Agg ais-slow teching-Observations and Reasoning

[peter450]

AGG AI was desinged to stop the AI being chop rushed at the start were a human would fast teck bronze/iron, chop some forest an get a very early strike force out, normal AI could be steam rolled by this were you could roll over 2 or 3 civ's, AGG AI you might roll one, but not two or three.

The downside to the AI's strong bias toward early units, is that it's start is sub optimal, to be honest i think all they need to do is sort out the DW system that seems to be a bit random to say the least and thats it, AGG AI is there for more warlike inclined players an for the way they play gives them a good challenge.

Standard AI playes a more balanced game, an i think is not the weak twin, although i think it's military spending should be marginally higher in both the very early (make it more difficult to rush) an late phases of the game (most tecks discovered, easy to wage long distance war at the end, more dangerous, more military spend needed)

 

[Roland Johansen]

Ok bug/exploit whatever, its just a name

BUT if I was attacked in this case, Ok sure, I'd probably whip the initial incursion point almost as far as I could, but then there's no way, I'd whip anymore in that fashion, I'd rather offer a city for peace, and keep my economy alive. There is a point when the whip frenzy becomes so counterproductive, as to be pointless. Most players could see this very quickly. The ai needs to be taught this.

If you have a dozen cities, can't afford them, and are further driving yourself into oblivion by massacring population, then why not offer one to save the rest and all importantly keep your economy alive? I would as a player. The ai wouldn't even consider this.(ok rarely. but only generally a really useless city later in the game). Someone decided that this was too "exploitable" in general. If you like to find and use exploits, then you will however hard programmers try to stop you

Anyways, apologies for the tone of any of my posts on the forum, made in the last couple of hours, that game really did get me (and Manning's looking unstoppable again)........

It seems that there were several problems for this AI except the fast expansion and the whipping. It apparently wasn't able to get you out of his lands because you could walk from one city to the next and he only whipped defenders. So I guess, he wasn't able to build other units then archers. Was he that unlucky with the resource distribution or didn't he have bronze working and animal husbandry?

This exploit/bug, will also mainly work on marathon speed. It just takes too much time to walk from one city to the next to get the AI to poprush and cities recover quicker at normal speed. Still, I guess the AI poprush behaviour is exploitable once you know that it will always do it when in trouble.

I am not in favour of the 'fix' that the AI should only whip defenders once in a war. Once players know that, conquering AI cities will become easy again.

I guess, the main problem for this AI was its inability to conduct a serious level of research because of its fast expansion and that resulted in other problems. I think this might again be connected to some of your settings. Settlers are relatively cheap at the marathon speed setting because units only cost twice as much while buildings cost three times as much as on normal speed (and there are three times as many turns). Furthermore, settlers move relatively quicker over the terrain as they have 3 moves for every 1 move at normal speed. All of this results in a relatively fast expansion speed at the marathon speed setting.
The problem is further compounded by the slower research (increased research costs) at huge maps. So the AI's level of development will be relatively backward compared to its level of expansion if it uses the same behaviour routines as on normal speed and normal sized maps.

About your peaceful game: Did nobody wage war or did nobody wage war with you?

Are you using Solvers unofficial patch now?

Oh and I noticed that I know nothing about American football. Who are these guys that you're talking about?

I continued with my game, it's 1000 AD now, and still no-one declared war on me (clearly, I play slower or less that you do as you've finished some games in the same time). There has been one war between 2 AI's which was relatively short and ended in one city captured. These 2 AI's have recently gone to war again. The wars haven't hurt them significantly. All of the AI's that I know have the same religion except one of them and that probably explains the peacefulness. One of the 2 civilisations that has been in the 2 wars was the civilisation that has a different state religion.

I'm researching Printing Press now and have all the previous technologies. Nice technologies like Liberalism, Economics, Replaceable Parts, Rifling and Democracy are going to be researched soon. The other AI's are at a similar level of technological advancement.

No strange AI behaviour so far in my game. But it's probably very dependent on how the AI performs at the starting stages of the game. If it self destructs there then it cannot recover. Or at least that seems to be true in some of your games.

I have been unlucky in the availability of strategic resources, no iron or horses or ivory, only copper and I already have 17 cities. The 8 further cities that I'm planning to build are also not going to give those resources to me. So I can't build pikemen, knights, crossbowmen and war elephants. This makes war a bit risky for me, so I haven't started one yet. Especially since my opponents are well defended (more units than me) and they do have a varied army. Maybe, I should try a war with riflemen. I normally would start a war earlier, but I had a rough starting position which slowed me a bit and I do have plenty of room for cities which makes war a less attractive option than peaceful expansion. And my neighbour has the statue of Zeus which doubles any war weariness.

