suggestions about ai strategy for firaxis

[bigvic]

on the hope that the programmers and whoever else is responsible for civ III check these places out, lets sort of set out a definitive, REALISTIC, set of suggestions about what could be done about some of the more lunk headed things tha ai does. i'm thinking more on the terms of specifics, especially militay tactics and strategy, but am open to any SPECIFIC suggestions in any area, political, builds, etc.

let me start it off........

1. teach the ai how to bombard - i don't mind having an advantage, but an ai that knew how to use good combined arms tactics would be really refreshing. there comes a point, before armor appears, when, like in the real world, offense is rather lacking. cannon/artillery/naval bombardment is the only intelligent option. the ai just sends human waves of knight/cavalry in a tediously predictable rampage.

2. teach the ai how to move forces in clumps, with a good mix of offensive and defensive troops. they do this sometimes, but more often seem to attack peicemeal so i, with far fewer forces easily pick them off one at a time

2.a. same with naval forces

3. stop stupid settler crap

4. more sane ai alliance structure - though this seems to have improved in 1.29 - also, a little off topic, some kind of option to tell ai to "call off its dogs" against your friends. miss that from smac/ac, even ctp2, i think.

5. better use of ai armies - i've seen the ai do the stupidest things with them. also, more ai attention to building military academies - i always seem to be able to build one in a high production city, then bust out army after army to crack those tough nuts, or guard that horde of artillery. ai should go there sometimes too.

ok, all for now - sure i'll think of more later - these were just the things that hit me on the current german military bustout i'm enjoying now - panzers rock!

[Cyclotron]

Originally posted by bigvic
3. stop stupid settler crap

Eh? What's wrong with 'em?

[Demerzel]

I'm guessing he's refering to the fact that an AI will send settlers all the way across your territory to build a city in a location which is totally unsuitable.

Annoying and dumb strategy.

[Myrddin]

Let us know what level you are playing at so that we can judge your complaints

Yes the AI does some stupid things but if unless you have concrete programming ideas on how to improve it....

[Cyclotron]

Originally posted by Demerzel
I'm guessing he's refering to the fact that an AI will send settlers all the way across your territory to build a city in a location which is totally unsuitable.

Annoying and dumb strategy.

I can't see that. If the AI doesn't build there, then you will... besdies, even a few gold or science a year is worth a measly settler. Why is this so bad?

[bigvic]

well, i was thinking more of the way during war, especially early in the game, all of a sudden theres a sertler accompanied by a spearman trying to scootch past your horde of swordsmen. sure great fun and all, getting loads of slave workers, but come on......who would do such a thing?

play on regent/monarch - like to speed play, limited on time. higher levels require too much time investment for me - got better things to do. but then i have not noticed a big difference in ai tactics/strategy - just more aggressive imo.....

[Zachriel]

Originally posted by bigvic
play on regent/monarch . . . but then i have not noticed a big difference in ai tactics/strategy - just more aggressive imo.....

Though the AI strategy may sometimes be the same on higher levels, the effect is not. For instance, what you consider piecemeal; that is, 5-6 units attacking, then another stack; becomes 50-60 units attacking, then another stack, then another stack.

Also, on higher levels, the AI will be more likely to treat you as an inferior. They will generally have more powerful armies, so will be more likely to make demands, more likely to refuse "reasonable" trades and treaties, and more likely to attack.

[Zachriel]

Originally posted by bigvic
1. teach the ai how to bombard -

This will probably be difficult to program, as it might be too easy to exploit by capturing AI artillery. If it could be made to work, it would be a great improvement to the game.

[bigvic]

Originally posted by Zachriel




This will probably be difficult to program, as it might be too easy to exploit by capturing AI artillery. If it could be made to work, it would be a great improvement to the game.

yeah, it would involve a ratio of defenders to bombarders, like 1 defender to every 2-4 bombarders, never less than 2 defenders.........

[Jaybe]

My main complaint* about the AI civs is that they often will maximize population without regard to production, WLTKD, or happiness. Size 25-30 cities with most of the cits being non-working and/or unhappy, and having 6-12(+) food surplus. Build more MINES, dammit!

