With that in mind, I think that a monte carlo program, which can play without lots of expert knowledge, could be employed to generate at least a moderately competent AI. The upshot is that the game will not be infinite in competitive play. This list may be long (I envision between 100 and 300 patterns), to ensure that the game retains the variety which is its strong suit. For the purposes of competitive play, the players will not be allowed to invent new patterns, but will rather have to choose from a list of patterns. Although I agree that programming Mind Ninja is difficult, I don't think it's as difficult as hk supposes. Nb ?: Hi! I'm the inventor of Mind Ninja.
Not only would an AI have to be able to evaluate and play an immense number of patterns, it would also have to be capable of proposing new patterns. So the game's state space complexity is theoretically infinite! In the most general form of the game, shape and size of the board is part of the negotiation phase, and there is no upper limit on size. The game starts with a negotiation phase, where one player defines the pattern to be built (within certain constraints), and the other decides whether he would rather attempt to build it, or attempt to block it from being built. Hk: Nick Bentley's "Mind Ninja" (an awful name for a great game) is fundamentally hard to write an AI for. Humans are still ahead of computers, but it looks like the gap is rather narrow - narrower than in Go even. The challenge occurred this year, and we were surprised to find that he lost three games, to two separate MCTS bots. In 2002, he offered $10,000 to any bot that could beat him in one game of a 10 game match. Hyperpape (): Recently, Christian Freeling (a top ten player at little golem) lost the Havannah Challenge. I don't think there is any obvious parallel to that basic approach in more complex games, but it might just be worth digging up the very first computer Go approaces, and see what one can do -) The (limited) theorem proving approach that good Hex programs use is based on a systematic way of generalizing a basic approach invented by Claude Shannon. Is a strong opponent possible with hex methods? - ab
0 kommentar(er)
