Some suggestions for improving the GOTM AI

alexman

Ancient Geek
Joined
Feb 28, 2002
Messages
792
Location
Mohawk
I have to confess that as I watched the GOTM staff add more modifications to the stock rules month after month, I was afraid that it would hurt player participation. I was obviously wrong! :)

Originally, modifications were restricted to three areas: Changes that bring PTW and vanilla Civ3 closer together, changes that remove exploits, and minor changes and unit additions that add flavor and change the effective map size.

But with the introduction of the Predator and Conquest classes, more modifications are being made to the behavior of the AI, so that a game of a given level can appear more or less difficult. Changes are also being made to some AI civilizations (GOTM19 Rome, GOTM20 Zulu), to make them behave in a certain way. This is where I have some suggestions for future games.

The following suggestions are tested extensively, and they do work. Some of them are more radical than others, so my only hope is that the GOTM staff reads through them, becomes aware of them, and do as they wish with them. The modifications can be made to the Predator-class AI, or they can be incorporated in all classes to make the AI rely less on bonuses and more on better decision-making.

  • Remove the AI offense flag from Mech. Infantry, Impi, Numidian Mercenary, Musketeer. For the AI, each unit can have a single purpose – either offense or defense. Most units have a single AI strategy, but there are some units, like Infantry for example, that can be built for either purpose. But is it smart for the AI to build Mech. Infantry for attacking, when it can be building Tanks? When the AI needs an offensive unit, it builds one from the list of available units with the AI offense flag checked. Unfortunately, the AI has a chance to build any unit, even units that are clearly inferior to others. For example, the AI sometimes builds Guerillas, even when it can build Infantry. Similarly, the AI sometimes builds Mech. Infantry for offense, even though Tanks are always available at the same time. More importantly, the AI usually has lots of Infantry running around, built for attack in the Industrial era before Tanks were available. When time comes to upgrade these units, they become Mech. Infantry with an offensive purpose, and they get sacrificed trying to attack fortified enemy Mech. Infantry. If the AI offense flag is removed from Mech. Infantry, the upgraded units get the AI defense flag, and the AI suddenly has a lack of offensive units after upgrading. Consequently, it has to build Tanks or Modern Armor to compensate. That’s what the human players do, and for a good reason! Similarly, some of the Zulu quota on offensive units is accounted for by Impi, the Carthaginian AI builds some Numidian Mercs for attack when it would be better off building swordsmen, and the French build some Musketeers, which are just too expensive for a 3-strength offensive unit.
  • Increase the Optimal Number of cities for all map sizes by 50%. Reduce the percentage of optimal number of cities for all difficulty levels by 1/3. The reason for this change is to make the AI more aggressive in pursuing land. It is sad to see how the AI does not even come close to threatening Domination on any map size greater than tiny. The combined effect on gameplay of these two changes is minimal. There is absolutely no change in corruption. The only change is that players need 50% more cities before being able to build the Forbidden Palace. But increasing the OCN encourages the AI to capture (not raze) cities. The hard-coded behavior of the AI to stop expanding at the OCN was put in when Civ3 was first released. Since then, Firaxis has provided more ways to reduce corruption (police stations, more powerful courthouses), so it’s worth it to expand well over the nominal OCN.
  • Increase attack strength of Riflemen by 1. Increase attack strength of Infantry, Guerillas, Marines, and Paratroopers by 2. Humans know to wait for Tanks (if possible) before going on a major offensive against Infantry. The AI doesn’t know that even though the above units are flagged as offensive units, they’re not that great, and often gets into costly Infantry-versus-Infantry wars, especially since it doesn’t know how to use Artillery. Increasing the attack strength of these units makes the AI seem like less of an idiot. :)
  • Remove from the Celts the ability to build Medieval Infantry. Humans recognize the value of the Gallic Swordsmen in the middle ages, and can avoid upgrading this “Light Knight”. Since no cash is required for the upgrade (Med. Infantry is actually cheaper), the AI mindlessly upgrades these units (unless it has not yet triggered its Golden Age). In doing this, the AI loses value towards the eventual Guerilla upgrade.
  • Frigates upgrade to Destroyers. Ironclads upgrade to Battleships. Humans know better than to invest shields into these early naval units that become obsolete so fast. The AI builds naval units as soon as they become available, and gets stuck with a fleet of Ironclads and Frigates in the Modern age.
