A Better AI.

[Zoolooman]

I think it'd be best if we simply artificially limited the AI's willingness to produce units. I'd like to see no more than 7 defenders in a city, no matter what, and I'd like to see no more than 20 units in an offensive stack, no matter what. I'd like to see peaceful AI's build even less. Does this mean that a human would outbuild an AI? Sure, whatever, AND WHY NOT? Humans ought to be able to win a conquest victory if they like. After all, the game is here to be beaten by the players however they prefer.

Honestly, I play Civ4 to win. I want to be challenged and interested on the way to victory, but I want to WIN. Every time the AI gets better in a way that makes it tedious to fight or even capable of beating me thoroughly, I get less interested in playing against it. For me, the game's replay value came in trying to win faster, harder, and better. This is why I enjoy the Games of the Month. I try to win faster, harder, and better than everyone else. But I never expect to lose unless I make some horrible mistakes. I wouldn't invest 9+ hours in a game if I expected to lose.

Now, before I get attacked, let me say that I used to play Emperor in Vanilla. I liked a hard AI, and I lost a few games along the way. But once I started winning, I always had fun. My finesse was beating the AI's stupid brute-force tactics.

When I play with the improved AI, sometimes I feel like my kind of game is gone. I'm playing someone's attempt to remake MP gameplay in SP, and you know what? I DON'T LIKE IT. It is pure tedium to conquer enemies in the Renaissance. I attack someone on a far continent and find their coastal cities instantly packed with 20+ longbows plus assorted gunpowder units. Even with an ABSURDLY large invasion force (16+ units), I can't conquer anything. This makes me want to uninstall not only Better AI, but the 2.08 patch as well. I don't even like the AI's improved tech speed. If I want a faster teching AI, I'll turn up the difficulty, which is exactly what I used to do.

I'm starting to miss the stupid AI with its huge brute force handicaps. I really am. I've given the improved AI a fighting chance and I've even recommended it to dozens of other players, but when I play a game and find myself utterly disgusted by the vast enemy defenses, I just quit. I could finish it out with yet ANOTHER peaceful victory, but why? It isn't worth it!

Please, please make a version of the Better AI that allows me to go on conquests deep into the game.

 

[KrikkitTwo]

Well if the issue is with the AI attempting to match the opponent power level, perhaps the AI needs to consider its goals when doing that.

If the AI has a largely Defensive Army, that should be considered 'more powerful' by the AI if the AI is planning to defend.

Essentially the AI should consider
1. Power Ratings of my Opponents (and how Opponenty they are..Humans should always be considered dangerous)

2. My Defense + Offense Ratings...
Defense v. my rival's "Power" tells me if I am safe.
Offense v. my rival's "Power" tells me if I can Invade.

SO each Unit/Tech/Pop building would have a Defense, Offense, and 'Power' Rating.

ie Trebuchets would provide almost no "Defense" Rating, and Longbows almost no "Offense" rating... So if AI had a Lot of Longbows, AI would feel safe.

The issue then becomes Risk
If you have two Builder AIs, they will build a defensive Army, which can be smaller than their neighbor's because it is optimized for defense. and it will get smaller and smaller as both of them keep their army small.

Now a builder next to a Warmonger, the Warmonger may optimize his Army for offense preparing to attack (sucessfully) because an army made for Attack will beat an army made for defense (standard defensive army being based on the assumption that the enemy army is mixed)

This opens the Warmonger up to attack himself from another Warmonger (or a dogpile of neutrals, but they don't know he is planning an attack.)

If the Warmonger tries for a 'balanced' army, then he will enter the arms race.

If you have Multiple Warmongers, they will continually 'up' their rival, until they decide to settle into 'neutral/defensive' armies.


Now the Ideal would be to organize diplomacy, so the AI could consider Allies and possible enemies (ie the computer should be much more willing to declare War on someone if you agree to declare War on them too, or if they are already at War)

The idea behind a War should be
1. Because you are weakened in a war, any participants in a war will suffer if third parties respond

2. Because you get weakened in a war, the power that benefits the most is the Fourth Party, the one that doesn't get involved in anyway... IF the war is inconclusive

3. Third Parties tend to make wars inconclusive. (as more people get involved the better it is to sit the War out)

So if the Zulus start attacking the French, then the other powers can either
1. Attack the Zulus to take advantage of their weakened position
2. Attack the French to take advantage of thier weakened position
3. Preserve their troops and focus on teching while the rest of the world blows itself up

In this way Diplomacy should be the alternative to troop buildup, and the way to solve it... so someone with no Friends should try to defend against the sum of the enemies...someone with good Friends can consider their Power part of their own defense.

