Honestly, most of the players are building tons of the strongest units with tons of siege units:
Macemen -> Riflemen (sometimes Grenadiers) -> Infantry -> Tanks -> Modern Armors/Mech Infantry
Cannot see any unit building strategy in stories.
The problem is: The AI is not able to counter those simple spamming "tactics". That's just stupid. A human player would adapt his strategy when he sees his enemy with tons of tanks.
Therefore I support the idea of AI scouting and then adapting the strategy.
Scouting and then adapting or winning more battles with a certain unit type and then adapting? These are different choices, so what is your proposal?
I don't think that adaptation based on victory percentage will work well. A few problems that might arise:
-Siege units cannot win a battle, siege units lose a lot thus siege units are bad. Build less siege units.
-If the AI were to build an army of 5 macemen, 2 crossbowmen and 2 pikemen plus 5 catapults, then that would be a nicely balanced stack. It would subsequently march it into my territory. I would sacrifice 3 catapults and lose 1 more unit while mopping up the stack. In this case, the macemen will likely be considered the best units since they likely beat my sacrificial catapults. But that's of course a nonsense conclusion. Building more macemen and less crossbowmen, pikemen and catapults will not create a better stack of units. It will become more vulnerable to counterattacks on the weaker areas where the stack lacks enough counterunits.
If the AI had added a few knights to that stack, then they'd likely have beat the catapults making them the preferred unit by the AI. But adding more and more knights to the stack while reducing other units wouldn't have made it a more dangerous stack. It would have likely made it more vulnerable to attacks by the counter unit of knights.
What I'm trying to say is that it's not necessarily the unit that does the killing that is most important in the stack of units. Some units are just there to avoid lethal counter strategies. These lethal counter strategies will then never be used and thus the unit proved its worth but that won't translate into a higher victory percentage of this unit and thus won't result in building it more. Shifting construction focus onto the unit that happened to win a higher percentage of battles can easily make the balance of units in a stack weaker.
However, I'm not against adaptation of the AI. I think it would be very good if the AI adapted to what it saw through scouting, but it's not easy to get the AI to do this well. For instance, if the AI can't scout your lands with units or espionage, then it has to guess and we of course don't want it to guess based on the 3 units of our army that it happened to see on its borders. So in that case, it probably will have to build a standard mix. This standard mix should then be adapted to what it encounters when it goes into a war and learns more about enemy troop composition. The AI should also adapt to multiple potential enemies at once with more weight towards the ones which it is more likely to get into conflict with.
Even if the AI can freely enter the lands of an opposing civilisation to scout, then I'm not sure if it will be easy to count the various numbers of units. It might be hard to teach the AI not to double count some units while missing others.