  • Add Communism-specific city improvement (secret police). Reduces corruption, produces veteran units, costs 40 shields, has zero upkeep. The AI selects Communism as its form of war-time government, even though Monarchy is usually a far better choice. But when built in every city, the Secret Police improvement brings Communism to the level of Monarchy for empires with a bad FP placement (such as most AI empires). Monarchy with a good FP placement is still much better than Communism, so the change would not significantly affect human strategy. The veteran unit ability is suggested so that the AI places a high priority getting the improvement built as soon as possible, especially since it is usually at war when in Communism.
  • Double the effect of entertainers. The AI does not use the luxury slider except to combat war weariness. When a city becomes unhappy, the AI sacrifices growth and production by assigning entertainers instead. While doubling the effect of entertainers is not enough to tempt the human to forfeit the use of the luxury slider, the AI will benefit because it will need fewer specialists to keep its cities happy.
  • Add the “Enables Alliances” ability to Code of Laws, Literature, Polytheism, Republic, Electricity, Atomic Theory, Radio, and Nuclear Power. Add the "Enables Communication Trading" ability to Literature and Medicine. To make the AI take a more varied approach to research, and to make less predictable the next technology that it will pursue. These abilities make no difference in gameplay, since they are already given by Writing (a prerequisite of all of them except Polytheism, but you need an embassy before you can form an alliance anyway).
  • Change the AI build-often list as shown below. Increase the aggression of Iroquois and Chinese by 2. Decrease aggression of builder Babylon by 2. Add trade to all civs to encourage harbors and marketplaces, which tries to alleviate their entertainer-mania. Encourage the AI to take advantage of their traits: Scientific have science flagged, Religious have happiness, Militaristic have units. Civs that are not Scientific, not Religious, and not Militaristic, build Culture.
    Code:
    ------------------------------------------------------------------------
      Civ Agg Off Def Art Stl Wrk Nav Air Gth Prd Hap Sci Wlt Trd Epl Clt
    ------------------------------------------------------------------------
      Rom  4   X   X                       X   X               +
      Egy  3                               X   X   +           +       -
      Gre  3                       X                   X       X       -
      Bab 4->2                                         -       +       X
      Ger  5   X                                       X       +       -
      Rus  4                           X   X           X       +
      Chn 2->4 +                           X   X       -       +
      USA  3                           X   X   X               +       X
      Jpn  4   X                   X               X           +
      Fra  1                                       -           X       X
      Ind  1                               X       +       X   X       -
      Per  4   X                                       +   X   X
      Azt  4   X                                   X           +
      Zul  5   X                                               +
      Iro 2->4 +                           X       X           +   -
      Eng  3                       X                       X   X       X
      Mgl  5   X                               X               +
      Spn  3                       X               +       X   X   -
      Sca  4   X                   X                           +
      Ott  3          X                            -   +       +
      Cel  4   X                                   +           X       -
      Ara  4                               X       +           +       -
      Car  2                       X                           X       +
      Kor  2          +                                +   X   X
    
      X = build often
      - = removed
      + = added
    ------------------------------------------------------------------------
 
Only comment I have is about the naval upgrades; upgrading ironclads > destroyers seems fair enough - but am not sure frigates > destroyers is really a good idea...

Too much of an exploit for the human players!

Apart from that, not really experienced enough to comment on the others... Some seem great, others seem a bit overpowering.
 
I like them except:
Increasing strength of rifleman, geurilla and infantry. Just ruins the fun of the industrial age of shelling cities. removes the challenge.
The communism one. I like the idea of an extra, cheap 30-40 shield, corruption reducing improvement for communism, but not the no upkeep cost. It has to require upkeep.
Double the effects of entertainers - no, no, no. Just no. Unbalances the game too much. e.g. newly captured cities will never flip back.
 
