AI flaws and needed improvements for Civ3

[Kevin Ar18]

In my games against the AI in Civ2 I usually encounter a number of basic flaws. Since all of my Civilization games are single player only, I would greatly love to see some solutions to these flaws in Civ3.

Flaws & Corresponding Suggestions
I. Battle Fronts.
This is the major flaw I often notice. A normal human player will send the majority of their units towards the edge of their territory to either protect itself from an enemy or to attack another enemy. However, the AI, often scatters their units throughout their territory making it easy to take cities one by one.
The way I see it, the AI should be made to focus all their units on battlefronts bordering their enemies thus providing much better protection with much of their forces.
BTW, when I refer to battlefronts, I am usually speaking of border cities, since they are usually the safest place to keep units.

II. Organized Attacks.
This is another flaw common to not only Civ2 but the majority of computer AI's. Often the AI will send one unit after the other single file or in small, unorganized groups. This of course is grossly ineffective to take over well defended cities and the AI usually makes little progress in taking over enemy cities even if the AI is more powerful.
In certain instances the AI should investigate the city or Civ to some extent that it is attacking to at least estimate what it would take to capture a city. Then instead of sending a single unit, the AI should build a proper force that it thinks can capture what it wants, then send all the units at once in one organized assault.
One thing that can be of use as well is to send a defensive unit along with the assault for the sole purpose of occupying and defending a city immediately after it is captured.

III.Proper use of defensive and attack units.
I can't tell how many times I've seen the AI not make proper use of the units it can build. Instead of using the units with the most hit points to defend the cities, they'll use cheaper but weaker units (even though the could have easily built the better ones). Then, the AI has the nerve to use those weak units that should be for defense to attack instead.
The AI needs to start using the strongest possible defense units to defend their cities, then the powerful attack units could be used in organized assaults.
As an added precaution, the AI should also keep a few powerful attack units in their territory to kill any incoming units before they even get to their cities. (I use this strategy against the AI.)
The AI should also send heavy defensive units to protect their weak attack units when it makes a major organized assault. (I too use this strategy.)

IV. Expansion across oceans.
This isn't exactly a flaw like the other three ideas, but I felt it was worth mentionioning nonetheless. Currently when the AI tries to send an attack force across a body of water in a transport, it it relatively weak. Also, on maps with many small islands it seems about the only way the AI will usually capture one of your cities is using paratroopers. Something needs to be done to get the AI to send ground units in a large scale in transports to perform ground assaults on new islands.

Advanced AI Suggestions
Other than the basic flaws I mentioned, there are a number of other more abstract areas in which I think the AI could get a little improvment.
Plug-ins for Civ3 tricks.
The more you played Civ2, the more you found a number of little tricks to attack the AI with. For example, using the railroad to perform massive assaults on cities or even kill the entire Civ in a single turn. Or you could use engineers to build a railroad directly to the front door of an enemy city in one turn; or use them to build a fortress on a new island, land a horde of units in the new fortress, and then be protected as the AI takes its turn.
There are just a few of the many "tricks" that can be used in Civ2.... As similar tricks are found out about Civ3 it would be nice if Firaxis would provide little AI add-ins to give the AI the ability to use these tricks (if the player so chooses).
[This message has been edited by Kevin Ar18 (edited February 13, 2001).]
[This message has been edited by Kevin Ar18 (edited April 11, 2001).]

[ChrisShaffer]

quote: So what do ya'll think about these AI improvements?

I think they focus to heavily on problems with the AI combat skills. If I were making an AI flaw list, problems with empire building would be higher on the list than problems with combat.

Main AI deficiencies -

(1) Knowing what units do what. Dammit! Fortify the mech infantry in a city and send the howitzer forward to the front. Better yet, stack a mech inf with the howitzer for defence. The AI acts like Col. Klinck.

(2) Cross oceanic invasions. Once in civ2 the AI brought over two transposts at once and dumped 8 or 9 units (they were even the most advanced units) and caused me hours of trouble late in the game. Civ3 should do this every game.

[Kevin Ar18]

quote: If I were making an AI flaw list, problems with empire building would be higher on the list than problems with combat.

Personally most of the serious problems I've noticed in the AI have been combat related. City building just doesn't seem a major issue to me. City building inefficiencies can cheaply be fixed by giving the AI better resources, allowing it to cheat or whatever (which you won't even see). But, the combat is a different story -- that part you encounter all the time and is basicly the only part that challenges you as a human player. If the AI's combat is unorganized you can just do practically whatever you want to kill 'em.
However, I'd be interested to see what city managment related issues you have in mind. I'll list a few I notice.
1a. Excessive use of improvements
Often the computer builds stuff on terrain that is useless. Like irrigation on hills and particulary railroads on EVERY square. To me this is just waste of the comp's resources.
1b. Use of fastest route
Many times, the comp will not make use of the fastest route. A better calculation needs to be made so the AI can make use of the railroads and the fastest route. I think one of the reasons excessive railroads are used by the AI is because the developers recognized the AI's inefficient routes. So this issue needs to be resolved before conserving improvement building.
2a. Lack of Terraforming
Rarely do I see the AI terraform terrain into something better in Civ2. I think the AI needs to make use this feature much more often in Civ3.
2b. Lack of shield producing terrain
This goes in line with terraforming, but the AI will also not make use of terrain that can collect much more resources (shields). Like terraforming grasslands without shields into hills and then making mines can give cities great boosts of production. The AI needs to be taught to go for these production benefits instead of just sticking with terrain it's got.
3. City Placement
Currently in Civ2 the AI places cities in a predictable pattern quite close to each other. However, this can lead to the AI putting cities in not too productive areas and thus wasting settlers on cities that will not grow to any worthwhile use in the future. Instead, the AI should take into consideration expansion, special resources, rivers, ocean, how much food producing areas, and how much shield producing areas when selecting a city location.
When I select a city location I pick areas that have good food producing spots as well as good shield producing areas or tiles that can be terraformed to such purposes. Also, if the city can't build ocean stuctures I try not to get any ocean in the city limits. These and other very basic considerations, that most human players consider when building a city, should be incorperated into the way an AI decides it's city locations.
[This message has been edited by Kevin Ar18 (edited February 14, 2001).]

