AI questions

[DoctorGonzo]

quote: (BTW Like I keep ranting on about in other threads, I think the AI should be open source. It could have a custom language, a bit like Civ2's events.txt, but it could very maybe use VB, even though that is slow - but at least it's easy to learn/program in.)

I agree with the need for an Open Source AI, however using a platform specific language is a bad idea. Mac and Linux users will miss out. They already have to deal with things like not having level editors for Unreal Tournament because someone thought it would be great to do it in the most Windows specific way they could.

I would advise them to make every effort to stick with the most platform independant solution possible.
[This message has been edited by DoctorGonzo (edited June 18, 2000).]

[UltraSonix]

Ok, how about creating a AI editor instead of a language. So using an interface similar Starcraft-editor's triggers-and-events way of setting up stuff, even novice users will be able to customise their own AI if they want. The same editor could be used to alter game settings/units instead of just altering a text file.

------------------
No, in Australia we don't live with kangaroos and koalas in our backyards...

[Ralf]

You can easily "expand your way to success" by shoving out buckloads of city-founding settlers i a row, early on in the game.
20+ settlers or more, founding just as many cities + another 20+ city-area settler-developers.
Effective, perhaps - but is it FUN?

Its certanly aint very realistic in historic terms, thats for shure. Also, if the AI should have the same early ultra-fast city-founding strategy for each of the (upto 7) computer-civs; the game would probably grind to a halt, in later endgames.

I have the following four expanding-rules suggestions. They should enable a much more realistic game, and allow the AI to compete more easily (i hope):

A/ Max two "empty" cities (= without any city-improvements) at any given time within that empire, can be allowed.
The AI/Human Player HAS to build temple in at least one of above cities, in order to continue founding a new city.

B/ Any city NOT road-connected with some/all of the other cities, gets an proportionally stiff corruption-penalty each turn.
Lack of road-connections to other cities should also give a more noticeable resource-/science-penalty.

C/ A big sized empire of 25 cities or more should be an unstability-factor in itself - increasingly prone to split-up federation-attempts. This should be especially true if the large empire is far ahead in terms of science-/production-/economy- and military might, then the other civs.

(It shouldnt matter if the empire is well maintained: these split-up tendencies should appear anyway, if the empire is self-sufficient and powerful enough (to far ahead the other civs).
The split-ups federations should consist of min 20% - max 40% av the empires cities. A real groundshaker, in other words).

D/ Also (important); the cities in small civilistic-perfectionist 8-12 city empires should in return have less problems with building huge 20+ mega-cities.
Cities in large 20-30 city empires on the other hand, should have increasingly bigger problems with developing indevidual mega-cities.

[SmartFart]

THE ART OF WAR: Ancient Military

-there is no effective war party without Crusaders,Pikemen,Catapults,Settlers and Diplomats
-all militarty units must be veterans
-pick your target carefully before attack,no use of taking a city which you cannot destroy or hold for a long time...it's a waste of time and resources
-use Diplomats to avoid zone of control
-use pre-worked Settlers on rivers/roads to build instant fortresses (for those who aren't familiar with this cute little trick: mine with your Settler 2 turns and click on it to stop working...now your Settler can make instant fort if you have construction)
-stack your Catapults/Crusaders with a Pikeman and move them to most deffensive terrain around your target city
-never left your invasion force unstacked
-explore distant shores with Triremes loaded with a Diplomat and one 4-legged unit. That will hurt early Republicans who prefere to leave distant cities undefended because of loss of shields
-play the cards you currently have : if you have Lighthouse - build lots of triremes, Sun Tzu War Academy - go conquering ASAP, The Great Wall - expand like hell and build defensive units or build barracks and attack civs w/o Sun-Tzu or build a city inside of your enemy territory, fill it with units and go conquering from there (WARNING: be sure you blocked all ways to enemy diplomats )
-defend your cities with at least 3 units,one Catapult is highly recomended,2 vet Pikemen are a must
-make as much embassies as possible,look which cities builds wonders and try to sabotage its production with Diplomats

Of course,all aforementioned depends on early expansion.
On Deity,AI have alot of benefits (smaller food/production boxes,more arrows,knowledge of entire map and cities,lesser unhappiness,etc.).

If you manage to teach AI how to use given benefits,you're doing a great job.

Regards.

SF

------------------
No i ain't doing much,doing nothing means alot to me (AC/DC)

[phunny pharmer]

When I attack and conquer a city, I use a large tresury to buy my attack units, so my attack 'booms' with the number of cities that I take. Not only is this historically inacurate, but the AI gets swaped when I do this.

I bring this up because I have never seen the AI do any sort of constructive attack like this. They should coordinate their attacks. Either the whole building system needs to be changed when a city is conquered, or the AI needs to know how to do this too.

A possible way to get around it: make cities that are captured by enemy civs should not have any production for ten turns. Furthermore, a reteating/defeated civ would probably sabotage factories, power plants, etc. These improvements should be considered 'off line' for these ten turns. Also, if the city walls are captured, they are probably damaged, and the enemy will know all the gates and weaknesses of the city just conquered. Thus, the city walls should only give a 50% bonus for these ten turns.