Originally posted by Gingerbread Man
I like them except:
Increasing strength of rifleman, geurilla and infantry. Just ruins the fun of the industrial age of shelling cities. removes the challenge.
You still have to shell cities. An Infantry attack factor of 8 against fortified Infantry in cities has low odds to succeed without Artillery support.

The communism one. I like the idea of an extra, cheap 30-40 shield, corruption reducing improvement for communism, but not the no upkeep cost. It has to require upkeep.
The whole idea is to give Communism some extra cash to be able to compete with Monarchy. The upkeep cost works against this goal, but this was only an example. The staff can fine-tune the proposal as they see fit.

Double the effects of entertainers - no, no, no. Just no. Unbalances the game too much. e.g. newly captured cities will never flip back.
It does not unbalance the game, trust me. Captured cities will not easily go into WLTKD with the unhappiness due to "aggression against their mother country", nor with a food deficit from assigning entertainers. And even when celebrating, cities do have a chance of flipping. This modification has been tested over the course of several months in comparison games and the only time it affects the player is when in Anarchy (cities starve less). But the benefits to the AI are considerable.
 
Alexman, you picked a hell of a day to introduce these stimulating suggestions. I assume that they are the gist of the AU Mod, and therefore have been tested by many experienced players over many, many games. It might help if you could provide some information as to how these mods have affected the game. I also have a feeling that Cracker is already aware of them, and has either already given them consideration, or is at work implementing somerhing similar.

These were my quick thoughts:

1. The MI defense setting makes complete sense. Would changing the setting for the AI UU's that you cited make it much less likely for these civs to have a GA?

2. Some change to stop the AI from late-game razing would make them more effective. Your fix makes sense, with the only downside being the longer wait for a human to build the FP. Is there an even more transparent way of achieving the same goal of improving the AI's domination chances?

3. Increasing the attack strength of industrial-era defensive units clearly helps the AI, and your extensive testing indicates that it does not take away from the human use of combined arms, to the overall game's detriment.

4. Forcing the Kelts to keep their UU rather than replacing it with a medieval unit is a good idea. Could the UU be replaced down the line by militia, however?

5. Adding the ability to upgrade ironclads and frigates makes "game' sense, and adjusting the upgrade cost would limit human exploitation of this (although why pick on this version of upgrade exploitation, given how often it's done with other units?).

6. Making Communism more effective for the AI, without unbalancing the government system for humans, is a great idea.

7. Doubling the effect of entertainers would be great if it could be restricted to the AI. Since it presumably can't, what effect has this mod had on the games in which it has been used? The tipping point would seem to be whether humans used entertainers more often than before.

8. The "enable alliances" mod seems like a great fix to predictable AI research patterns. It could be argued that it will benefit the human even more if that pattern were more varied, as it would greatly aid the astute trader. Do these flags keep the civs researching in a pattern that fits their traits, or do they result in a truly random AI choice pattern?

9. Enhancing AI behavior by altering civ patterns is something seemingly evident in the GOTM already. These suggestions seem to make sense at first glance. How have they affected your games?
 
Originally posted by Txurce

5. Adding the ability to upgrade ironclads and frigates makes "game' sense, and adjusting the upgrade cost would limit human exploitation of this (although why pick on this version of upgrade exploitation, given how often it's done with other units?).

I for one am not happy about humans being able to upgrade ANYTHING to a battleship. It's the most expensive combat unit in the game (stealth bombers are flying arty), and humans could easily knock up a kick ass fleet in only a dozen turns.

How well does the AI use navies now? Not very well.

Limiting the AI updrade to just Ironclad > Destroyer forces the AI up that path, and prevents the human player from getting an invincible fleet in a relatively short period of time
 
Originally posted by Txurce
1. The MI defense setting makes complete sense. Would changing the setting for the AI UU's that you cited make it much less likely for these civs to have a GA?
Of course, the less you use your UU, the lower the chance of getting a GA. But for the AI, the chance is not much lower at all. If an AI civ gets into a war, you can bet it will bring its defenders in its attack, surely winning a battle when it gets counter-attacked. So Zulu, Carthage, and France will almost certainly get into a GA when they enter a war at the time their UUs are available. It's only the human who sometimes launches attacks without defenders (e.g. pure Cavalry).