 

[DrewBledsoe]

It seems that there were several problems for this AI except the fast expansion and the whipping. It apparently wasn't able to get you out of his lands because you could walk from one city to the next and he only whipped defenders. So I guess, he wasn't able to build other units then archers. Was he that unlucky with the resource distribution or didn't he have bronze working and animal husbandry?

Yes he had BW otherwise he wouldn't have been able to whip BUT he didn't have any Copper. He had horses and some chariots, but once I'd pillaged the resource, and he'd expended all his horses in the initial incursion, all he had was archers.

This exploit/bug, will also mainly work on marathon speed. It just takes too much time to walk from one city to the next to get the AI to poprush and cities recover quicker at normal speed. Still, I guess the AI poprush behaviour is exploitable once you know that it will always do it when in trouble.

Exactly, many a time I've just wandered past a slaughtered (whipped) city, deeming it to big a loss for my troops, and gone on to another city. I'm not meaning specifically to exploit anything in this type of action, but it has the same effect.
I am not in favour of the 'fix' that the AI should only whip defenders once in a war. Once players know that, conquering AI cities will become easy again.

I guess, the main problem for this AI was its inability to conduct a serious level of research because of its fast expansion and that resulted in other problems. I think this might again be connected to some of your settings. Settlers are relatively cheap at the marathon speed setting because units only cost twice as much while buildings cost three times as much as on normal speed (and there are three times as many turns). Furthermore, settlers move relatively quicker over the terrain as they have 3 moves for every 1 move at normal speed. All of this results in a relatively fast expansion speed at the marathon speed setting.
The problem is further compounded by the slower research (increased research costs) at huge maps. So the AI's level of development will be relatively backward compared to its level of expansion if it uses the same behaviour routines as on normal speed and normal sized maps.

Settlers now cost 300 on Marathon, this is not cheap 1.5 settlers = the Great Wall! The problem may have been compounded because Cyrus is IMP (hence cheaper settlers), but I've also had Ragnar and a few other ais do this too (all forests seemingly chopped for settlers and cities that can't realistically yet be supported)
About your peaceful game: Did nobody wage war or did nobody wage war with you?
Wars were the order of the day. As I said, the Dutch lend themselves to "peaceful" play better than anyone. I was just extremely lucky, to have Brennus's religion (as did everyone on my continent), thus all were at least pleased with me (on our continent), and someone else always seemed a better bet to pick upon. As I said, probably a game in one thousand.
Are you using Solvers unofficial patch now?
No, not yet, because from correspondence with others, it breaks corporations to the other extreme (I think minus the extreme inflation, which Alexman said is being addresses in the patch, they work well enough already for huge/marathon). Plus apparently a patch is due in 2 weeks or so.
Oh and I noticed that I know nothing about American football. Who are these guys that you're talking about?
Ha Don't worry, the NFL season started Thursday Night. Peyton Manning is the QB for the Indianapolis Colts, last year's Superbowl winners. He, and they looked very good again in Thursday's opener.
I continued with my game, it's 1000 AD now, and still no-one declared war on me (clearly, I play slower or less that you do as you've finished some games in the same time). There has been one war between 2 AI's which was relatively short and ended in one city captured. These 2 AI's have recently gone to war again. The wars haven't hurt them significantly. All of the AI's that I know have the same religion except one of them and that probably explains the peacefulness. One of the 2 civilisations that has been in the 2 wars was the civilisation that has a different state religion.
I work mostly at home, and do so in huge spurts of work, and thus give myself a lot of free time (and spend a lot on playing civ--often don't play for long stretches when Im really busy)
I'm researching Printing Press now and have all the previous technologies. Nice technologies like Liberalism, Economics, Replaceable Parts, Rifling and Democracy are going to be researched soon. The other AI's are at a similar level of technological advancement.

No strange AI behaviour so far in my game. But it's probably very dependent on how the AI performs at the starting stages of the game. If it self destructs there then it cannot recover. Or at least that seems to be true in some of your games.

