Let's talk ciV AI (Suggestions/Fixes) for the Developers

[Lord Parkin]

- Social Policies are separate for every team member, not combined for the team as a whole. Not sure if this is intentional or not, but it seems wrong. It contributes more to the feeling of teammates playing separate games with their empires than managing a cooperative strategy together.

After some further thought, I think that perhaps the most sensible solution would be to combine overall culture for teams (obviously raising the thresholds for adopting policies as required for balance), but let teammates adopt different policies.

This way, all teammates get the opportunity to pick a new policy at the same time. This would make it a bit less restrictive on the social policy side, but would still give a feeling of coherence and cohesion to playing on a team.

In the present form with separate culture though, there are numerous potential exploits. Just one potential exploit off the top of my head is that one teammate can sit with a couple of cities and mass culture and policies, while the other expands like crazy (through Settlers and war). Then, partway through the game, the civ with lots of the cities and few policies switches most of the cities over to the civ with lots of policies and few cities. I'm sure you can see the obvious problem here.

 

[Lord Parkin]

More problems with teams:

- Confirmed that wonders don't give shared benefits. The Great Library was built by an AI teammate, and it didn't appear to give me their free tech. The Pyramids was built by an AI teammate, and it does not seem to have sped up my Worker speed. I suspect this is going to be the case for all wonders. In Civ4 they had shared benefits in teams, which made sense. What's up with Civ5?

- AI teammates offer excess resources to other civs with preference over your own. This is backwards; in Civ4 they always offered excess resources to you first.

- No control over which City States your AI teammates go to war with. This not only means they can attack a city state you were hoping to ally with soon, but they can ruin your chances to gain future city state allies because of their warmongering ("City States band together" event). The power to control wars with City States in a team with one human player and the rest AI players should ALWAYS reside with the human player.

- The above also makes me suspicious that AI teammates may be able to declare war on other civs without your consent. As before, in a team with one human player and the rest AI players, this power should always rest with the human player.

 

[DaveGold]

In terms of AI movement I'd suggest some very basic things

1) Have a designated city defence unit as soon as a city is threatened. That unit will then bombard from the city until there's a better unit to put in or the city is going to fall.
2) Don't engage with a unit if it's not going to do any damage. On one turn I had three cavalry die attacking my mechanized infantry on a hill doing, 0, 0, and 1 damage. Not a good use of units.
3) As a rule of thumb, don't move isolated units into places where they will die. Clearly harder to code than for a human to analyse but it would help.
4) Don't shoot at city with 1 health. It won't take any more damage. Heal up instead.
5) Don't attack a city with a unit that will always die from the city's first strike.
6) Don't embark units without confidence they can survive.

City States can currently get lucky with their tactics and capture cities. They are very profligate with their units though and should retreat them more often. They should only ever try to attack undefended cities or half sieged cities, as a guide.

I'd suggest that instead of liberating an eliminated player in any of their cities, the capital (only) can be liberated and that capital can become a new military city state. This will leave it in the UN for voting, it will have standard city state behavior to follow, and not have all the complications of a liberated AI nation.

 

[dexters]

Problem:

As I noted in this post, the AI has all the mechanisms to catch /spot what the player is doing but there are areas where it lacks coherence in specific situaitons which I suspect will come up quite often with 'human' player.

The situation:
- Player have Pact of co-operation, and open borders with Civ X, and have signed a secret treaty with Civ X against a much larger Civ Y which is threatening Civ X
- Player moves their troops inside Civ X's territory in hopes there will be an opportunity to jointly declare war against Civ Y or simply to protect Civ X from Y.

The Issue:
-Civ X will fail to look at the power balance, proximity of Civ Y and the deals at hand when evaluation relationships
- Civ X will see close military proximity as a threat (its a potential threat) but fails to consider mitigating treaties
- Civ X will see player aggression against weaker Civs a world away as a threat to itself (also logical generally, but not logical in this case)

Solution:
-Pact of military co-operation against <insert Civ here> as a pact in addition of pact of secrecy.
-This pact:

• Is public. AI will see a global announcement. Human players can also see it if AIs sign one against them.
• Can be signed without pact of secrecy, but combined with secrecy grants a double modifier which allows AI to be more willing to go to war with the target Civ.
• Enhances relationship bonus for gifting units
• Penalty of having your troops near their cities reduced but not removed completely
• Penalty of aggression of on the Civs removed for the duration of the pact
• Strong global penalty for breaking this deal and attacking the partner Civ. (global war? no trades allowed etc.)

 

[Defacont]

I just wish to see AI with a better judgment on their situation.

As in, They have 0 army, and I am the worlds greatest military empire. With 1 city left, half health, surrounded by units, I should be able to open a trade menu, and offer a peace treaty. Instead, they want me to give them stuff. They should be willing to give me everything they have.

 

[Scilly_guy]

I recently played a game where a bug cropped up, its like the reoccurring "peace treaty has ended" bug except it was "Horses to Ghandi have ended", each time it popped up I got back the horses I had traded to him, I didn't notice this until I suddenly had 84 horses. I may have then cheated slightly and sold 40 horses to Ghandi for all of his Gold, I did this over a few turns, I got a lot of GPT and a few luxury resources, I then noticed where all these horses had come from, every turn after that I would get "Horses to Ghandi have ended" +40 horses, at one point I had also traded him 100 horses for which he was really grateful, he would've sold me his Grandmother for them. To cut a long story short I ended up with 2583 horses, clearly a bug.

Now for my observation, the AI doesn't accurately understand the value of resources, it obviously thinks that 1 horse is worth 40 gold or whatever, and maybe 1 horse is, perhaps 2 are worth twice that, but 10? Does Gandhi actually want to use 10 horses?

Suggestion:
AI should calculate how much of a strategic resource it wants.

