Suppose I want to create an AI that can observe and arbitrary RTS map in game and decide how to play it. Locating chokepoints, setting up an SC Missile Wall style defense of an island and other things like that.

What if I want to avoid stuck units by having the AI create good paths through the base and use the size of the largest relevant unit to calculate how large of a path from the constructing building is needed?

Just general not sucking at building placement stuff? How would I go about doing these things?

I also want it to evaluate units on its own. So like look at their stats and compare them to other units. Say it could decide if it has longer range than enemy units of various classes so it can decide what stuff to build against a given enemy faction, especially since I support a ton of modding, so I couldn't just code in those decisions personally. And humans are not too smart, so a modder might not realize their faction was broken and certain tactics just wouldn't work for it.

Probably some of this isn't possible, especially if I can't shunt the AI into a separate core to avoid lag when certain decisions are made.

Influence mapping for terrain analysis and utility modeling for the decision-making.

For chokepoint detection, see this thread:

http://www.gamedev.net/topic/601508-choke-point-detection/

How much setup time are you allowing the AI to do this.

If its alot, you can precalc paths (ie- some monte carlo random sampling to feed start+end points to build up a routing network map)

Will the AI be given data containing where the enemy is starting so that is can figure which way THEY have to go to get to where the AIs units will be ??

Then you can project where they can get to so the AI can pick good defensive positions (if that is to be the strategy) or where good ambushes might be probable or even just which path is best for a Zerg Rush of the enemy (if there is a target that must be taken ...)