AIs and the Art of War

It would be nice to see at least the aggressive AI having a more traditional medieval siege force. Or a Defensive AI building up Castle related buildings.
Click to expand...
That's what flavors are all about really. But I don't know how much flavors now influence ai builds and other choices.

@Koshling: ? How much do flavors now play a role in building selections and warfare behaviors? Do they weight things at all anymore? It'd be nice if they played a minor role in tinting the behavior somewhat, particularly after some adjustments soon to be made to deepen that mechanism a touch. Obviously it shouldn't be the primary determination in those things by any means... just a slight weighting would be nice.
 
Hydromancerx said:
Medieval Era spans X35 to X41. Here is a timeline of the units with strength ...

X35 - Guard (9), Warden (10), Flailman (10), Longbowman (8), Crosbowman (9), Trebuchet (8), Cog (6)

X36 - None

X37 - Shaolin Monk (12), Heavy Swordsman (13), Heavy Pikeman (9), Ancient Flamethrower (12), Ancient Hand Cannon (12), Galleass (8)

X38 - Heavy Calvary (13), Carrack Merchant (6)

X39 - Knight (11), War Elephant (12), Caravel (9)

X40 - Mailed Knight (14)

X41 - Adventuerer (13), Sloop of War (14)

Note this list doesn't count Heroes or culture units.
Click to expand...

But the Town Watchman is Early Ancient. It may well be that the later units that the watchman upgrade into are not so much of a problem for the AI to figure how best to use.
 
Hydromancerx said:
Do you think that is the issue? That the Town Watchmen are the best for the era so they make them?

I also do agree that I have seen the AI do better when it comes to attacking back in the early days of the project and even RoM/AND. However more recently the AI has been doing a bit better than it was doing say around v25 or v26.

It would be nice to see at least the aggressive AI having a more traditional medieval siege force. Or a Defensive AI building up Castle related buildings.
Click to expand...

From what Koshling posted (i'd have to say NO) but this is the feeling/(seeing games in play) i have, sorry.

I am trying a NEW game with Raging Barbs on, and this seems to be where the core should be presently.

I am also presently working on a re-structure of the units per Era in the file CIV4EraInfo or CIV4HandicapInfo. (Cant remember which one right now):blush:
 
Koshling said:
In my experience, provided crime is reasonably under control, it does choose archers once it has the archery promoting buildings. Recently however, the addition of crime to slave specialists has broken ai crime management, so it rarely is under good control and it winds up having to continuously build town watchmen.
Click to expand...

I did not want to have the slave specialist have crime (or disease) on them. Unhappiness and unhealthiness i could understand but not the crime. Should I remove it?
 
Dancing Hoskuld said:
I did not want to have the slave specialist have crime (or disease) on them. Unhappiness and unhealthiness i could understand but not the crime. Should I remove it?
Click to expand...

I dont see why the specialist would have any UNhappiness or UNhealthiness?? Back WHEN, they were actually better off than the peasants were. At least THEY had a decent place to live, eat food, all they needed to do was WORK for their Dominus.
 
I found some AI behaviour that I think should be improved:
As the save shows, my stack is parking on a hill near the leading AIs capital and I also have some Ambushers around the in erea. Not only does the AI (Babylonians) send scouts and hunters there that all get killed each round (it doesnt "remember" losing units in that erea, so it doesnt avoid it after a few losses) but most importantly they try to resettle the city I razed there again and again by sending a settler together with just one weak stone axe right past my big stack. So what AI lacks is: scouting for hidden units with trained dogs before it sends in other units; realizing that sending a settler in a region with a dangerous enemy stack (AI could evaluate the strength as it was on its territory) is no good idea, let alone with just one weak escort. (playing the latest SVN I also saw one tribe being ecorted by just 2 healers)
Sooo what could it do otherwise? A) build a huge stack to clear out the enemy stack before trying to resettle the occupied erea B) leave enemy there until better units are available and settle towards other directions (the Babylonians could have settled some islands or conquered the Dutch) C) trying to create a "Über-Unit", attach a GG to a bear rider or something and also send a great commander next to that unit as well as other good units (the Babylonians, long time tech leader built the Deer-, Bear- and Giraffe-Trainer I think whereas my best unit was Atl-Atls [starting on GEM is so much easier because there the AI don't get free techs when you start immortal and turn deity first turn].