[Smash]

The AI needs to improve in the trade area the most.It just doesn't seem to maximize trade arrows at all.

Cities are built haphazardly.I can't tell you how many AI cities I have seen that are missing 2 nearby specials.It need a higher priority for resource specials.And it should be by resource type rather than resources in general.

Food is a problem.It always goes for food squares over trade squares.It will build Collosus in a desert.Cope's Observatory in a city that will never produce more than 10 science beakers.Caravans don't seem to be a priority and don't follow any plan if they are sent.

Tactically it is not good,but it would be better if it didn't attack armor with elephants.This is where trade comes in.Perhaps putting a higher priority on trade arrows rather than food would help.

The ai does better at spacerace than conquest(for reasons mentioned above) but if I didn't give 80% of the techs it would be stuck in ancient times forever.

Despotism.The ai takes along time to get out of it.It needs to be told to get out of Despo ASAP.And when it does,it seems to change governments at the drop of a hat.Its almost as if it arrives at a decision 1 turn then arrives at a totally different one next turn because some values have changed.It needs to be longer term in it's strategy decisions.

Growth-yes it needs to expand more.And grow cities larger.It is pretty rare I see AI cities beyond size 12.Most are size 8.Although the AI has irrigated and farmed every square in sight.

I don't know how to do it but if the AI could recognize which improvements to build in which city.It will build a factory in a 5 sheild,30 beaker city while a 20 sheild,5 beaker city is building a university.

[Phutnote]

I find wiping out computer civ's fairly easy late in the game. This is my best strategy:

1) Build up average size civ. Avoid wars. Exchange technology. Build caravans like crazy to build good economy.

2) Connect civ to others via railway, or obtain/build bases/city within 1 turn of transport ship range. let them develop their terrain before invading, so it's less work fixing it up for me.

3) Get the robotics advance so I can build howitzers. Sometimes I wait till I have stealth and can sweep the skies. The only units i really need are howitzers, fighters, transports, a few spies, AEGIS cruisers, and a few armour units to clear out light units from cities (missiles, etc), without wasting howitzer attacks.

4) Build a lot of howitzers. Mass them at the border. Get the SDI defense if nukes exist. Build them in every city, get lots of cash to buy em.

5) Attack an enemy civ. Take 10 or so cities, or their entire empire in one turn. Make peace after taking the last city I can. Buy SDI defenses in all cities.

6) Wait for units to repair, then attack again. Can destroy most computer civs at King level in 1 or 2 turns, especially if only on one continent.

Phutnote

[DaveV]

I agree with the early expansion. If the AI could be taught to ICS, that alone would make it dangerous.

More on placing city workers optimally: the AI doesn't seem to value trade arrows - it will put workers on forest before silk. It will also produce a huge food bonus in a size 8 city with no aqueducts. I think the AI should emphasize shields instead of food, to capitalize on its production advantages.

My biggest advantage over the AI is that I have the initiative over the AI for almost the whole game. Two tricks that would help the AI: 1) fortifying a couple units on strong defensive terrain in the heart of an enemy empire. 2) seaborne invasion with 2-movement units (like the barbarians do now). Land a stack of crusaders/dragoons/howitzers on a clear square next to an enemy city, then stomp it.

[Father Beast]

quote: Originally posted by Ralf on 06-18-2000 04:10 AM You can easily "expand your way to success" by shoving out buckloads of city-founding settlers i a row, early on in the game. 20+ settlers or more, founding just as many cities + another 20+ city-area settler-developers. Effective, perhaps - but is it FUN?

Ralf,
I posted the best fix (IMHO)for this problem in the ultimate ICS thread. (can someone give the link, please? I don't know how.)(ell, I huess I do now.)
it's basically to limit the tiles being produced to the number of citizens. no free city square.

of course if ICS is still a factor, then the AI should take advantage of it. I can just see being plastered by the mongol hordes while I'm trying to perfectionist build.
Edited numerous times while learning how to link and fixing the link.
[This message has been edited by Father Beast (edited June 20, 2000).]
[This message has been edited by Father Beast (edited June 20, 2000).]
[This message has been edited by Father Beast (edited June 20, 2000).]
[This message has been edited by Father Beast (edited June 20, 2000).]
[This message has been edited by Father Beast (edited June 20, 2000).]

[StormLord]

Hi,

Nearly every major glitch has been mentioned, except this one, which seems to be a unique strategy invented by myself:

Use aircraft/helis for defense! They're fast and effective. If the enemy lands with a small force, bomb them. If they land with a large force, it'll probably be a cheap one, so send in the helis... they can attack several targets per turn.

By the way, there could be more difficulty levels with less difference between successive steps. I beat the Prince-level AI with one hand tied behind my back, while I seem to lag behind at King difficulty.

One last thing... why's everybody so obsessed with bottlenecks? Ships sail around them without trouble. If you had enough naval power to get to the enemy, you have enough naval power to get around a bottleneck. It's as simple as that. (Except on giant linked pangaea style maps ofcourse )

Bye/See you in Civ!
The StormLord ( maybe it should be WindLord, because the wind brings the storm...)

