So I'm filing this under "Brilliant if it works, but I have serious doubts whether it would work." Not because I doubt your design chops at all--it's just fundamentally I'm not sure either the human players or the algorithms will behave the way the concept requires.
My personal observation is that, once you take the human element out of a social game, what you have is no longer a social game, but a strategy game, more like, eg, Crusader Kings. First because it's no longer symmetrical--the computer doesn't know it's playing a game and won't be gaming the player the way the player is gaming it. And second because, even if the gameplay is truly identical, players will act differently if they know it's not another human. (For example, players will test the algorithm and try to find stock phrases that always yield a favorable response, something that's inherently not possible with humans.)
That said, all that is just my skepticism, and by all means, prove me wrong and I will happily eat crow, because it is a very cool idea. I like your setting and a lot of the details of the concept, like making the interface resemble a tabletop game. And it definitely rises to Willem's challenge, because if it works, it'll definitely be a whole new thing that many other games are guaranteed to copy.