TL;DR:
Simplify the AI choices to 3-5 and the AI will have a 1 in 5 chance of picking the correct decision.
A) The tweaks need not come from the AI crunching thousands upon thousands of choices. It should come from simplifying its choices so it can make the correct ones.
AI should have 3 victory flavors, 1 flavor is chosen at random by every AI civ at the beginning of the game.
- diplomacy (befriend all city-states and make alliances with trading partners, build culture)
- science (out-tech everyone in a race to apollo program, build science, fielding a small but technologically advanced army capable of dealing massive damage to outdated ones)
- expansionist (eat land for 60% of the world domination, fielding massive but outdated army to conquer more land, typically warmongering)
- The AI will go to war (or cold war/espionage) if those victory conditions are being impeded.
all the other choices the AI is to make is in the service of those 3 victories and if it keeps the AI on track for the victory conditions. That also means the AI has to have pre-set build orders, priority buildings, etc. for those conditions.
That doesn't make the AI play better. It makes the AI look like an idiot. Someone already explained why.
B) AI's been called stupid because it doesn't defend cities well enough nor can't prevent being overrun by a larger neighbor. well here's a simplification of choice that would work for the AI.
1) Have line of sight for AI (assumes AI can only see part of the map: sphere of influence + fog of war similar to the humans)
....I'm not quite sure what you're saying. Have the AI have what it already has?
2) AI will check units in line of sight near borders and up to 3 hexes away, is number of melee/range above X? if so change over city production from AI victory flavor to accelerated war production and/or rushbuy. Status on Guarded.
Might work.
3) if AI has tall empire, call on allies for a round of denouncements to discourage from DOW. If possible, have pact of friendship drag allies to war or face a very harsh reputation penalty (this is better than a warmonger rep penalty i believe for many reasons).
Sounds like something the Ruthless AI modcomp from Civilization IV already does.
C) AI can't use 1upt. Well 1upt has to be altered so AI can be on equal footing. Make it 2upt for melee and workers only, range is only 1 upt.
AI couldn't use stacks correctly either. If you keep changing the system so an AI can use it, it will get to the point where there is no terrain and all units are the same.
1) Load melee units into Army ala Civ3.
You didn't explain why this system would be better.
AI respects the HP so more HP means they'll put them on the front lines more and stop leading with range.
Clearly easier said than done.
2) Range has less HP than Melee.
Why? Range already has less melee strength.
The AI will stop leading with range because they're weak.
No they won't.
3) Range has longer hex arc of fire but weaker effect
Range can fire up to 4 hexes away but with weaker effect.
Why?
4) Range cannot fire on close proximity.
Range must have a distance of at least 2 hexes away from its target at all times. It's all in the service of putting melee in front of range so the AI stops leading with it.
Except when it doesn't. The devs have already made several attempts at trying to make the AI lead less with ranged units. The results have been positive. What you're suggesting are radical changes that would result in a redrawing of the AI. If the question is "how would you design the AI" then the answer should be a simple "make the AI capable of using existing mechanics".
It was a theoretical question, not a practical one. However, you're not giving practical solutions either.
5) Range damages any unit in a stack worse due to collateral damages (not applicable to cities - cannot stack in a city)
I'm sure the collateral damage system of Civilization IV made the AI think twice about creating 60+ unit stac-oh wait it didn't.
It increases range units' firepower against melee without increasing their actual HP creating a balance and encouraging the AI to do hit and run tactics with range as its supposed to do.
Except range weapons aren't supposed to be hit and run half the time. Again,
human players (at least, the less intelligent among us) have a problem positioning their units as is. Now you're telling them you can't range bombard up to two hexes adjacent to your unit, making your range units crappy defenders AND attackers. Not only does your idea not make sense to scale (and that's saying something, given how much the Civilization games hate scale), but they don't make sense half the time from other realistic standpoints.
Range units are
not hit and run units except for ranged mounted units. Ranged siege units are meant to be deployed and fire from position, preferably guarded. Archer units are supposed to move, fire, and then fire fire and fire again. Preferably without running into a melee unit. The solution, if you had the capability of designing AI, would be to simply make the AI more capable of guarding ranged units instead of going ass-out-of-the-way to create arbitrary rule after arbitrary check to force players and the AI to perform even
more calculations.
D) AI uses espionage a lot.
1) Spy, sabotage, steal tech, steal maps, steal troop movements
Stealing troop movements would be hilariously annoying and can be countered by players easily. Solution? Move your units randomly. Humans are far more capable of finding a pattern in a sea of randomness than an AI.
2) AI evaluation of other Civs before going to war
Using espionage helps the AI gauge strength in taking on another Civ or even the human.
The AI already (apparently) checks their military strength against yours when deciding to attack. However, you can easily mess with the AI by doing the random-movement thing and make them attack when you're actually strong or not attack when you're weak. Basically, it's a coin flip.