[BustaMike]

I usually stick to the SMAC forums, but since I'm planning on getting CIV3 even if it only gets decent reviews I figure my input can't hurt. I have beaten the crud out of both CIV2 and SMAC on the hardest skill levels with my eyes closed so I have a good idea of where the AI falls short.

I have also seen a lot of good comments on the AI from a CIV standpoint but almost none from a SMAC standpoint. For this reason I will try and give my input from a SMAC standpoint.

First off let me say that the AI is much much weaker in SMAC than it is in CIV2. This is because SMAC has many more choices to be made than CIV2 did. I can only assume that, like SMAC, CIV3 will also be much more complex than CIV2. This means that it's going to be a much bigger chore to make a good AI.

Anyway here are my biggest beefs with the SMAC AI:

[*]Supply Crawlers:

The AI has absolutely no idea how to use these properly and rarely if ever builds them. This allows a human player to gain a ridiculous advantage though the proper use of crawlers. Crawlers are the second most important unit in the game (1st being formers). They allow a city to gain production without using a worker and can be used to rush prototypes and Secret Projects (the same as caravans and freights in Civ2). If the AI could understand how to use these even a little bit it would gain a huge advantage over an AI that does not.
[*]Pop Booming:

The only time the AI will ever pop boom happens totally by accident. A pop boom, for those of you who are unfamiliar with the term, means that your city or base is increasing a size every turn. In SMAC this is done by getting +6 growth, in Civ2 by getting golden ages in Republic or Democracy. Pop booming will create quite a sizeable advantage if timed correctly and I know several people who absolutely refuse to do it any more in single player SMAC because it puts the AI waaaay to far behind and the game is no longer fun.
[*]Support costs:

I have regularly seen the AI build so many units in one base that it no longer has any minerals free to build anything else. This is dumb. This happens a lot in Civ2 also.
[*]Terraforming:

The AI is an idiot when it comes to terraforming. 'Nuff said
[*]Attacking:

I know that this has already been mentioned, but it is important. The AI has no idea how to launch an attack. Instead of sending one powerful attack force it sends a steady string of units that cannot defend each other and can easily be wiped out by a much smaller strategically placed set of troops. Several times I have defended against well over 30 units with two or three quick units and a monolith to heal every turn at.
[*]City Placement:

The AI in SMAC will regularly spend a ridiculous amount of time moving a colony pod to found a city near a landmark rather than founding a perfectly good base close to home. This eliminates greatly any turn advantage that could have been gained by having more cities quicker and also makes it extremely difficult to defend the new base because it is by itself. The faction in SMAC that is most notorious for screwing itself this way is the pirates. The pirates will always run as fast as they can toward the monsoon jungle to try and get the nutrient bonus; however, this is in vain 90% of the time because if someone elses territory already includes the jungle then the pirates cannot harvest any resources from the jungle because it is outside their territory.[/list]
Anyway those are my biggest gripes. I know that probably only a few of them will actually be relevant to CIV3, but I hope they are helpful.

------------------
"Luck's last match struck in the pouring down wind." - Chris Cornell, "Mindriot"
[This message has been edited by BustaMike (edited June 20, 2000).]

[Sparky]

testing. . .

[Sparky]

Ah, here we go. I'm an experienced civ player -- played civ 1, civ 2, SMAC, and everything in between. I've also observed the AI throughout each version.

The Civ 2 AI improved much over Civ 1's. Civ 2 had costal blockades and smarter pathfinding, as well as nukes, majing the game harder than Civ 1. Still there's much room for improvement:

-Railroads. The computer doesn't know where to build them. The AI doesn't connect specific cities, and instead wastes precious engineers' turns putting down rail and making a sprawling mess. The result? Land invasion is EASY. Almost every square my forces are on is connected by railroad. If I can bomb their cities into being unoccupied, I can easily move all my troops a vast distance inside their territory without losing even a single movement point. Even if their cities are still occupied, I can always find a way to move a howitser from one end of their homeland to another, just because prevalent railroads make movement too easy. Recommend: build roads everywhere for trade, but only connect cities with minimal railroads. Make a web of roads so that you can't get to one city without going through another, or make a wheel-and-spoke style layout where a central (well-defended) city hub has railways leading to every other city making a large circle. An invader must control the central city to make the rail infrastructure useful. Also, don't build rail on coasts. Makes landing from ships very easy: any two-movement unit can land and use the rail to move inside enemy territory within the same turn.