So I'd say by
1. Proper consideration of Actual Military and Diplomatic position (my Offense, Defense, Friends' Power, Enemies' Power, and General Strategy)
2. A measure of Risk taking by the computer (in which a builder next to a Warmonger may likely lose, but two Warmongers next to each other may Both likely lose)

 

[sveint]

and I'd like to see no more than 20 units in an offensive stack, no matter what

I strongly disagree. This project has always been about making the AI play better. You argue that you want to win - fine play a difficulty level where you will.

I want a challenge at the lowest difficulty level possible.

 

[JimMac99]

Very interesting comments from the above recent posts, and in general I have to agree; the amount of units the AI currently builds and doesn't appear to be be using is simply too many.

I remember reading heated criticism of the Civ III AI on Apolyton, and Soren himself weighed in with a post to say that he could have made the AI much more challenging, but it wouldn't have been any more fun. In fact, he argued this would have detracted from the enjoyment of the game.

Anyway, I am sure Blake is aware and adjustments will be made. It sure aint easy to keep all strategies balanced.

 

[shadyforce]

I think one of the main problems the AI has is that the power rating doesn't accurately reflect a civ's ability to defend themselves or attack someone else. Thus they have no easy measure of how much units and of what type to build. So if the AI thinks it needs x power before it feels safe attacking a civ with y power, it will spam units even though it isn't always necessary. I don't know how easy it would be to program the AI to spy to evaluate the defensive strength of a target?

Also, when the AI decides to attack someone, it slows down research to amass a huge army of the present day best unit, and when it has enough it sends an enormous stack of doom about 3 times as much as defenders in the target cities.

However, when *I* plan on attacking someone, I group all my current attack units lying around and beeline to the next military tech (eg. Iron Working, Civil Service/Machinery, Chemistry, Assembly Line, etc). Then once the tech comes in, I freeze research and run 100% cash, I have a short burst of building top of the range units, and once I think I have enough, I upgrade all my old xisting units to their newest models as well. Then I send my much smaller but more effective stack-of-doom to the target cities. I can get away with a stack about 1/3 the size of the original. And Assuming my beeline has worked, and I use my siege weapons, etc, properly, I can usually take a few cities quickly and effectively.

The converse could also be true... that if a civ decided it was going to go on a builder phase, or even a cultural phase, then it should bee-line to the next defensive military tech, have a short burst of building defensive (and counter) units, run cash to upgrade existing ones, and once it feels safe, then ignore military completely and go on a building spree, or science boost, or whatever. And if done so succesfully, it should feel fairly safe for the next half era at least.

If the AI warmongers and builders did this, then they would be a much more efficient attacking/defending. The result of this is that the net amount of units across the board could decrease and there would be safety, not in numbers, but at a high enough level in the tech tree.

Of course, during a war you can't exactly beeline. You have to just spam units. But that's the way it's meant to be right?

 

[Arlborn]

There is no point in building more than 8 units per city unless some are used to offensive-defensive(like siege units going on the attacker stack just outside the city). Collateral damage will effectively destroy it all AND all AIs start to lose their personality.
Im looking forward to see the next building where Blake said the building of units will vary more with the personality of the AI. Even then I want the more militar AIs ot build a bit less..

But if its an offenssive AI, I dont see why dont build a huge offensive stack IF they are really going to use it..

 

[klokwerk]

Honestly, I play Civ4 to win. I want to be challenged and interested on the way to victory, but I want to WIN. Every time the AI gets better in a way that makes it tedious to fight or even capable of beating me thoroughly, I get less interested in playing against it. For me, the game's replay value came in trying to win faster, harder, and better. This is why I enjoy the Games of the Month. I try to win faster, harder, and better than everyone else. But I never expect to lose unless I make some horrible mistakes. I wouldn't invest 9+ hours in a game if I expected to lose.

I couldn't disagree more. I stop any game as soon as I'm sure I'll win. I have no problem with losing, it makes me better the next time. If I'm sure I'll win, then why should I play ??