 

[dexters]

I have decided to revive this thread with current issues. The original issues/problems log can still be found in the OP, hidden under a spoiler flag.

Below are current issues as of [1.0.1.383] As I'm not an AI programmer, these are written in lay person's POV



Economy
Issue:
AI governors believe tiles within a city radius needs improving and continue to suggest workers as an economic build choice even when the empire have enough workers and the tiles around a city are improved. This is likely the source of numerous reports from human players observing the AI having too many workers

• Wasted hammers
• Wasted gold upkeep

This issue can be easily observed with how city governors work for human players.

Spoiler :

Solutions

• This issue appears to have been introduced during one of the recent patches/updates (There was a previous issue where City States would not build workers after their initial builds were captures and pillaged tiles are never repaired). This Could be the unintended result of this fix or another fix
• One possible solution is to fix how AI decides if tiles around a city needs improving and crosscheck it with number of workers on hand

From my AI Puppet Empires - the Paper Tiger thread
Issues:

• AI will not annex conquered cities ;
• 'Gold' Focus is often sub-optimal for larger cities which could build gold producing buildings/science and culture buildings at a faster rate on 'default' focus. Getting multiplier buildings up sooner is more effective! High production/pop cities could be put into better use producing MORE culture annexed; and units
• Large AI puppet empires are often susceptible to 'total collapse' if their main armies are defeated or occupied elsewhere (say 2 front war with the 2nd front on the other size of their large empire) and their productive core is attacked on the other side. Leaving the rest of the empire prostrate and undefended. This quicky destroys their main armies and they are left feeding in a few units at a time per turn.
• Capitol AI will routinely place a Civ's 2nd and 3rd capitals into a puppeted city, often of decent size. But the AI seems unable to deal with this - ie: notice their capital is a puppet.
• The AI appears oblivious that they have puppeted cities

Solutions:

• Allow AI to annex and rush buy courthouses in high pop/production cities, especially those in resource rich/food rich/production rich areas. This won't be universal, but they will do it for SOME of them. preference could be form size/growth potential/existing infrastructure/distance from capital (for symmetry)
  
• Target cities that symeterically balance the current core, to spread out production over more nodes in a large empire
• Do the annexing in the 'peace' phase after a conquest, not in the heat of the battle - The idea is for 'consolidation' rather than the current issue of a relatively small core dragging along a large empire that is completely dependent on the core to produce units and non commercial buildings

Tactical AI
Issue:
AI will rush in a defender and worker on a red-lined city which would have fallen regardless.

Solutions:

• Check City HP before sending in Sierge units to defend
• Check local enemy superiority to decide if the risk it work taking. And if enemy is locally superior, check to see if units (air/land/sea) can be combined to weaken attackers
• The idea is not to have the AI NEVER play this gambit, but to do so prudently.

Issue:
When attacked and having their field armies destroyed, AI will feed in units piecemeal, instead of retreating remaining damaged units and mustering a force to counterattack. This hastens, collapse of AI empires.

Solutions:

• Related to my earlier point about annexation of puppet cities, AI needs to annex cities and develop those cities to increase 'real' production proportionately with growth of empire; as a corollary, Annex cities as needed when production is lost elsewhere.
• Evaluate threat level of invasion and formulate a force required to respond. This will be a variable bounded/balanced by the by gold rush buying units/production and the time required.
• Engage in disruptive attack behaviour by attacking flanks, and supply lines and not directly at the next besieged city. A Human player can usually redirect besieging forces to destory enemy units. The working assumption must be that a counter-attack is to disrupt and not to directly attack invasion forces.

Victory
Issue:
AI have a sticky cash problem - (they like to hold on to their cash far too much) This has led to the following scenarios.

• Despite having a 2.5 to 1 gold advantage an AI willingly let me buy up all the city states 1 turn away from UN victory and win a game I would have easily lost
• As their forces are whittled down during a war, AI gpt will often improve and gold reserves will go up. It is not uncommon to see an AI with 10,000 or 20,000 gold reserves sitting like a lame duck as their cities fall around them. Though this may be due to the fact the AI lack enough unpuppeted cities to rushbuy those units.

Solutions:

• Set gold reserve ratio based on perceived threat level over both long and short term. While holding a large stockpile is a good long-term strategy, not using this cash becomes a short term liability when those long-term problems become short-term problems in the situation of say Total War or a clock running down to UN victory
• Corollary to my previous point about city annexation*, city annexation is key to create more possible nodes of production. This will enable AI to use this cash to rushbuy infrastructure and units. And AI that can rushbuy units in its core + annexed cities will always have more units, and ultimately more usable production (that the AI civ can direct as needed) than the same AI relying on its core to support the rest of his empire.
  
  *As noted earlier, annexation needs to meet a criteria (growth potential, size etc) and this isn't a suggestion for a carte blanche request to have the AI annex most of their cities (that is bad too)

 

[apocalypse105]

One thing that ols bugs me is the AI thinxs its a good Ai to declare war on someone who is twice there power (for example I have rifleman and he has pikeman)

All my rifleman are on the left side of the map he then send all his pikeman to a city where are no troops and then declare's war and get destroyed by the counter atack


However the msot anyoing thing is if you pick production focus in one of you're cities the AI thinxs its a good idea to starve you're people to dead

 

[Zhahz]

Combat AI needs help.

The rest of the stuff people want is really pushing it, IMO. I'd rather them not screw around with anything until the combat AI is top notch - and that could take some doing.

City States surely don't need to be robust participants in combat. They're an occasional irritant now, which works. They're city states, not civs.

I hated the AIs in Civ 3. It wasn't that they were ruthless, they were completely random, easily bought with gold, changed direction/allegiance every other turn, etc. Total garbage.