Another thing are chockepoints. Its just so easy to block enemy access to whole parts of the continent by stationing one unit on a chockepoint. I can see that its hard for it to realize it just has to sacrifice one unit of horses and the next could clear the chokepoint... (if it understands that it is a chokepoint that when occupied by an enemy unit blocks its expansion path but then again AI had some ships in its port so why not load units in there and drop them off behind the chockepoint as if its a new continent or island, can't this routine be added? On one hand it feels to be cheating to block chockepoints to not be attacked but on the other hand there is no other way to avoid the huge enemy stacks or to come by them else than using chockepoint blocking or having a GG attached to a unit and a great commander close to it...

as I sugegsted a bit more AI naval invasion in the Hinterland might tweak it in the short term... It doesnt have to be whole inavasions, dropping of a unit or two and pillaging can also be devastating (at least for players like me who only leave one garrison in their cities for the most time early on)


Regarding the pathing system. What about this: once a AI realizes that the best path is occupied it tries to find another path if no other path is possible it evaluates what strength it has to invest clearing the shortest path and then it reavaluates which strenngth i would have to invest on alternative paths (so the human just can't block the shortest path with many troops and the alternative path with just one) If more than 2 paths are blocked the one with the lowest strength is the attacking point. Now what Humans do is to trick enemy stacks - for example you see that the AI sends it massive stack to your weakest chockepoint what you do? You open one chockepoint so the AI stops its pathing to the weak point and tries to go thru the now open erea. After a few turns you block the path again and AI goes back to the weak chockepoint and so on - by opening closing you can trick AI in stupid behaviour. Best would be that once AI has evaluated a chokepoint decision (evaluating the strength needed to clear it by attack) it sticks by it, instead of canceling the order once a new path is possible...And once a chokepoint is cleared it will occupy it. Also, getting it to preemptively occuping chokepoints itself would be nice.
 

Attachments

  • AI settler behaviour and chockepoints.7z
    2.1 MB · Views: 169
Hydromancerx said:
Medieval Era spans X35 to X41. Here is a timeline of the units with strength ...

X35 - Guard (9), Warden (10), Flailman (10), Longbowman (8), Crosbowman (9), Trebuchet (8), Cog (6)

X36 - None

X37 - Shaolin Monk (12), Heavy Swordsman (13), Heavy Pikeman (9), Ancient Flamethrower (12), Ancient Hand Cannon (12), Galleass (8)

X38 - Heavy Calvary (13), Carrack Merchant (6)

X39 - Knight (11), War Elephant (12), Caravel (9)

X40 - Mailed Knight (14)

X41 - Adventuerer (13), Sloop of War (14)

Note this list doesn't count Heroes or culture units.
Click to expand...

Right, and guards get +50% city defense, so thy are rally 12 strength in the city defender role + crime fighting. I suspect they are considered the best (city) defensive unit for that reason throughout this entire period. Build cost also plays a role - what are the relative costs?
 
DRJ said:
I found some AI behaviour that I think should be improved:
As the save shows, my stack is parking on a hill near the leading AIs capital and I also have some Ambushers around the in erea. Not only does the AI (Babylonians) send scouts and hunters there that all get killed each round (it doesnt "remember" losing units in that erea, so it doesnt avoid it after a few losses) but most importantly they try to resettle the city I razed there again and again by sending a settler together with just one weak stone axe right past my big stack. So what AI lacks is: scouting for hidden units with trained dogs before it sends in other units; realizing that sending a settler in a region with a dangerous enemy stack (AI could evaluate the strength as it was on its territory) is no good idea, let alone with just one weak escort. (playing the latest SVN I also saw one tribe being ecorted by just 2 healers)
Sooo what could it do otherwise? A) build a huge stack to clear out the enemy stack before trying to resettle the occupied erea B) leave enemy there until better units are available and settle towards other directions (the Babylonians could have settled some islands or conquered the Dutch) C) trying to create a "Über-Unit", attach a GG to a bear rider or something and also send a great commander next to that unit as well as other good units (the Babylonians, long time tech leader built the Deer-, Bear- and Giraffe-Trainer I think whereas my best unit was Atl-Atls [starting on GEM is so much easier because there the AI don't get free techs when you start immortal and turn deity first turn].

Another thing are chockepoints. Its just so easy to block enemy access to whole parts of the continent by stationing one unit on a chockepoint. I can see that its hard for it to realize it just has to sacrifice one unit of horses and the next could clear the chokepoint... (if it understands that it is a chokepoint that when occupied by an enemy unit blocks its expansion path but then again AI had some ships in its port so why not load units in there and drop them off behind the chockepoint as if its a new continent or island, can't this routine be added? On one hand it feels to be cheating to block chockepoints to not be attacked but on the other hand there is no other way to avoid the huge enemy stacks or to come by them else than using chockepoint blocking or having a GG attached to a unit and a great commander close to it...

as I sugegsted a bit more AI naval invasion in the Hinterland might tweak it in the short term... It doesnt have to be whole inavasions, dropping of a unit or two and pillaging can also be devastating (at least for players like me who only leave one garrison in their cities for the most time early on)