 

[icemanjsg]

A few things i noticed when using a better AI - By the Iceman

Hey guys great work i love the mod , i have played 3 games all the way to the end now and just thought i would give my input to try and help you guys with your next release. This is no way meant to be offensive criticism! (i always play your mod )

1. All sea units have blue circles like fishing boats? I am not sure why

2. A lot of my units do as well.

3. As mentioned a lot for some reason units can bombard others can't

4. from the 1st of January to this release i have noticed that my AI have stopped building wonders? I got the oracle before i got stone hedge and when i got stone hedge no one would build it and a lot of other wonders for that fact.

5. I am not sure if it was because i had frigates and Monty had triremes but he would not move his ships out of his cities to engage my fleet.even when i was attacking in land He just let his entire fleet in the city to get destroyed (his power bar drastically dropped)

6.In 2 of my games (I'll talk about the first) Julius Caesar and Mao Zai Dong Shared an island it was huge they could of both easily had 15 cities, for some reason Mao built 4 cities then stopped he hadn't built any wonders either. Caesar had 8 then wiped him out (this was early as well) He then had the entire island to himself and ended up having around 30 cities! but by the end of the Game Caesar was so far behind because he had so many cities he was not a threat to anyone. I don't no why he declared war when he had room to share? This happened in another game of mine with Monty and Churchill


Anyways once again thanks for making a great mod hope something i wrote can help you.

R.e. for the record i totally disagree with Zoolooman in almost everything he said. I think the smarter the AI the better the game there is one thing that stops me playing this game and that is because its to easy. Sometimes playing this game is like jousting with a child, and on higher levels the computer don't get smarter they just get more help. If winning in a computer game is so important perhaps playing counterstrike is a better option.

 

[marioflag]

I totally agree with what Krikkitone says.
I wouldn't like that builders AI would be weakened, as Krikkitone said AI should always take in account a Risk factor, so also builders AI should always be able to defend themselves based on their "percepption" of the risk.
IMO the main difference between aggressive AI and builders AI should be that warmongers should have a greater army composed by aggressive units (trebuchet, cavalry)and defenders, a builder AI should have a smaller army all composed of units which are good to defend their homeland like Longbows or Pikeman.In this way builders would be a lot more ineffective in offensive wars but at the same it would be as hard as attacking a warmonger civ to take their cities.About diplomacy i also think that AI should take advantage of wars between others AI.When for example 2 AI fight each other , other AI should be more bent to declare wars on the 2 AI based on their relations, if they are warmongers and most important factor if one of the AI in war is a neighbour.

 

[Arlborn]

Builder AIs should have a maximun of 2/3 of the defensive power of an aggressive AI. Unless in war, then they should spam units just like any other AI, because ITS realistic. No CIV left they country to be dominated without fight of some sort.


And yes I enjoy losing. If the game is already won in thew middle of it, its boring and I mostly quit.

 

[Quagga]

Builder AIs should have a maximun of 2/3 of the defensive power of an aggressive AI. Unless in war, then they should spam units just like any other AI, because ITS realistic. No CIV left they country to be dominated without fight of some sort.

The question I have is what to do with the units after the war? (Presume the defense was a success, elsewise the point is moot.) Carrying all those units means a slowed economy.

I believe that is one of the things happening to the AIs. They build up for dagger attack or war defense then have a bunch of units hanging around forever.

 

[Tiberias]

I want a challenge at the lowest difficulty level possible.

I agree, overall, I want a more challenging AI. Players who only want to win should consider the difficulty level that they're at. The Improved AI is supposed to make the harder difficulty levels truly hard, and guess what? It does that extremely well.

In evaluating AI strategies, a valid question is, would a human player do the same? The "too many units" complaint doesn't really resonate with me, because in many situations human players will amass units, especially if they're bent on a conquest win. Consequently, if the computer player is shooting for a military win, massing units is the right move.

Conversely, if the computer is shooting for a spaceship win or a cultural win, it doesn't make sense to build so many units that its tech is crippled. Past the unit support limit, building of units should only be when they're critically needed for defense.

Of course, as with everyone else I'd like to see the computer player make more intelligent use of collateral damage (and protected against collateral damage) units, and to make better use of fewer units. If that happens, both aggressive and defensive AIs can safely scale back the sheer number of units to more managable levels, which would allow them to tech faster and be even more competitive against the human player.