-Forts. This issue is serious. The AI has no idea how to use them. I assume in Civ 2
the idea was for the AI to build lots of forts since it built none in Civ 1. The combination of fortresses and rail indeadly to the computer. Here's what I always do to the AI: take a large force of offensive units (ex: howitzers) and add a single defensive unit. The AI builds tons of forts withing its homeland, none guarded very well. I occupy a fort next to the city I want to capture. I fortify my defensive guy and stack all my howitzers. The AI rarely makes a serious counterstrike, so I can always count on my defensive unit surviving (even if he doesn't, since my stack is in a fort, it's only a single loss). I then use the railways to attack with my entire force at once without any movement penalty. If I fail to capture the city, I retreat with my damaged forces, heal them in the fort, and attack again! Never fails to crush the AI. The remedy: The AI needs to understand the danger of a large army in its midst -- it needs all-out war to destroy them. Sever the railways, bring in troops from other cities to reenforce the garrison, counterstrike the enemy force. As for forts: build fewer so that the engineers aren't wasting their turns. Also, recognize strategic locations and make bottleneck forts in mountains at the edge of the empire, so that the enemy doesn't get in in the first place. Always occupy forts with a good defense, not a single unit. Destroy any old forts that could be used against you.

-I use the above strategy to crush the AI on diety in the WWII scenario. Build many howizers. Load on transport. Sail to Netherlands (there's a single fort on the coast next to Amsterdam). Unload in fort the same turn. Fortify 1 turn. Use German rail against them. March into Berlin. Never fails.

-The AI needs to understand total war. Sometimes, during a devistating attack on your base, you ammass all your men in a last stand against the enemy, pulling every rifleman from across the empire to stop the enemy in his tracks. The AI never understands this and lets the human invade with only marginal resistance. The only exception I know of is the Zulus. They fight like mad to protect their homeland. The AI must recognize *real* danger (full-scale invasion, massive aerial bombing, etc.) and do everything possible to stop it. Anything less and they'll loose.

-Expansion. The AI has trouble with cities. It should build them a little farther apart, four spaces between, instead of two or three. Also, it needs to expand ealry-on faster. I focus on occupying my continent as soon as possible. Then I build infrastructure after all the cities are down. This may not, however, be appropriate for all civs: the Chinese and Babylonians are usually pretty contained, while the Russians sprawl.

-Freight makes great spies. I wander around enemy rails with "peaceful" freight to get a look at enemy territory.

-The AI must coordinate tricky massive air/land/sea attacks simaltaneously to be effective. I emphasive MASSIVE because huge invasions of just a few unit types (howitzers) seem to work best. I say tricky because the AI needs to understand the element of surprise: attack the rea of the island, paradrop *inside* the homeland, capture all the undefended colonies, do blitz on the far-away city that just made a Wonder.

-The AI never paradrops outside cities. I often drop troops in all the surrounding city squares cutting off food, trade, reenforcements. Then I invade with my paratroops already on the ground. Also, seiges are important. The computer never approaches cities with hordes of defensive units to surround it and starve it into submission.

Thanks for reading

[markusf]

The AI is really bad....


Build 30 cities don't get into a war.
build massive amounts of market places/banks/stock exchanges/
Go into dem
80% tax 20% luxeries and then BRIBE 7-8 cities per turn. Once your making 1500 cash per turn none of the AI's can stand up to you. I have taken out 15 cities in one turn before. Bribing is just way to powerfull.

[Sparky]

-Ben Larrimer reminded me: the AI does not use cruise missiles enough. Yes, it strikes ships, but it never fires at cities. If I don't have the means to actually wage a war, I build missiles and hit cities over and over, softening them for invasion.

[Sparky]

Build MANY bombers. If you have enough attacking at once (8), you can attack a city from every one of its surrounding squares. This cuts off reenforcements and avoids the problems of stacking -- if a fighter shoots you downs, you don't loose all your planes. The AI always attacks with multiple bombers in the same square -- a single fighter of mine can shoot down all their stacked aircraft and (if repeated) destroy their entire airforce within a few turns.

[Gord McLeod]

It seems to me that there's one big problem with most of the suggestions I'm seeing in this thread so far. They're mainly very valid points, but they're also very limited. I disagree that, for instance, "The AI" should use more of "unit type X" for (attack/defense/exploration).

What *I* would like to see is more diversification in the AIs. SMAC made some good improvements in this area - there were somewhat distinct personalities at work in the way the AIs would behave.

I would like to see some civilizations that take on different roles. Some should be insular and xenophobic, building to a point and jealously guarding its borders, almost never venturing out to attack others, instead concentrating on holding what they have and ensuring that nobody, NOBODY gets in. Others might care more about moving quickly to gobble up territory, while avoiding battles. Some might be attempting to build a giant world-encompasing empire.

What I'm getting at is that not all civs should be out to rule the world, not should all civs be attempting to play a war game. Civilization is not, to me at least, a war game, and not all cultures/countries/civilizations should have military aims. Of course for those who do prefer to play it as a war game, you should be able to influence which types of AI personalities are used.

[jpk]

One thing that happens often in CivII for the Macintosh is that AI units wander around aimlessly. This is especially bad when the unit is a settler. Build aroad, irrigate, mine. Almost anything is better than wandering around.

The AI does not appreciate the advantages of a massive amount of caravan production. Try to get every AI city have three trade routes.

I agree with comments about dropping piecemeal attacks in favor of massed attacks. Since the AI is unlikely to ever be as good in a battle of maneuver as a human, a massed attack is probably the best strategy for the AI. Build twenty units and then attack from as many directions as possible.