(*Other than combat bombardment ineptness, already mentioned)

[The Puny Celt]

The AI needs to upgrade units more often and/or disband obsolete units. If the AI could be programmed to disband its units in high corruption cities so much the better.

[Arrian]

Jaybe is right. I think that the worker automation orders are more of a hindrance to the AI than the bombard issue. Too much food, too little production.

I wonder if this could be done: Order the AI to grow a city as fast as possible until it hits a wall (size 6, no construction/size 12, no sanitation). Then maximize production. Once the wall is breached, maximize food again until hitting the next wall. Stop growth around size 25. In addition, forbid tile improvement orders that provide no benifit under the current system of government. This is primarily aimed at irrigating under despotism, but also applies so mining cows on plains under despotism. That, coupled with fine-tuning the city build orders (when to build marketplaces, etc.) would make the AI much tougher.

Of course, if that was the case, I would also favor lowering the AI's advantages on the higher levels.

Bombard would be ok... even just convincing them to mass their artillery and use it to beat up on approaching invasion stacks would be good, but I don't want the AI spending too much time & shields building arty stacks... most of the time it would be better served by building large numbers of fast attackers. I know I am. Arty is nice, but is only really important when fighting Cav vs. Infantry or Tank vs. Mech.

-Arrian

p.s. Oh, one more thing... has anyone ever seen the AI use the luxury slider? It seems to use entertainers instead, which is often a mistake.

[jshelr]

My impression, correct me if wrong, is that the AI station defensive troops relatively evenly throughout their cities. We focus our defenses in border cities and respond to a threat with massive concentration in the city being attacked (relying on transport advantages on our own road or RR system to prevent being "flanked"). If the AI did a bit more defensive force concentration, it would be much tougher to roll over their civs. That might be programable although we would immediately try to fake them out.

[Gastric ReFlux]

Actually, the AI coud be adjusted to mine more, but if it were, it would get kerfloogled by something that would need to be programmed.

The AI never, _never_, uses the luxury slider to make citizens content. It will always hire specialists to prevent a city from rioting. Which means that what we regard as extra irrigation is in insurance against starvation. So before the worker AI code could be adjusted, there would first have to be code written to teach the AI to use the luxury slider.

[JohnM2433]

Fix the way the AI evaluates gold per turn deals. As it is now, it seems they will only accept such a deal if the gold per turn is no more than the current income of the party requiered to pay it. They do not take into account that spending rates can be adjusted nor that debts can be paid from the treasury directly.

[ALPHA WOLF 64]

Its too bad the AI doesnt evaluate its units end position. For example, i invaded babylon with multiple invasions, disbanding those cities in order to place my own better located cities. However, these spots are still within babylonian borders so i have to wait another turn to build my cities after moving. The babs plop a city with 1 spearmen right next to a stack of 5 tanks/3 MI. At tthe same time, its scattered numerous spearmen and cavalry recapturing their workers that I had far in the rear building roads to the front. Each of these units were easily picked off. Had those units remained in the cities, I would have had a much more difficult time wiping them out. Spearmen in the modern age dont have much chance to win, but 10 of them in a city will make it difficult to take that city in one turn as each spearman eats up 1 of my limited number of attacks per turn. Holding that one city would have given them another turn to draft infantry and they would have had a fighting chance then.

[bigvic]

great suggestions - exactly the kind of stuff i was looking for - ai tweaks that are realistic, could be done w/o eating up too much space. if it makes the ai tougher - good! i would rather face an ai that is a challenge at lower levels of play because it is smarter than simply mathematically beefing it up at higher levels. hate feeling i'm being unfairly ganked.

perhaps even some kind of multilevel strategy - though i realize this involves geometrically more coding, so.........

but stuff like tweaking ai routines involving attack/defense styles, production, worker utilization, etc., in other words all we are talking abouut now - that is doable, heck, has been done in the series of patches firaxis has released thus far. look at what humans do right and emulate it in code.

[Jon Shafer]

You realize 1 guy has programmed the entire AI? Do you think he purposely left these things out? Do you think he would make an AI that could beat the best humans out there if he had the time and resources?