I have been unlucky in the availability of strategic resources, no iron or horses or ivory, only copper and I already have 17 cities. The 8 further cities that I'm planning to build are also not going to give those resources to me. So I can't build pikemen, knights, crossbowmen and war elephants. This makes war a bit risky for me, so I haven't started one yet. Especially since my opponents are well defended (more units than me) and they do have a varied army. Maybe, I should try a war with riflemen. I normally would start a war earlier, but I had a rough starting position which slowed me a bit and I do have plenty of room for cities which makes war a less attractive option than peaceful expansion. And my neighbour has the statue of Zeus which doubles any war weariness.

Sounds an interesting game, Recently, I've found myself able to win by having far fewer cities than some ais, but making them better cities, and then soaking up any aggresion made against me (except in that 1-1000 Dutch game).

Thanks for interesting reply

 

[Roland Johansen]

Sounds an interesting game, Recently, I've found myself able to win by having far fewer cities than some ais, but making them better cities, and then soaking up any aggresion made against me (except in that 1-1000 Dutch game).

Thanks for interesting reply

Oh, I was wrong about the settlers only costing double the cost of normal speed. Apparently, they cost triple as they should. It shows that I don't have any experience with the marathon speed setting. Still, relatively faster movement at marathons speed will increase the speed of expansion. And the relatively slow research speed at huge maps also doesn't help for a balanced AI gameplan.

I really think that the lack of flexibility of the AI coding relative to the gamespeed makes it behave worse at the extreme settings. Higher difficulty levels like we're playing of course overcomes some of these problems for the AI, but it remains a basic problem for the AI. It is designed to play optimal under certain circumstances.
For instance: If you modded the game in such a way that there existed a starting technology called urbanisation which would require a huge research cost before you could research anything else, then I wouldn't be surprised if the AI would overexpand before finishing this research and thus doom itself.
The balance between the various AI build options is based on certain assumptions on how a game develops and in modded games or under somewhat extreme game settings, the game develops a bit different and the game plan that was coded into the AI is not working that well.

 

[DrewBledsoe]

Oh, I was wrong about the settlers only costing double the cost of normal speed. Apparently, they cost triple as they should. It shows that I don't have any experience with the marathon speed setting. Still, relatively faster movement at marathons speed will increase the speed of expansion. And the relatively slow research speed at huge maps also doesn't help for a balanced AI gameplan.

I really think that the lack of flexibility of the AI coding relative to the gamespeed makes it behave worse at the extreme settings. Higher difficulty levels like we're playing of course overcomes some of these problems for the AI, but it remains a basic problem for the AI. It is designed to play optimal under certain circumstances.
For instance: If you modded the game in such a way that there existed a starting technology called urbanisation which would require a huge research cost before you could research anything else, then I wouldn't be surprised if the AI would overexpand before finishing this research and thus doom itself.
The balance between the various AI build options is based on certain assumptions on how a game develops and in modded games or under somewhat extreme game settings, the game develops a bit different and the game plan that was coded into the AI is not working that well.

Agreed completely. Don't get me wrong, some games have been really entertaining, and interesting, just some have been spoiled due to aspects mentioned in previous posts of this thread.

I'm hoping the patch will fix at least a few aspects, then when I've played with the patch for a couple of months, I'll probably start a specific marathon/huge "fix" of my own.

 

[Nor Me]

The AI does stop building attacking units when it thinks its military costs are too high. It does this by calculating the percentage of its per turn costs that are spent on unit support and stops training units when this goes over a certain number (depending on personality, wars, etc. Agg vs Non-agg AI counts a mere 4%).

One of several flaws with this is the denominator: the number of units (above those it has free support for) it is prepared to build is proportional to total costs rather than empire size or economy. You don't need me to tell you that the big early game cost is maintenance and that is rises faster than linearly with the number of cities.

So an AI civ, like Joao in the OP, who over-expands is prepared to build more units, even compared to the size of his empire than a smaller civ. A small civ, even with a sound economy and at war, might be building nothing in its cities in the same position.

I've been testing modding the unit cost percentage to something different by trying to include upgrade costs, which are much bigger than support costs on higher levels, and using total science and income rather than costs.

For instance: The AI has a chance to declare war dependent on the size of the military of it and its potential enemy, the setting aggressive AI or normal AI, the distance of the various civilisations and the diplomatic relations. This chance is probably checked every turn. But because the marathon speed setting has far more turns, the chance that it will declare war in one of those many turns is much higher. I'm not sure if it works this way as I haven't looked at the code, but I wouldn't be surprised if this was one of the issues.