Yesterday the AI wanted to spy on me. In order to get into position it stacked 5 diplomats on a nonfortified square. One hit got rid of all of them. If they had been spread out I would have had to use all the nearby fire power for one turn to get rid of the threat.

[jsorense]

Hi Gord!
Long time no see. (-:

------------------
To The Hijack Police: I don't know what you are talking about. I didn't do it. I wasn't there. I don't even own a computer.

[Rusty Nail]

There are many good suggestions on this thread with which I concur, but the best one, which in a sense incorporates all the others, is to offer a really good AI editor. Allow Apolytons and others to cutomize the AI behaviour at no charge. Let them do the work for you. This could revolutionize the concept of AI. To get optimal benefit of the many willing programmers, it is important to define a really wide range of system indicators so there is lots of raw material to work with. AOE does this to a very limited extent and several improved AI algorithms are available on the net.

[Sparky]

Very important: I often wage "scorched earth" warfare. Occasionally an enemy (often the French) decides to mess with me. If they attack first, I ALWAYS make them regret it. I wage scorched earth warfare: I don't capture their cities - I destroy them.

This is EXTREMELY effective. Here's why: the enemy looses a city and for all intents and purposes, cannon get it back. At the same time, the city makes me a ton of money to wage war against the enemy. Best of all, on huge maps this is a method to conquer the world without running into the problem of being overwhelmed by too much city management.

I capture a well-developed, world-class city in their homeland, around size 14 or so. I immediatly have the city build defensive military units to keep the city from enemy hands. I turn ALL citizens into taxman (taxmen = muchos $$$) and laugh at the starving people. Every turn I sell the most valuable city improvement, starting with factories. In just a few turns, I can make an enemy's most powerful city into a worthless villiage. Even if they capture it back, it'll be useless -- they'll have to divert needed war resources to rebuild the city.

In summary, the enemy is doomed to lose. I leave nothing worth recapturing, and devestate the enemy nation. I also avoid the time-consuming act of massive city micromanagement as my forces conquer the world.

I believe that the Russians and Germans in particular should use this tactic in Civ III.

[TheLimey]

AI Scripting: This has already happened in CtP with the 'Awesome AIPs' project, which I took the very first steps in.

AI list processing files are very complex... but rules allow you to do a lot, provided enough is 'exposed'. Many, or maybe even most, of the AI problems in CtP have been solved or reduced to a large extent by this work.

Having a dumbed down AI editor is something that can't compare to a conditionally hinting AI script, as you'll find in CtP.

The problem is, you have to build the AI from the ground up with this intent, and another problem, is documentation; it needs to be provided to start with (as it wasn't in CtP,) to avoid the months of painstaking research to work out how stuff was done, and how it works.
[This message has been edited by TheLimey (edited June 24, 2000).]

[Glostakarov]

Things I do to the AI players:

1) Treachery: Sounds basic? It is, but I don't know how many times I've made peace with a computer player, given him a few turns to work on other things while I build up and position units, then swamp him. Declare peace, repeat.
2) Explotation of Zones of Control: There's a lot in this, the nastiest being using diplomats to allow military units past enemy zones of control to attack the cities directly. However, I also on occasion use a nice "creeping offense" with two lines of units staggered on diagonals, moving forward by rows to make a nice dragnet the enemy can't slip through. This can also be used to defend wide spaces if you have the units to spare.
3) Use cities to claim large amounts of territory, then fill in the gaps: Send out piles of settlers, obviously, but also scout heavily to find where the computers are slowly expanding, then put cities right in their path. Build back toward your original cities from there, and spank any cities the computer builds in your expansion area.
4) Settler Bonanza: Don't you just love it when the AI builds two cities right next to each other? Not just within each other's radii, but right damn next to each other. Take them both, and have the smaller settle itself to death. Move all the workers onto the best production/least food squares and build settlers until it's gone. Use the other city to support the last couple so you don't lose them when the city dries up.
5) Use Crap cities as unit supporters: This is an offshoot of ICS, really, and only worth it in the early to mid game, but as you space cities with the lovely 21 square thing, there are going to be some unused squares in-between. I like to drop a city in there and limit it to just the squares that are not in the range of other cities, give it almost no improvements, and mine the squares it uses. Then, I bring units through and set their home base to the crap city, leaving my other cities their full production.

Things I do that the AI should too:

1) Roman Roads: When I send an attack force, I have at least 2, generally 4 or more settlers along with them, creating a nice road/railroad connecting my cities with the AI cities so my reinforcements can come right up the pipe. This was really a hoot in SMAC, where my formers would raise me a land bridge to the enemy if they were close enough.
2) Air Power used defensively: Keep a couple bombers in cities at the edges of your empire and patrol with them, knocking down enemy stacks before they get close enough for your ground units to worry about. Keep fighters in the cities themselves in case something gets through.
3) Bombardment/Sea Power: I take control of the seas early, sink anything I don't like, and circuit enemy islands attacking anything I can reach, even if I don't plan on attacking soon.
4) Use the Terrain: If an enemy city is surrounded by 3 mountain squares, 3 plains, and 2 water, and I am bringing up 3 attacking units, I'm going to want them in those mountains, even if I have to build roads and dance with diplomats to do it. Defensive bonuses are a wonderful thing.
5) Large-Scale Offensives: Land 2 or 3 full transports at once, then immediately send them home to where more troops should be piling up ready to board. Space them and spread the units to attack several cities at once. If the assault is proceeding by land instead of sea, make sure everthing is heavily railroaded enough to keep the reinforcements coming steadily. Try to take the least heavily guarded cities first, reducing the enemy's industrial capabilities while increasing your own.