[Martinet]

Forget the programming, folks! The best ideas are usually developed when one's brain isn't smushed on the coding issues like a cheap eraser. If Firaxis really expected all the fans here to write out the code to implement suggestions, we'd be on the payroll instead of posting here. If you're clueless on the technical side, just use your judgement in making your suggestions, and those with real programming knowledge can tell you what's possible and what's not.

A game like Civilization lives and dies by its gameplay, and in single player the gameplay depends on the AI. There's really no excuse for not improving it so that it is more (or even any) fun to play against.

My personal suggestions for the AI break my own rule and do get into the technical sphere. I think the AI's entire ground rules are flawed, and that a system that makes decisions based on inherent values of things in the game (cities/terrain/units) would come up with satisfying behavior on its own without arbitrary rules. For example, in my last game I took the Greek capital without blocking rail routes. The AI let its most important city battle my whole army with its peacetime garrison plus a few draftees. A few size six cities nearby had strong units available. The AI could not make the decision to save its capital with reinforcements. If the AI was told to place a high value on its capital, it could, after evaluating my threat, decide to move in units from other cities until those cities were insufficiently garrisoned according to their values. Then the AI could make an effort to save the capital without abandoning the rest of its strong cities. Common sense, eh?

[bigvic]

Originally posted by Trip
You realize 1 guy has programmed the entire AI? Do you think he purposely left these things out? Do you think he would make an AI that could beat the best humans out there if he had the time and resources?

yes

i understand it was a monumental task and greatly respect his abilities

as stated earlier, i kind of hope this feedback might possible get back to him, so that he can continue to improve an amazing product, since most of what is being suggested is not beyond reasonable programming abilities

i think he would agree and like the fact that good suggestions are being made............

[Bamspeedy]

There should be a line of code added in that basically says "If a particular city/capital is not losing at least 1 gold or 1 shield, DO NOT build a courthouse". So the AI isn't wasting time and shields building a courthouse in the capital that will do nothing but add more expenses to the AI. Sure, the AI may need a courthouse there if it ever switches to communism, but in some mods, there is a point in the game where the AI is in a particular government and would never switch to a communal form of government in any situation, so a courthouse would be useless.

Have the AI actually aim for ONE victory condition! Right now, the AI only wins by pure luck, as it never really had a specific goal and their victory only came either by given enough time (spaceship, UN, cultural), or by being triggered into war by other circumstances/civs (domination, conquest). It would be great if either at the beginning of the game, or at the start of the industrial era (after they had a chance to build a variety of infrastructure), to choose randomly, or by assessing certain criteria (if they already own 40% of the world, shoot for domination/conquest, or if they have a 2-3 tech lead, shoot for spaceship) and shoot for that particular victory condition. When it knows what victory condition to shoot for, it prioritizes certain things more highly that would get them the victory easier/sooner. Like libraries and universities would be a high priority for a spaceship win, Offensive units would be a high priority for conquest/domination. And would be more likely to initiate wars with anyone who is on pace to beat them to that victory condition.

Prioritize factories higher. One of the reasons that the human catches up to the AI on deity is because it is so slow on getting those factories up and running.

The AI doesn't build enough workers to make up for the AI's inefficiency of choosing the best and most efficient worker tasks.

Never have the AI build Wealth, unless absolutely necessary to balance the budget for the short-term. The AI should be building more military units and conquering someone if they run out of stuff to build. If they aren't building units because of money problems, they should be disbanding those obsolete units and replacing them with updating ones. Quick question: I see there is a flag in the editor for the AI to never build wealth, but doesn't that also prevent them from building banks? Or is banks considered trade?

Be able to change production of something when circumstances change. The AI will almost never change production of anything it is building unless an EXTREME emergency comes up. I always attack the AI in the ancient era, right after they start building a wonder. Because the AI won't stop building that wonder, and so he has at least 1 of his best cities tied up building a happiness wonder instead of military units. He'll only stop building it if I kill off some of his defenders in his capital, than he will change to a spearman and even pop rush it if he has to. And even than, sometimes the next turn he goes back to building the wonder .

