I was referring to the AI deciding that, for instance, a fur trapper's hut may be more valuable than the lodge.
If it
always decided like that without and thus
destroys balance of Improvements, it will be really bad.

If it decides it
reasonably by checking current situation, that that is good.
Generally AI needs to have a
good balance of Improvements.
A) It will need some
Food
B) It will need some
Lumber
C) It will need
Cash Yields
D) It should then worry about
Ore (but also Cash Yield) <-- Already in C
E) It might then worry about
Stone.
All we need to do is to teach AI how to
balance the amounts of different Improvements it builds.
It does not really need to consider at all how these improvements will look
in 50 Turns ...

(Since ALL of them will grow over time.)
Summary:
AI should not be messing around by replacing Improvements it
already built.
AI should be concerned about building
more Improvements in a
balanced sequence.
Example:
Priority 1: Food -->
If (Low Food): Find best plot (without any other Improvement and not destroying Bonus Ressource) to build
next Farm.
Priority 2: Lumber -->
Else If (Low Lumber): Find best plot (without any other Improvement and not destroying Bonus Ressource) to build
next Lodge.
Priority 3: Cash Yields --> Else if (a
Plot worked but no Improvement on Plot): Find best plot (without any other Improvement and not destroying Bonus Ressource) to Improvement for most valuable Cash Yield.
Priority 4: Stone --> Else if (no Stone and Plot with Bonus Ressource Stone): build Quarry
So basically we just need to set up a good priority list.

(And have some good decision factors what AI should build
next. If it starts replacing it will just loose time inefficiently anyways.)