[Lodi]

The thing that has to be implemented in Civ III's AI is the aforementioned by Gord McLeod in enemy civs overall strategies. Instead of all civs eventually going to war with you there should be other means of your enemies trying to win the game like having the other civs develope their civ without aggression. This never happens since Civ II and SMAC favored the warmongers. Why the hell can I not build a civilization in Civilization games, but I am forced to build a war machine?

------------------
I am the Roman Emperor, and am above grammar.
-Emperor Sigismund

[Gord McLeod]

Hey there, jsorense. Looks like I'm not the only migrant from the old Firaxis forums.

A further notion that crossed my mind regarding my earlier posts about different personalities in the AI - some might argue that it's unrealistic to have AI personalities that will never go to war. I'm not suggesting that, but I think it'd be very interesting to have AI personalities that display dislike for another civilization in alternative ways. For instance, say you have a nation that's putting intense effort into becoming a trade empire and have become a power of the world as a result. They might express dislike for a more powerful rival civ by cutting off trade with that civ, potentially even attempting to influence other civs into ceasing trade with them. Alternative types of warfare such as this, economic warfare, should be well represented in the AI 'arsenal'.

[Ralf]

I have posted a thread which named "Arguments why its nearly impossible to program an "almost human" AI"
The thread contains several new posts from me, and reactions one those, but its to big to be replicared here.
Go to:
http://apolyton.net/forums/Forum6/HTML/001531.html

My main idea is about Pre-Made Templates - how and when to use them, in order to free up the AI as much as possible, so it can be used more effectively elsewhere.


[This message has been edited by Ralf (edited June 25, 2000).]

[Matthew]

I'm not a computer programmer, but I realize that there is onlyso much one can practically do. I think some of these more complicated suggestions would go well in a strategy guide thread, but not here. I do have a few suggestions which I imagine would be fairly easy to impliment.

Things the AI lets me do:

1. Stack some single movement units next to a city, and let them live to attack the next turn. I don't do this much, but when I do, I often get away with it.

2. Likewise, if I have a large group of mobile units within 1 movement point of the AI city, the AI should really kill them off, peace or no peace.

3. This has been mentioned before, but worth restating. The AI should not agree to peace or a ceace fire right after losing a city. I've used this to take over some very powerful civs with relatively little military force, one or two cities at a time. I take a city I couldn't possibly hope to hold, then ask for a cease fire. or better yet, more often than not the AI will ask for peace immediately. Then on my next turn I can often get the AI to declare war on me by demanding tribute, so my reputation goes unscathed. It would be better for the AI to at least wait for the end of it's next turn to make a cease fire offer, after it has tried to counter attack.

4. The AI is almost always far to ready to give tribute in the form of money or technology. Sometimes one can build almost all the early wonders just by buying caravans with money supplied from tribute. Tribute should be a factor in the game, but shouldn't be so imbalancing.

5. The warrior defense. Often my entire civ is defended primarily by warriors, and maybe the occasional phalanx. No way in hades a civ with monotheism should let me get away with this one. (Don't get any freaking ideas, RAH. this is against the computer, not you.)

6. The AI should probably never be willing to give you a wonder tech if the wonder hasn't already been built. Possible exception for allies, bronze working, and map making. Otherwise techtrades might be extremely limited early on.

Things the AI should do.

1. Expand. Perhaps not do much more than expand until a certain number of cities is reached. For balance there should perhaps be some civs that aren't so expansionistic, but some certainly should be. Maybe have a couple civs plant roots, build 4 cities real quick, and start working on wonders real quick, while 3 others try to get 12 cities down before they build much else. Gobble up some land before I get to it.

2. Government.

There's this thing called Monarchy. It's kind of important. There is probably no such thing as a decent human player that does not make either monarchy or republic his first priority. The AI should try to get out of despotism as soon as possible as well.

3. As mentioned before, the AI should not shun putting workers on trade arrows. every turn it should evaluate a city's worker distribution and adjust it according to maximal return.

4. Not a major point, but why investigate my city 5 times on the same turn?

5. I've seen AI cities hang around size 8 or 12 for very long times. Perhaps make construction/sanitation a high research priority once a substantial fraction of AI cities reach size 6/10, and prioritize aqueducts/sewers at size 7/11, provided there is enough food production potential and space availible.

6. I've never seen the slightest evidence of an AI usingit's hordes of goldto speed production. Perhaps it should put the gold to some kind of use.