2. Some change to stop the AI from late-game razing would make them more effective. Your fix makes sense, with the only downside being the longer wait for a human to build the FP. Is there an even more transparent way of achieving the same goal of improving the AI's domination chances?
Unfortunately, there is no way to change the AI's desire to expand without affecting the human. In practice, a 50% increase in the required cities for a FP does not affect human strategy in a big way. That's 12 cities instead of 8 on a standard map. Most players are not ready for a palace jump until after they have 12 cities anyway! ;)

4. Forcing the Kelts to keep their UU rather than replacing it with a medieval unit is a good idea. Could the UU be replaced down the line by militia, however?
Yes, it could. The upgrade path would not be changed. The Celts would simply be unable to build Medieval Infantry.

5. Adding the ability to upgrade ironclads and frigates makes "game' sense, and adjusting the upgrade cost would limit human exploitation of this (although why pick on this version of upgrade exploitation, given how often it's done with other units?).
First of all, I'll admit that this change might be changing the game too much for the GOTM. Adjusting the upgrade cost would not be necessary, as it is purely a function of the difference in shield costs of the two units. A single Frigate-to-Destroyer upgrade would cost 120 gold without Leo's, and a Ironclad-to-Battleship upgrade would cost 240 gold! Hardly exploitable, given the relative weakness of naval power in Civ3.

7. Doubling the effect of entertainers would be great if it could be restricted to the AI. Since it presumably can't, what effect has this mod had on the games in which it has been used? The tipping point would seem to be whether humans used entertainers more often than before.
It still definitely does not pay to use entertainers under most circumstances. Obviously, if you have one city that is much larger than most of the other cities in your empire, and it also suffers from corruption, you (like the AI) would need fewer entertainers to keep it happy. But a properly managed human empire does not get into that situation in the first place, and it's always better to use the slider, even with more powerful entertainers. The only time when the change really affects the human is during Anarchy, when luxury slider is not used.

8. The "enable alliances" mod seems like a great fix to predictable AI research patterns. It could be argued that it will benefit the human even more if that pattern were more varied, as it would greatly aid the astute trader. Do these flags keep the civs researching in a pattern that fits their traits, or do they result in a truly random AI choice pattern?
Civ traits really do not affect the AI reseach path. The AI makes a random choice, weighted by the value it places on each available tech (see the article). This change brings the AI values of techs closer together, so that the choice becomes more unpredictable. Some of the changes also encourage the AI to research technologies that are normally ignored because they do not offer important benefits but lie in an important branch of the tech tree (Electricity towards Scientific Method, for example).

9. Enhancing AI behavior by altering civ patterns is something seemingly evident in the GOTM already. These suggestions seem to make sense at first glance. How have they affected your games?
Actually, the AI preferences do not have as great an effect as one might think, but they definitely help improve the decision making of the AI a little. For example, a human Ottoman player would never in his/her right mind build a temple before a library, but the Ottoman AI has happiness, not science, flagged. Again, this list was only an example. In our games we have had production added to all AI civs to encourage building factories, but it's not clear if it helped the AI.
 
Originally posted by alexman



First of all, I'll admit that this change might be changing the game too much for the GOTM. Adjusting the upgrade cost would not be necessary, as it is purely a function of the difference in shield costs of the two units. A single Frigate-to-Destroyer upgrade would cost 120 gold without Leo's, and a Ironclad-to-Battleship upgrade would cost 240 gold! Hardly exploitable, given the relative weakness of naval power in Civ3.

Give this I would argue that the AI would actually be hampered by this. The AI tends to upgrade even things that are not strategically useful. 4 Iron Clads could effectively bankrupt an AI for little to no actual use. At least when they upgrade Medieval Infantry they get a decent defender/skirmisher.

I like most of the other mods though.
 
Back
Top Bottom