Android Vs Human

yes, it is possible.

---

ai-junkie.com

Actually, the MMORPG "Ballerium" (still under development)
Is supposed to implement this.

Here''s an article:

AI helps gamers keep on playing

Updated screenshots can be found here

I plan on using such a recording system to create AI for a multiplayer RTS I''m working on. If I ever get it working, it will function by recording a player''s actions and store context information about each maneuver. It will then(possibly after the game is complete) analize each maneuver and ''guess'' if it gave that player an advantage for at least X period of time. For example, it would record that a player moved a sniper on top of a hill and the context data would show that the move let the sniper have a clear view of nearly half the enemy''s army. Since the sniper was able to take out many important units before being taken out, the computer would (hopefully) assign a ''good'' rating to that maneuver. From then on, the AI would check if a simmilar situation exists, and if so is there a sniper nearby to take advantage of it. If yes to both of those conditions, then the computer would try that maneuver. Since it would also analize its own movements, it might learn that it isnt so good at pulling a certain move off and so use it less and less. Or, it might discover that moving a sniper in to place and charging forward with a large force lets the sniper live longer and pick of all the important enemy units, so it creates a link between them (or actually, a more complex condition and a more complex action, so each action individually still exists and a new combination action has been added to the list of things to do)

The amount of data that accumulate would make deepblue shiver, or cant we just say that there is such and such actions that the player can do, and then see which he does the most, offcause a pathfinder is on top, so we wont have rambo''s runing i circles slicing air, but in design have a math that measures if a player likes to lay flat when he shoots on large groups of enemies.
I think that a AI will have a problem calculating the enviroment.

I don''t think the amount of data that will accumulate will be that large. It would only take one byte to store each action (move, attack, etc) and a few to store the parameters of the action (to a high place, where you can see several enemies, etc), a few for conditions (is the enemy army gathered together? is there a sniper that isn''t currently busy? is the sniper near the high place) etc. Its kind of like a genetic algorthim, except that its not evolved by trial and error of the AI itself (though it would take data from itself to add rules).
I might try to make it a scripting language, where the bytecodes would be instructions like "pick a high place: from which a: sniper: could see: several enemy units", "pick a: sniper: that can reach: chosen location: in X amount of time: AND: has at least: 10: bullets", "order: chosen unit: to move to: chosen place: AND: shoot: enemy officer". I used colons to break apart the different ''instructions'' that could be used to create actions. Each byte might represent one of the sections. Each action would be less than 150 bytes I think, so it would take quite a lot of them to fill up RAM.