7. After the first few cities have been built, and the AI is out of despotism, forests and hills should recieve a higher value as potential city sites provided there will be some minimum food availibility in the city radius. This would make a city much more difficult to take.

8. Keep single movement attacking-defending units, like legions, preferentially on defensive terrain when in or near your territory.

9. Feudalism should be a high priority for civs nearest the human player. Also, masonry and city walls.

Things that it might really suck if the AI did.

WARFARE.

1.I know that it's probably hard to get an AI to assemble a battle group, but this might be a simple solution. Give the AI a build option of battle group in a coastal city, provided there is some sea route from that city to human territory. This battle group option should consist of perhaps 3 state of the art vet mobile units, best availible defensive unit, a diplomat, and a boat. ( or more boats if you develope qualms about an AI being able to overload a boat). To speed production perhaps the AI could either use some of that gold sitting around or use some of those shields saved up for a wonder or two that was discontinued. Then move as a group to enemy territory, and, if at war, attack the first coastal city it comes to where the mobile units can land and attack on the same turn. Dip first, with a high probability of taking out city walls if there are any. Seems like it would be a fairly simple algorythm to impliment. In the event of feudalism perhaps the battle group should be changed to the dip, required boat, 2 good single movement attack-defense units like legions, and 6 single movement units like catapults. And of course the dip with an exceptional ability to bust walls. Again go to nearest likely enemy coastal city. Land in 2 groups on two squares adjacent to the city and bust walls with dip. If the human only has 1 offensive unit in the city one of the stacks has a good chance of surviving. Next turn, first attack with 3 offensive units. Then hurl boat or boats at city. Finally attack with lastoffensive unit. Note: I have found very few people who defend a city with more than 3 units. These seem like 2 fairly simple proceedures that may not be too much to get an AI to do. It doesn't have to figure out exactly what to build and how to assemble it, because it is all one build order in 1 city. Granted, these two proceedures are easily defeatable, but certainly not without cost. Even if it forces the human to go for feudalism early, and have at least 2 attacking units in each city near a coast, it's taken effort and resources.

The next few suggestions are about defense, because AI citiea are usually pathetically easy to take.

2. If the AI has engineering, any time a city reaches some threshhold in size, developement, and food availibility, make an engineer. When that engineer is built, It terraforms the city square either into forest,if already on a river square, or otherwise into a hill, then a mountain.

3. Forest all coastal squares adjacent to a coastal city, if food availibility won't be hurt too much. That way landing and attacking on the same turn is not quite so easy.

4. Around any city where food availibility won't be hurt too much, forest all 8 squares adjacent to the city. Have at most 2 roads on any of these adjacent squares, always with a fort on each and occupiesd with 2 units. Forests will slow 2 movement units, but won't provide too much of a defensive bonus. I wouldn't think this would be too hard to do. Just have the AI try to optimize the city radius of any city that does not overlap another. Try to make it look as much as possible like this: inner square the city, next 8 squares forest or mined hills, outer 12 squares grassland. If a minimum food availibility potential is reached, it stops foresting the inner squares. It also forests any squares adjacent to both the city and the sea first. Next, it forests the squares adjacent to the city that are closest to the enemy. If food availibility falls below a certain point, it starts to turn outer squares into irrigated improved grass-farmland. Once the food availibility is restored, it goes back to forresting adjacent to the city. If The food availibility can't be brought up enough, it stops. This all will involve a lot of terraforming, so give the computer a bit of a cheat to speed things up.

5. Once a city reaches a certain threashold, maintain 2 good attack- defense units
(like legions), an attacking mobile unit (like a crusader) and the strongest availible attack unit(like a catapult). They need to be upgraded when appropriate tech becomes availible. These units need to stay in the city.

6. A diplomat fortified in at least half of all cities. When a city is taken, it can be cheaply bribed back on the next turn, if the AI doesn't do something stupid like accept an immediate cease fire. Also, a dip, though low in defense, costs nothing to maintain and is just 1 more unit that has to be taken out before a city can be taken.

7. Expand preferentially towards the enemy until cities begin to overlap or the coastis reached. Then expand elsewhere. This will make expansion harder for the human player.

8. Have an AI player or 2 use a fast expansion policy; building only warriors and settlers, or phalanxes and settlers until either room for expansion is gone or war makes a different policy necessary.

8. Coordinate AI activities more to the detriment of the human player in a single player game. This can be done largely simply by the personality of the various AI's. For example, in a 7 player game, perhaps the 2 civs closest to the human player should have a militaristic expansion personality, provided they share the same land mass as the human. Some sort of balance is sought between building settlers and building attack units to send towards the enemy. The next two closest AI's, or the AI's with the most room for expansion, should perhaps adopt a rapid expansion personality; avoiding war and just trying to get those cities down til seriously threatened or availible land is gone. Thelast 2 civs, either the furthest away or the ones with less room to grow, should try to build 3 or 4 cities relatively fast, then start building wonders, preferentially in or close to the capital, and once built, defend the wonder cities well. They should go for the early wonders people usually like the most, like HG, GL, and perhaps MPE. Leave the Pyramids and GW for the larger AI's to build. (And one of the larger AI's should at least make the effort.)