Have it's navy travel a little more in stacks. If only 5 ships stopped at your coast and unloaded troops it would be a lot more menacing than 1 ship unloading 1 little swordsman. Same goes for land units. More S.O.D!! (Stack of Doom). There still should be at least 2-3 defenders in all cities, but perhaps if offensive units were gathered in the border cities it would be better? Actually not all IN the cities (in case of culture flipping), but near the border. Right now all their offensive units are spread all around, so if they suddenly decide to send them all to one spot (at the human's cities, for example), they all arrive at different times, thus leading to their units 'trickling in one or two at a time'. If most of them were near the border they would be all together and you would have 20-30 knights to fend off right away instead of dealing with them 1 or 2 at a time.

[JohnM2433]

This is a small thing, but...

Sometimes, when I establish an embassy, I see that an AI's capitol is building a settler, and that the production box is full. The city only has a population of 1 or 2, so settler production is delayed until it reaches 3. This is a stupid move on the part of the AI.

And it should be fairly easy to fix. Have the AI calculate how many turns it will take until the city reaches population 3. Then have it build wealth until the number of turns needed to complete the settler is the number of turns remaining to achieve the required population. If there's enough time before reaching pop 3 to build something else before the settler, so much the better: consider doing that, too.

This wouldn't be such a big deal, except that I think it might cause the AI to lose some production in the early game. I don't think many human players would waste sheilds in this way.

[Wormwood]

Originally posted by The Puny Celt
The AI needs to upgrade units more often and/or disband obsolete units. If the AI could be programmed to disband its units in high corruption cities so much the better.

This can be somewhat fixed by modding your game so that swordsman/longbowman/etc. upgrade into rifleman. They won't always upgrade those units, but it removes the units from late-game build queues.

[Akaoz]

The military AI has three really bad flaws IMHO:

1, It attacks headlong with it's fastest units up front, often putting them on your Rail net without defensive units or artillery to help defend them.

This often leading to the SI loosing it's teeth in the opening engagement.

2, It has no idea how to soften up a defensive position or reduce an assault with artillery/bombarding.

3, When you place a bunch of units on it's rail net it doesn't use nearly enough of it's resources to break the beachhead.

-Alech

[BillChin]

Originally posted by bigvic
1. teach the ai how to bombard ....
2. teach the ai how to move forces in clumps, with a good mix of offensive and defensive troops ...
2.a. same with naval forces
3. stop stupid settler crap
4. more sane ai alliance structure ...
5. better use of ai armies ...

For 1) and 2) I have suggested some ways to implement this. One way is to set percentages for production for each type, instead of having flags for often, rarely, never. This is easy to do if the idea is put in early in the design. However, with the nest of code that exists now, it may be a huge programming task.

For 3) one idea is to have civs expand to a certain number of cities based on time and difficulty level plus a random factor. Once the expansion is complete, prepare for war and keep settler production at a minimum, until another trigger event retriggers settler production. This is how many good human war monger players operate. The AI can mimick this to improve its game. I believe that predictable AI expansion is one of the weakest links. Many opening war strategies are dependant on this predictability. This one seems relatively easy to implement.

I also like BamSpeedy's idea of aiming for a type of victory. Right now, the AI stumbles around, mostly claiming land, without any kind of plan to win the game.
- Bill

[bigvic]

Just remembered something else - minor, but irritating. Leave one little worker out, unguarded and too close to the enemy and he will gleefullysend out a single, valuable, often low defense troop, desperately needed elswherre, to capture the worker. Irritating. you kill the poorly defended dangling worker catcher, and usually the worker to boot. Serves no purpose in 9 out of 10 times but to deplete the defense of the ai. I've gotten to guard my workers pretty well, not so much cause i value them - hell its to my advantage to dangle a few slaves out there to bait the ai into stupidity - but rather i hate to see the stupidity!

again - i realize the ai is the reult of time, effort, and a type of genius i am incapable of, just offering up suggestions to make the next incarnation better.

