There is a HUGE difference between something effective and something legitimate. Many things are effective, few are legitimate. The test is simple: if it works well on a human competitor under similar conditions, then it's legitimate.
This is not correct, a strategy's legitimacy is always determined by it's end goal. Nothing else matters.
For example: If I wanted to finish the campaign as fast as possible (ie. speedrun) then a lot of what you just described as "cheese" would in fact be the optimal strategy's to use. And had I tried to use "multiplayer tournament tactics" in a speedrun I would have been laughed off the stage.
Beating Alexander by stacking barrels is, whether you like it or not, definitely a legitimate strategy if your end goal is to defeat him as easily as possible within the context of the campaign. Saying its not legitimate because it wouldn't work in a tournament setting makes no sense, because you are just cherry-picking standards by which you judge the strategy out of personal preference. Therefore, the only objective way to judge the legitimacy of any strategy is to see how well it achieves it's stated end goal.
Then we agree to disagree. Let me lay out two assumptions I am going to make:
1. Given the choice of having an AI GM or a human GM, a human GM is likely to provide a better experience for all players involved.
2. The better experience is largely due to the human GM having the flexibility to respond to the various challenges the players pose to him strategically.
So if I am the AI programmer and I learned that people are stacking barrels to get rid of enemies they can't beat otherwise, the first thing I do is to figure out a way to code the AI into behaving similarly as to what a human GM would do on the receiving end (ie: run away/destroy barrels when they see them being moved over). This, as we have argued already, will most likely stop the barrel abuse altogether. At the very least it would have challenged the players into figuring out how to utilize barrels to the next level before the GM catches up to what they are doing. Thus as an AI programmer my end goal is for my AI to behave as closely to a human GM as possible. Not to artificially pose a restriction on how things like barrels can be used, but to have the AI behave realistically and intelligently in order to avoid going down the same pitfall again and again. Achieving that would have provided players better challenges and better overall experience. Sure it would have taken out the fun in figuring out the cheese to abuse, but like I explained, they are called "cheese" for a reason. It's nothing more than exploiting a repeatable behavior problem in the computer code. It might be fun for the first couple of times when you found out it worked. By the time you do it for the 100th battles it might as well be called the "skip this fight" button.