It does work like this. The biggest difference between Aggressive AI and normal AI is that it doubles this chance. So non-aggressive AI on marathon is possibly more similar to aggressive AI than not on normal speed. Aggressive AI is much more balanced on normal speed. Would scaling this per turn chance down with the increased unit cost make sense?

 

[Roland Johansen]

The AI does stop building attacking units when it thinks its military costs are too high. It does this by calculating the percentage of its per turn costs that are spent on unit support and stops training units when this goes over a certain number (depending on personality, wars, etc. Agg vs Non-agg AI counts a mere 4%).

One of several flaws with this is the denominator: the number of units (above those it has free support for) it is prepared to build is proportional to total costs rather than empire size or economy. You don't need me to tell you that the big early game cost is maintenance and that is rises faster than linearly with the number of cities.

So an AI civ, like Joao in the OP, who over-expands is prepared to build more units, even compared to the size of his empire than a smaller civ. A small civ, even with a sound economy and at war, might be building nothing in its cities in the same position.

I've been testing modding the unit cost percentage to something different by trying to include upgrade costs, which are much bigger than support costs on higher levels, and using total science and income rather than costs.

It seems to me that the AI shouldn't spend more than a certain percentage of its income on units, not more than a certain percentage of its spending (just like you say, I agree).

The best way (for me) would be if the AI would calculate what its maximum income is at 100% gold and then take a certain percentage of that number (dependent on aggression level of this AI) and never use more than that percentage of its maximum income on units. This percentage may be higher when at war.

There should be a few exceptions to this rule so that the AI is still able to build a single defender for a new city and such things.

In the same way, an AI should stop expanding by conquest or settling if more than a certain percentage of its maximum income is used for civic+city upkeep.

In all cases, inflation should be added to the basic cost before calculating whether the maximum percentage is being used.

It does work like this. The biggest difference between Aggressive AI and normal AI is that it doubles this chance. So non-aggressive AI on marathon is possibly more similar to aggressive AI than not on normal speed. Aggressive AI is much more balanced on normal speed. Would scaling this per turn chance down with the increased unit cost make sense?

I would adjust it directly for the number of turns in the game, not the unit cost. In that way, you'll get a similar number of war declarations during one marathon game as during one normal speed game.

Note that such an adjustment is not perfect. Example:

At marathon speed, a game has three times as many turns as on normal speed. If you want an equal number of wars during an immortal game, then you might think that dividing the percentage chance of war (each turn) by three would work.

Assume that the percentage were 30% at normal speed. Then the marathon percentage would be 10%. The chance that there is a war declaration during a set of three subsequent turns at marathon speed then is 27.1%, not 30%.

Assume that the percentage were 90% at normal speed, then the marathon percentage would be 30%. The chance that there is a war declaration during a set of three subsequent turns at marathon speed then is 65.7%, not 90%

Assume that the percentage were 9% at normal speed, then the marathon percentage would be 3%. The chance that there is a war declaration during a set of three subsequent turns at marathon speed then is 8.7%, not 9%.

As you can see, adjusting the percentage linearly for number of turns works quite ok as long as the chance of a war declaration during a turn is small. If the chance is large, then such an adjustment is pretty poor.
(The above statement can be mathematically proven. The adjustment works better at small percentages.)

Since wars don't erupt every turn, I guess that the chances aren't that large and such an adjustment would would pretty well.

 

[DrewBledsoe]

The AI does stop building attacking units when it thinks its military costs are too high. It does this by calculating the percentage of its per turn costs that are spent on unit support and stops training units when this goes over a certain number (depending on personality, wars, etc. Agg vs Non-agg AI counts a mere 4%).

One of several flaws with this is the denominator: the number of units (above those it has free support for) it is prepared to build is proportional to total costs rather than empire size or economy. You don't need me to tell you that the big early game cost is maintenance and that is rises faster than linearly with the number of cities.

So an AI civ, like Joao in the OP, who over-expands is prepared to build more units, even compared to the size of his empire than a smaller civ. A small civ, even with a sound economy and at war, might be building nothing in its cities in the same position.

I've been testing modding the unit cost percentage to something different by trying to include upgrade costs, which are much bigger than support costs on higher levels, and using total science and income rather than costs.



It does work like this. The biggest difference between Aggressive AI and normal AI is that it doubles this chance. So non-aggressive AI on marathon is possibly more similar to aggressive AI than not on normal speed. Aggressive AI is much more balanced on normal speed. Would scaling this per turn chance down with the increased unit cost make sense?

Thanks very much for this It's given me a definite way to start improving matters (as I said once its been tested with the new patch).