 

[Uncle_Joe]

I strongly disagree. This project has always been about making the AI play better. You argue that you want to win - fine play a difficulty level where you will.

I want a challenge at the lowest difficulty level possible.

I fully agree with wanting a challenge with the lowest possible handicap. However, I dont feel that the AIs simply spamming more units makes it more challenging overall.

Does it make it harder to invade and win by conquest? Sure, in most cases it does. But the AI is NOT playing a 'better' game doing this. The main reason I see for this is that ALL the AIs are busily spamming units regardless of being at war or planning on it. So if the player is fighting a war now, he really isnt falling much behind the other AIs since they are slowing themselves as well.

Also, as I indicated above, the AIs having more units doesnt not necessarily translate to making them harder to defeat. Its still possible to defeat larger armies of AI troops with fewer, well used forces. I suspect that that will ALWAYS be the case when playing against the AI (through no fault of the programmer). But the result is that the more units dont really offer more security past a certain break point.

When I first played with these newer builds and was forced to build massed units myself, I thought 'wow, I'm screwed if one these huge AI armies attacks me'. But after having played a few times, I know that I can still easily defeat these massed armies (on attack or defense) PROVIDED that I have also massed a huge army myself.

But this is where the game breaks down now. There is less opportunity cost for building a large army now since EVERYONE is forced to do it en masse. In earlier builds, massing an army meant potentially falling behind non-engaged AIs. Now, that is not as much of an issue.

Also, many 'interesting decision' have been subdued because the only real decision is to mass armies of your own. And the cost of those armies completely dwarfs the cost of the other investments in building etc. Consider a typical 2.08 game. Lets say you have 6 cities. To defend them adequately, perhaps you might spend 180-250 hammers on each for defense (buying a few Longbows, a Maceman, a Catapult or two to add to the obsolete troops you might have). Total investment to defend is about 1080-1500 hammers. To attack a neighboring Civ, you mght invest maybe twice that again (call it 2000-3000 hammers). Compare that to cost of building, say, a University or a Market in each city (call it about 1000 total resources). That is a fairly solid trade off. With the need for massed troops however, that 2000-3000 investment in troops becomes more like 8000-15000 hammers needed to attack (at the MINIMUM) and 4000-6000 to defend adequately (again, at the bare minimum against 50-100 enemy units that might typically attack). Now that investment for the econ pales in comparison. And you never really 'stop' building troops to fit those buildings in.

But the biggest problem I see is as mentioned above and that is that you can freely build up those massed armies because all of the AIs are doing the same (and they must or else they die). And to me, that just kills a lot of the basic gameplay of CIVILIZATION. It makes more of a simple wargame than anything else. So, in the end, the game isn't really any HARDER as a result of all of these units (since I have them too), its just more tedious to play, destroys much of the balance where non-scalable factors are present, and removes many of the troops vs improvements decisions from the game.

All that said, I definately think that the 'stock' AIs can use a few more defensive troops. Two-three per city is not enough on the borders, but that doesnt mean they need 15-20 either. And building up for the attack should not entail needing 50-100 unit stacks either. If the pacing of the game is progressing normally (ie everyone is not concentrated on needing monster defensive stacks), then stacks of 10-12 should suffice with more on the way once at war.

Perhaps if repeatedly at total war, an AI might start to seriously build massed units, but as an every day, all the time occurance its not helping the AI become 'better'. Not at all.

 

[HardCoder]

Builder AIs should have a maximun of 2/3 of the defensive power of an aggressive AI. Unless in war, then they should spam units just like any other AI, because ITS realistic. No CIV left they country to be dominated without fight of some sort.

That's just wrong. The number of civilizations throughout history that have been overrun due to a lack of military or lack of willingness to fight far exceeds the number that have done the overruning.

 

[HardCoder]

So, in the end, the game isn't really any HARDER as a result of all of these units (since I have them too), its just more tedious to play, destroys much of the balance where non-scalable factors are present, and removes many of the troops vs improvements decisions from the game.

Right.

And in response to a post above, someone playing only to lose (as in, who never finishes a "winning" game) is perhaps playing a game that is too tedious to begin with.

