it would be cool if the AI were updated to be more challenging. Especially Elite Four members and gym leaders. but an app like that would be challenging to add I think, although I don't know enough about that kind of stuff to be accurate.
It wouldn't be THAT hard to record matches over the Internet, and it would only be moderately difficult to translate metagame strategies into computer-readable formats. The hard part would be analyzing them. Here's how I would go about it:
A team of about 20-30 employees would then read the Serebii Pokemon of the Week info and other sources to learn about the strategies in human-understandable language, then they would search through recordings to see that strategy in action. (Since the battles would be stored as sequences of actions (a la
Brawl replays), it would be perfectly easy for the team to do a simple search for Pokemon X using move Y.)
Once a number of replays showing a given strategy in use are identified, the team would assign the strategy some sort of ID number or other tag that the computer would be able to read and learn. This tag would also identify which Pokemon can legally use the strategy. Repeat until most ordinary strategies are recorded in the system.
Here comes the tricky part: the team would then sit down with a list of strategies and watch them in use, then judge their overall effectiveness and difficulty to counter. This highly subjective rating would be averaged across all employees who rated a given strategy and then compiled into the database, essentially giving each strategy a "difficulty rating." From there, a simple code will assign each strategy a probability of use in various lookup tables (each balanced for a certain level of difficulty). When the player encounters a Trainer, that trainer's file contains a variable that tells it which lookup table to randomly select a strategy from, as well as any relevant modifiers (i.e. early-game trainers' random "dice rolls" would carry a heavy penalty, ensuring they only select strategies that are easy to counter). If a given strategy is illegal for the Pokemon that Trainer has, the game will select a new one.
This is, essentially, the same A.I. currently used for in-game Trainers, just tweaked a bit to pseudo-randomly choose
series of moves, instead of just random moves.
Also note that, to prevent in-game Trainers from ONLY using really effective battle strategies, the team would have to come up with various "dud" strategies and place them into the database and lookup tables. This would have the effect of representing poor training or poor prediction, and would serve to randomize moves a little to keep things interesting for players. TL;DR not every Rattata should be a FEAR Rattata.