9. Perhaps defend any city with 2 or more wonders with at least 1 more unit than it has population. That way if the city walls are destroyed the city will be destroyed, along with the wonders, before it can be taken. Dips could make up a large part of the defense force since they cost nothing to maintain.

10. AIs should place a much higher priority on getting JSB or MC, as well as the tech to get these.

[This message has been edited by Matthew (edited June 26, 2000).]

[Basil]

This post is mainly about basic things the SMAC AI doesn't do

In Civ2 I won on Deity using a mild-moderate ICS strategy.

In SMAC I win by default: the AI is completely out of its depth after the discovery of air power.

Crucial flaws:
1. The AI doesn't know how to terraform.
1a: Assigning each base a random chance of building a former doesn't work: often, bases are unproductive because the land around them hasn't been terraformed, occasionally excess terraformers on an island sit idle using support. Ideally, the AI should figure out how many formers it needs; less ambitiously it could simply assume it needs 1 former per city.
In Civ, where settlers/engineers require population points as well as maintenance, this would be more tricky.
1b: The AI should figure out which tiles within a base's radius will give the best production, given its current technology, and then terraform those. At the moment it does silly things like building mines instead of forests before mineral restrictions are lifted.
1c: Probably not relevant to Civ3, but it should realize when it needs sea formers, instead of waiting until a port base happens to build one.

The consequence of these things is that over the long run, during the mid-game, the AI grows at about 1% per turn whereas I grow at maybe 2 1/2% per turn. Over 100 or 150 turns that 1 1/2% per turn difference gives me the game.

2. The AI doesn't know how to use aircraft.
At the moment it seems to have two strategies for using aircraft: either (a) attack easy targets, such as unarmoured formers or rovers, or (b) conserve the aircraft, not risking them. These are both good strategies to use at the right times and places, but it needs a third: all-out attack in support of the army, using bombers to kill enemy garrisons that are holding bases the army is attacking, and being willing to lose bombers during these attacks. The consequence of not risking bombers coupled with the AI's focus on building bombers and not ground assault troops is that the AI can't launch anything beyond nuisance attacks after the discovery of air power.
Ideally, the AI's aircraft should also fly patrols. But this is tricky: they shouldn't fly long patrols into areas where they are likely to get attacked by fighters, etc.
Really, of course, bombers should attack like artillery.

3. The AI doesn't assign the Social Engineering factor of Efficiency a higher priority as its colony grows larger. In one of my games about 30 bases of the 60-base Usurper faction were producing 0 energy due to inefficiency.

Lesser flaws:
4. The AI doesn't build supply crawlers in order to gather resources. When it has supply crawlers left over from building a SP, it doesn't know which tiles they should be used on. (E.g. it will harvest 0 factors of production from a square.)

5. The AI doesn't realize that it should build a new headquarters if its old headquarters is destroyed. (In SMAC a headquarters may -or may not- be added to a build queue somewhere sometime. Instead, it should realise that it needs a HQ, calculate the best spot, and override the governor in that base.)

6. The AI doesn't know how to pop boom. This is a secondary thing, because not knowing how to terraform means that bases' maximum populations aren't very high anyway. (The Usurpers built the Cloning Vats in the game mentioned above and I was surprised to see that it made little difference.

To be really effective, the AI needs to have a better idea of how to launch amphibious invasions. But this would be very difficult. Get simple things like terraforming working first.

[This message has been edited by Basil (edited June 27, 2000).]

[SmartFart]

Wow...this thread growed fat

Alot of suggestons 'use bombers,use cannons' type,but main problem stays:

My AI never see anything stronger than pikeman/crusader/catapult.

Or my disk is invalid?

------------------
No i ain't doing much,doing nothing means alot to me (AC/DC)

[Tiberius]

quote: Why the hell can I not build a civilization in Civilization games, but I am forced to build a war machine?

I agree very much here with Lodi and Gord McLeod. There are many-many posts here on how to beat the AI, and every one is about how to beat it in millitary way.

I think we should have the possibility to win the game without going to war. In Civ2 the AI ALWAYS sneak-attack me, very often in a kamikaze way, no matter how peacefull I play and how strong my army is. One turn they are "enthusiastic, peace", the very next turn "enraged, war"! That doesn't seem to me realistic.

It would be very nice to have AI "personalities" which try to win in alternative ways (for ex. becoming a trade power). Making trade embargos should be a powerfull "weapon" to force your enemies to make peace. I'd like to see the time when the AI will be real "scared" when I'm threatening him with a UN meeting or a global trade embargo! I'd like to see that the reputation of a country is really important and influences the behaviour of the AI, not just in relation with the human player but also between 2 AI civs.

I'd like to see the time when we will discuss about how the AI (or at least a peacefull one) can PEACEFULLY win: makes alliances to isolate his enemies, unions to become a trade power, UN-meetings to force peace treaties or embargos, share research projects or wonders and so on.

I know it's a hard job, so good luck and thanks for listening !