I think that, given enough time, an AI would be better at arena than constructed. Arena is about drafting, the basics, and knowing how to reasonably play around key cards that come up in arena that would screw you over. But playing around these cards is easier to program for because you can just list them out and make exceptions in your AI like if playing against a mage going on 7+ mana don't play a 3rd minion that will die to flamestrike and so forth. But once you've written out the logic for playing around the main suspects you're just kind of done.
The rest of arena is mainly about how to sequence creatures and trade. There is very rarely an overarching goal or cunning strategy.
In constructed you need to play the man more. Get a read on what they want to do. Make reads based on their plays and if they've been holding a card for a long time. These skills don't matter as much in arena because the card pool is so much larger that you can't reasonable know what your opponent has. If they've been holding a card for a long time it might be a key card in their strategy... or it might a hungry crab they were forced to draft.
Arena requires a much better memory than constructed in order to be aware of all the possibilities and the AI will be better at that than a human.
The single most difficult part of writing an AI for arena is drafting. Aside from some kind of machine learning algorithm its hard to imagine how to write an AI that wouldn't suck at drafting on its own... except the machine learning part of the problem is entirely bypassed by user supplied data (arena value and such)
edit: although I guess its hard to quantify what it means to be 'better' at arena than constructed. 'getting to legend' and 'averaging 7 arena wins' are not the same kind of measurement. In the former, skill can be compensated for with time (to a degree) of which an AI has... well... lots. In the latter time is irrelevant.
The rest of arena is mainly about how to sequence creatures and trade. There is very rarely an overarching goal or cunning strategy.
In constructed you need to play the man more. Get a read on what they want to do. Make reads based on their plays and if they've been holding a card for a long time. These skills don't matter as much in arena because the card pool is so much larger that you can't reasonable know what your opponent has. If they've been holding a card for a long time it might be a key card in their strategy... or it might a hungry crab they were forced to draft.
Arena requires a much better memory than constructed in order to be aware of all the possibilities and the AI will be better at that than a human.
The single most difficult part of writing an AI for arena is drafting. Aside from some kind of machine learning algorithm its hard to imagine how to write an AI that wouldn't suck at drafting on its own... except the machine learning part of the problem is entirely bypassed by user supplied data (arena value and such)
edit: although I guess its hard to quantify what it means to be 'better' at arena than constructed. 'getting to legend' and 'averaging 7 arena wins' are not the same kind of measurement. In the former, skill can be compensated for with time (to a degree) of which an AI has... well... lots. In the latter time is irrelevant.