Post by Hólmar Sigmundsson on Feb 15, 2009 18:00:24 GMT -5
1) You talked about combining the best of both worlds - simulations and traditional game-tree search. Since the paper was published/written, have you researched and/or implented that in the CadiaPlayer? If so, how effective is the combination?
2) Has the CadiaPlayer impoved in games that converge slowly (e.g. Skirmish) since the paper was published/written. If so, was it because you implented an evaluation function for assessing non-terminal states or for some other reason?
Post by steinarhugi on Feb 15, 2009 19:01:22 GMT -5
Humans examine their opponent strategies and try to use it against them. Have such methods been implemented in a GGP engine? What are the hardware limitations in the competition? Is the CadiaPlayer still in development? If so, what are the improvement plans for this years competition?
Post by Haukur Jónasson on Feb 16, 2009 19:05:03 GMT -5
1) How does CadiaPlayer fare against human opponents?
2) More general, regarding AI players in various computer games. Almost all of them have some sort of difficulty setting (some even have the option to scale depending on how the player is doing). How is this achieved? How do you make the AI 'dumber' to accommodate for less skilled players?
"Artificial intelligence is no match for natural stupidity."