[Kevin Ar18]

Here's a few more Civ2 related tricks:
Using caravans to build a wonder at the last second before the computer gets it. Using units or airports to keep nuclear missiles from getting your cities.
While these same bugs/tricks may not be in Civ3 it still would be nice for Firaxis to provide little AI add-ins as similar strategies are found for Civ3.
Since some of these strategies are rather foul I think you should also have some way to select which tricks the AI can and cannot use.

So what do ya'll think about these AI improvements?

[ChrisShaffer]

Even at Deity level in Civ2, by 0 AD my empire is at least three times as big as the closest AI. The demographics screen shows me in 1st or 2nd in every significant category. At that point, it doesn't matter what the AI does with its military - I'm so far ahead in technology and building capacity that I can roll over them with impunity.

Also, I like playing "race to space" games much more than I like "conquer the world" games. Aside from anything else, race to space is more realistic - at least we've sent people into space. Nobody has ever come close to conquering the world. And I'm a builder at heart. I want to have a challenge in games where war is not a significant factor - on Giga maps for example. Civ2 doesn't provide that challenge.

Wouldn't it be nice to see an AI that can expand effectively? That knows how to build a super science city? That doesn't build nearly obsolete wonders? That really knows how to take advantage of terrain? That doesn't build libraries in cities that will never, ever, have significant trade? The list of bad decisions made by the AI in empire building goes on and on and on...

[Windborne]

The biggest problem with the computer AI is that is it just ONE ai, I finally retired civ II, took it off my computer when I got SEIV because that game allowed you not only to create different distinct races, but different AI files for each one, each opponent plays differently, values different things, builds different kinds of units, and says different speaches to you. I usually play the game hot-seat with a few friends, but the ai, especially the one for the terrans makes even single player games much more interesting. You get to know the individual races and their flaws, not just the nameless AI of civ II and before.

quote: Originally posted by Windborne on 02-16-2001 03:41 AM I finally retired civ II, took it off my computer when I got SEIV

Please tell me what's the 'SEIV'?

[Grumbold]

I would guess Space Empires IV. A good low budget 4x game with a HUGE tech tree. It doesn't quite have the 'just one more turn' factor for me, but it has the biggest and possibly best tech tree of any game so far. It's only available for purchase via the net, I believe (www.malfador.com).
[This message has been edited by Grumbold (edited February 16, 2001).]

Humm...Never heard of it

[Roman]

quote: Originally posted by Kevin Ar18 on 02-14-2001 10:48 PM Personally most of the serious problems I've noticed in the AI have been combat related. City building just doesn't seem a major issue to me. City building inefficiencies can cheaply be fixed by giving the AI better resources, allowing it to cheat or whatever (which you won't even see). But, the combat is a different story -- that part you encounter all the time and is basicly the only part that challenges you as a human player. If the AI's combat is unorganized you can just do practically whatever you want to kill 'em.

I agree totally. I am not a very good Civ2 player, so the AI on deity actually outproduces me significantly unless I use ICS. The only reason I can survive is that the AI uses their vastly superior numbers of units wastefully and especially because it does not "understand" the principle of active defence.

[Kevin Ar18]

I see people saying something about a harder difficulty level in this thread:
http://www.apolyton.net/forums/Forum6/HTML/002488.html

I think teaching the AI to attack properly as outlined here would give a much stronger AI on the combat side.
[This message has been edited by Kevin Ar18 (edited April 11, 2001).]

[SerapisIV]

A lot has been made of the ability in the AI in Black and White to learn fromt he player and get better. Wouldn't it be easier to program a learning AI (hardest initially, but more robust in the long run) and then have it learn from play-testers who are actually humans. You can have players play in different styles and those styles can be kept as seperate AI patterns. Well, this is all theoretical, Civ is much, much more involved at the micromanagement level the B&W, so I don't know how prohibitively expensive in terms of programming and CPU/RAM such an AI would be.

[markusf]

Space empires 4 is good, i like the mods that add 1000's of techs to the tech tree... but it gets kind of boring to play.


AI capitol needs a build script for improvements/wonders

Another Build script for the rest of the cities in the empire... If not attacked build... temple,marketplace,caravan, caravan,caravan,library, aqaduct etc etc.

No matter what firaxas releases there are going to be people who find out ways to beat the AI consistantly and easilly. One of those people is me

AI should be scriptable. What i would like to see as "scriptable" is what units make up a AI army, build q's, and a set of AI strategies... For instance "Diplomat nightmare" AI brings a ship of 3 dipomats and sabotages your biggest cities. "AI bribe strat" ai sends out a dipomat in a ship and tries to bribe every city it encounters.