Freaking impossible! I bet it'd be anyway. Sort of a masters level thing, or a PhD dissertation.
Well... basically you could play it with statistics. You have easy statistics for the chance of winning a roll, and you could program in statistics for... say, someone has 30 dudes parked on your border, and he keeps plunking down more every turn, statistically he's threatening you. But then you'd get into questions of quality, judgement calls... you'd quickly need to program a personality in. If I had to make the call between Europe and Africa, and it were an equal call, I'd go for Europe every time, but that's only because it's position is superior to the deadend of Africa. If no one was in S. America, I might go for Africa... You'd need to build a statistical model of the entire board. I guess it would be like chess computers, who work from an analysis of thousands of historical games, just like humans do.
Of course, remember, that it took bazillions of dollars and years of research to build a computer that could consistently defeat a player in chess. Barring some secret resources of lackattack, what you're saying here is, let's create an idiot, someone that will play predictibly and stupidly, and then assign points for defeating them.
I've given this enough patient thought to dismiss it as ludicrous without sounding abrupt, haven't I?