[bigvic]

ah, something new...
i don't like how the ai makes useless alliances/war

example - everybody's ganking me cuz i'm the aztecs and supposedly "sneak attacked" the limeys early (sheesh!)and after beating off the 13 to 1 ai dogpile on me, notice germany, big, healthy ai power, allys with babylon, bigger and badder, against the amerricans, who are down to 1 city (partially my fault - hehe), on the other side of the world from germany, and lanlocked, and about to go down hard (d@mn - wish i could get one of my 4 armies there in time to grab it).
now this does not bother me cause i'm missing out on a satisfying wipe out, but rather because
a) why would babylon want help from someone who cannot really help (everyone else seems to LOVE the freakin babs)?
b) why would the babs pay for help - which i assume is assumed
c) why would germany give a rat's @$$., cept why not, give me something for nothin? sure!

or the obverse, sort of - when ai's pay good $ to another that has no interest in ganking someone, and instead of ganking the obvious, a power near them that holds the potential to truly enrich and strengthen their long term position, they send a horde of troops half a world away to gobble up untenable cities. i know they don't make a real effort to build a forbidden palace there and make thedisconnected second homeland a viable entity.

[Jaybe]

bigvic, regarding useless alliances, it seems that alliances are a "bonding ritual" which even humans can get involved in. Notice how AI attitudes improve while the alliance is in force, and declines when you finish it? Personally, I could have won a U.N. vote if I hadn't terminated a couple of such alliances beforehand (I now understand better for next time).

Of course, the downside to extended (more than 20 turn) alliances may be corresponding war weariness while you stay in a rep government while everyone else has gone through their anarchies and gone commie.....

[vmxa1]

Originally posted by bigvic
Just remembered something else - minor, but irritating. Leave one little worker out, unguarded and too close to the enemy and he will gleefullysend out a single, valuable, often low defense troop, desperately needed elswherre, to capture the worker. Irritating. you kill the poorly defended dangling worker catcher, and usually the worker to boot. Serves no purpose in 9 out of 10 times but to deplete the defense of the ai. I've gotten to guard my workers pretty well, not so much cause i value them - hell its to my advantage to dangle a few slaves out there to bait the ai into stupidity - but rather i hate to see the stupidity!

again - i realize the ai is the reult of time, effort, and a type of genius i am incapable of, just offering up suggestions to make the next incarnation better.

I do not see a lot of this anymore, about two patches ago. Of course I do not leave workers out to be grabbed, unless it was an overshight or I had some worker on auto (not often).

[Coracle]

Originally posted by bigvic
on the hope that the programmers and whoever else is responsible for civ III check these places out, lets sort of set out a definitive, REALISTIC, set of suggestions about what could be done about some of the more lunk headed things tha ai does. i'm thinking more on the terms of specifics, especially militay tactics and strategy, but am open to any SPECIFIC suggestions in any area, political, builds, etc.

let me start it off........

1. teach the ai how to bombard - i don't mind having an advantage, but an ai that knew how to use good combined arms tactics would be really refreshing. there comes a point, before armor appears, when, like in the real world, offense is rather lacking. cannon/artillery/naval bombardment is the only intelligent option. the ai just sends human waves of knight/cavalry in a tediously predictable rampage.

2. teach the ai how to move forces in clumps, with a good mix of offensive and defensive troops. they do this sometimes, but more often seem to attack peicemeal so i, with far fewer forces easily pick them off one at a time

2.a. same with naval forces

3. stop stupid settler crap

4. more sane ai alliance structure - though this seems to have improved in 1.29 - also, a little off topic, some kind of option to tell ai to "call off its dogs" against your friends. miss that from smac/ac, even ctp2, i think.

5. better use of ai armies - i've seen the ai do the stupidest things with them. also, more ai attention to building military academies - i always seem to be able to build one in a high production city, then bust out army after army to crack those tough nuts, or guard that horde of artillery. ai should go there sometimes too.

ok, all for now - sure i'll think of more later - these were just the things that hit me on the current german military bustout i'm enjoying now - panzers rock!

"tediously predictable". We know.

"Stupid Settler Crap" is the unrealistic and irritating Settler Diarrhea/Rapid Early Expansion nonsense which is now the worst aspect of the game since Flipping can be toggled off in PTW.

All your suggestions are valid. I can think of dozens more; we have gone over them for the past ten months.

Firaxis long ago had the chance to get it right in their patches and in PTW. But they didn't, and they won't. There are to Firaxis enough people willing to pay for their flawed merchandise to NOT justify the time needed to make the improvements. Too bad.