2, 3 or more moves ahead which can backfire. To paraphrase @zorbot after losing to a game in which Tiny used 3 Swaps "Tiny is always willing to play decks that can get away with not doing anything for 50 turns". I believe having the patience to think things through and engineering strong BGs is why Tiny has consistently been top tier.
Chess ai is not a neural network - it knows all possible moves for endgame, all predefined debutes, and uses algorythm with huge coverage of possible moves but few evaluation parameters for everything in between. For pox that won't work - you'd need a real thing, which would require shiftload of manual training and midlevel definitions even to jumpstart, and who knows how much to perfect it. But it should be totally doable, and I believe even dumbest thing with minimal training would be worth it.
