I've also done script AI programming for custom maps for games, I've done class one brains (learning machines) for games and business purposes, I've done fuzzy logic systems mainly for industrial purposes, I've toyed but no serious use of organic algorithms but I might get to soon for a healthcare company...
I am no AI expert, perhaps a journey man of sorts, and I don't know completely how elements is designed and programmed; however, I am pretty certain it uses a client/server like function where flash is the client side... this does not lend it to learning systems. The game itself does not lend itself to fuzzy logic either. Scripting each deck is both tedious, requires additional knowledge outside the game, and takes away time from playing the game for the player and a lot of time away to update and re update the AI decks for each update, only the algorithm method would work here outside of hard fast rules programming, this requires the most knowledge of which not every developer has.
Again I am but a journey man, but something like HTML5/JavaScript front end with a Java/C# backend recode job, might lend this game to be able to make use of a database backed learning system that organically build's it's schema and relationships, and bind to a series to three algorithms, a planning algorithm that makes awareness of it's deck calculating things out like highest damage rate and what combinations in the past based on the learning portion result in good use. A second algorithm to track possible events based on what the opponent plays, so the opponent has a aether mark, phase shield is a possibility, based off of previous games it was best to hold defrags, Then as an overall a field control algorithm that takes these two into account but makes the final decision based off of a third set of rules, that handles the field, ahhh a sundial, is in play, don't spend the sky blitz yet.
Just my two cents