 

[Uncle_Joe]

In evaluating AI strategies, a valid question is, would a human player do the same? The "too many units" complaint doesn't really resonate with me, because in many situations human players will amass units, especially if they're bent on a conquest win. Consequently, if the computer player is shooting for a military win, massing units is the right move.

This is not necessarily true. If it were, then the 'right' move would likely be to never stop building Axemen/Swordsmen etc. And if every game were played on a tiny Pangea then that might even work. But that isnt the way most games are played vs the AI so that is NOT a valid strategy IMO. Expansion/conquest (and hence military expenditure) must be tempered with growth of the econ. If your industrial output is ~10 city, then building even cheaper units is not going to compete in quantity or quality with Civs that are advancing faster (again, unless EVERYONE is spamming the same unit numbers).

Conversely, if the computer is shooting for a spaceship win or a cultural win, it doesn't make sense to build so many units that its tech is crippled. Past the unit support limit, building of units should only be when they're critically needed for defense.

The 'support limit' is a factor, but that is certainly not the only factor at work here. The simple massed HAMMERS needed to get those levels of troops is what is killing the AIs more than the support numbers. The simply fact is this...if the AIs make that investment, they must get pay-off for it. Otherwise a 'builder' AI building huge amounts of defensive troops will lose to another builder not making that investment (unless the former attacks the latter and wins). And therein lies the problem...once those types of military investments are made, the AIs MUST attack any lesser defended Civ or else they are surrendering a HUGE competitive advantage in development, making the game more of a simple wargame than anything else. And being a wargame is not Civ4's forte...not by a longshot. So indirectly reducing it to one should not be a side effect of a 'better' AI'.

 

[Uncle_Joe]

Earlier I tried to use some Hammer cost examples to illustrate the point, but I dont think it was very clear. Let me attempt to simplify:

Lets assume that Civ 'X' pursues a balanced path of troops and improvements. He spends 5000 on troops and 5000 on improvements.

Civ 'Y', decides to try and attack his neighbors as his way of bettering his lot in life. He spends 9000 on troops, and 1000 on basic improvements. If he attacks, he has a 9000 to 5000 advantage or almost 2 to 1 in troops strength.

Enter the newer builds. Now, a typical defense for a builder must be closer to 20000 on troops and 5000 for improvements.

The attacker, spending the same amount of resources once again would have 24000 and 1000. Now his margin for attack is only 24000 to 20000 or only 1.2 to 1.

The reason? The cost of troops now completely smothers the cost of developments. There is no longer any trade off here. Civ 'Y' will almost always lose.

In the earlier case, there is a real decision to be made between development and troops. And if you go heavy on development, then you are trading on either 1) not being attacked or 2) that your developments will give you enough of a long-term edge to be able to win a protracted war against a less developed Civ possessing a larger military. You must balance one against the other.

In the latter case, its far harder to get a military advantage over Civs that are continuously pumping out military units. It is possible for the player to do, but MUCH harder for the AI to do so. Without his 2 to 1 advantage over the builder, the attacking AI will likely lose hard. And being behind on development as well, he will eventually fade out and die like many of the Aggressive AIs I see in the later builds.

Hopefully the above illustrates the point properly. When massed military units are the norm, there becomes a 'right' way to play. The intricate trade-offs crucial to the game's design become far less meaningful. And the AIs aren't playing any 'better' as a result.

 

[Arlborn]

That's just wrong. The number of civilizations throughout history that have been overrun due to a lack of military or lack of willingness to fight far exceeds the number that have done the overruning.

Lack of military is one thing, probably or they didnt expect the attack or not enough population, but lack of willinness? Give me some expample of stabilished civilizations that let themselves be destroyed for 'lack of williness' 0.o Sure must be exceptions, but or the population just didnt like it rulers or there was some weird cases..

Ands about your other comentary, I like to lose yes. Why? Then I can improve and push my limits. It is specially nice when you play mostly peacefully as I do
And yes if the game is 100% won, it IS boring and I go search for challange elesewhere.

Best game is when you are in a race to win. Doesent matter if you lose or win, the race is real exciting.

 

[Airny]

This thread went so large, is Massive!

