I think I get the idea, a smarter AI isn't necessarily the one that best replicates the actual experience of playing Tic Tac Toe with someone. I let AI 1 win, because it was cute. But maybe it was just the most calculating of all!
I wonder, I got a different conclusion: the "best" AI is the one you think is the best. It could be any of these - they're definitely not the same, they are all distinct, and have their own strengths and weaknesses. It's kinda human! I like your conclusion too.