I just want to tell you, that I'm annoyed and happy at the same time, because all the strats and scores in the Hall of Fame are not reachable and comparable any more. I can almost throw away my knowledge of civ4. :/

 

[Roland Johansen]

Earlier I tried to use some Hammer cost examples to illustrate the point, but I dont think it was very clear. Let me attempt to simplify:

Lets assume that Civ 'X' pursues a balanced path of troops and improvements. He spends 5000 on troops and 5000 on improvements.

Civ 'Y', decides to try and attack his neighbors as his way of bettering his lot in life. He spends 9000 on troops, and 1000 on basic improvements. If he attacks, he has a 9000 to 5000 advantage or almost 2 to 1 in troops strength.

Enter the newer builds. Now, a typical defense for a builder must be closer to 20000 on troops and 5000 for improvements.

The attacker, spending the same amount of resources once again would have 24000 and 1000. Now his margin for attack is only 24000 to 20000 or only 1.2 to 1.

The reason? The cost of troops now completely smothers the cost of developments. There is no longer any trade off here. Civ 'Y' will almost always lose.

In the earlier case, there is a real decision to be made between development and troops. And if you go heavy on development, then you are trading on either 1) not being attacked or 2) that your developments will give you enough of a long-term edge to be able to win a protracted war against a less developed Civ possessing a larger military. You must balance one against the other.

In the latter case, its far harder to get a military advantage over Civs that are continuously pumping out military units. It is possible for the player to do, but MUCH harder for the AI to do so. Without his 2 to 1 advantage over the builder, the attacking AI will likely lose hard. And being behind on development as well, he will eventually fade out and die like many of the Aggressive AIs I see in the later builds.

Hopefully the above illustrates the point properly. When massed military units are the norm, there becomes a 'right' way to play. The intricate trade-offs crucial to the game's design become far less meaningful. And the AIs aren't playing any 'better' as a result.

You have been writing very many posts on this subject and I can say that I disagree with many parts of them. No I don't think that it would be a good thing if the AI would build 20-30 defenders per city. But I haven't seen the AI do this. Yes, I've seen cities with 20-30 units in them and even more. But those were not the defenders of that city. That was an attack stack stationed in that city. The normal number of defenders that I see is more around 8-10 (continents, huge map, aggressive AI, emperor level).

About your post above. In the middle ages a city can have a granary, barracks, market, grocery, forge, temple, aquaduct, monastery, cathedral, lighthouse, harbor, monument, theater and colloseum (maybe forgot something).
In 2.08, my well developed centre cities would have a majority of these buildings and 1 defender as they are not threathened to be attacked. The border cities would have 3 defenders and there would be an army (10 units) that could go to the rescue of a threatened city. Say on average 3 units per city and 8 buildings per city (I usually have less than 3 units per city in version 2.08 of this game but I might be building fewer units than the average player.) That would mean that I had spent about 5 times as much on buildings as on military. (Units are cheaper than buildings in this age, and some are upgraded old units and costed even less hammers but I don't take that into account.)
So I think the 5000 on buildings, 5000 on units is a gross misrepresentation of a middle age empire in version 2.08 of this game. Maybe you are far more militaristic than me, but you must have build some buildings in your cities in version 2.08 of this game.
In this version of the betterAI mod, I will build about 2 times as many units as in version 2.08 of civ4. So my construction speed on buildings suffers a bit, but not that much.

I'm not really concerned with the AI building many units as long as their construction of buildings doesn't suffer too much and as long as their unit upkeep doesn't kill their economy. Now there are some instances reported of this happening and I'm concerned about that, but not about all the other stuff that is mentioned as being negative about the AI building many units.

For instance, I think it's great that 5 turns of unit construction with your entire empire doesn't create a stack of units that is unstoppable for AI. You need more than that and that's a good thing in my opinion. Conquest of a few nations was seen as the easiest road to victory for the human player and if the AI is building more defensive units, then the conquest option is less attactive.

I also still see cities being captured by the various AI nations, so the defences are not impregnable. Some nations are conquering others. It's not easy and it's not a quick conquest, but there are some nations winning and some losing. On the other continent, there exists a nation that threatens to dominate that continent.

I'm eagerly awaiting the next version of the BetterAI mod to see how the military defensive and offensive AI was changed. In one of his post, Blake said something about letting the personality of the AI (aggressive or not) have an effect on the number of troops being build. I like that as it makes the civilizations more unique. As long as there is no nation that leaves its cities almost undefended.