Regarding the pathing system. What about this: once a AI realizes that the best path is occupied it tries to find another path if no other path is possible it evaluates what strength it has to invest clearing the shortest path and then it reavaluates which strenngth i would have to invest on alternative paths (so the human just can't block the shortest path with many troops and the alternative path with just one) If more than 2 paths are blocked the one with the lowest strength is the attacking point. Now what Humans do is to trick enemy stacks - for example you see that the AI sends it massive stack to your weakest chockepoint what you do? You open one chockepoint so the AI stops its pathing to the weak point and tries to go thru the now open erea. After a few turns you block the path again and AI goes back to the weak chockepoint and so on - by opening closing you can trick AI in stupid behaviour. Best would be that once AI has evaluated a chokepoint decision (evaluating the strength needed to clear it by attack) it sticks by it, instead of canceling the order once a new path is possible...And once a chokepoint is cleared it will occupy it. Also, getting it to preemptively occuping chokepoints itself would be nice.
Click to expand...

Good observations, but there is little to no low-hanging fruit here:

1) Memory of recent danger is available, but only used in extremely limited ways (almost exclusively by 0-defense units like workers, and then only to ensure thy have an escort). The use of this information could be extended more generally, but it's tricky to do and may have unintended side-effects

2) There is currently no mechanism in the AI to make units take cooperative actions (like scouting before settling). I'll be adding the basis for this with the S&D AI underpinning, but until that is done there is really no good way to tackle this apart from forcing the stack concerned to wait for a see-invis unit and include it in the stack (which is far from optimal)

3) Choke-points are a difficult problem because units have no memory. They evaluate paths afresh every turn. Some sort of deeper where-was-I-going-and-why mechanism is required. An element of that is available but it's very lightly used, and not really deep enough to solve this sort of problem.

I think it can ultimately all be addressed, but each of the above is a lot of work, and likely to have a lot of consequences to settle down afterwards, so I'd say that ach would be the focus for a full release cycle on the AI side. In light of the amount of time I have, and the need for critical bug fixes to take priority we can maybe look to address all of the above by about release 35 or 36 or so.
 
Thanks for taking notice of this and I am looking forward to your further AI improvements!

By the way what really impressed me was the retreat mechanics you added, they prevent the AI to lose commanders and small stacks that wander around your cities are much more dangerous now, as you have to keep some troops close by so they dont come back once you turned your back on them^^
 
I do find it amusing that not too long ago that the build cost of units was increased in order to prevent the AI to have an overabundance of them which would drain their economy and stagnate them under high unit maintenance cost and horrible city infrastructure.

I think it tended to favour canine units then, which was amusing to see the army of dogs descend upon me.
Crime was recently introduced which tended to destroy the AI by the late classical period...
 
strategyonly said:
I dont see why the specialist would have any UNhappiness or UNhealthiness?? Back WHEN, they were actually better off than the peasants were. At least THEY had a decent place to live, eat food, all they needed to do was WORK for their Dominus.
Click to expand...

Back when, ie classic era and earlier, most slaves were owned by the peasants, just because there were a lot more peasants than there were rich people. There was not much difference between the living conditions of the slaves and the peasants that owned them. Exceptions occurred when the Pharaoh/king/emperor decided a grand construction was needed.
 
Koshling said:
In my experience, provided crime is reasonably under control, it does choose archers once it has the archery promoting buildings. Recently however, the addition of crime to slave specialists has broken ai crime management, so it rarely is under good control and it winds up having to continuously build town watchmen.
Click to expand...

That is bad, I was unaware that DH had done that with slaves. The AI needs to have crime under control to do economically well, and if it can't do that anymore we've lost the economic progress you made last September. Maybe the AI should use fewer slave specialists for the time being.
 
Koshling said:
Right, and guards get +50% city defense, so thy are rally 12 strength in the city defender role + crime fighting. I suspect they are considered the best (city) defensive unit for that reason throughout this entire period. Build cost also plays a role - what are the relative costs?
Click to expand...

| Strength | Base Cost
Archer|5|25
Town Watchmen|5|40
Spearman|5|35
Javlineer|5|35
Arsonist|8|100
Maceman|5|45
Axeman|5|40
Light Swordsman|7|80
Light Crossbowman|6|50
Pikeman|6|95
Skirmisher|6|95
Ancient Rocketeer|6|60
Swordsman|10|110
Guard|9|120
Flailman|10|130
Longbowman|8|100
Crossbowman|9|120
Heavy Axeman|13|135
Heavy Swordsman|13|135
Heavy Pikeman|9|135
Ancient Flamethrower|12|125
Ancient Hand Cannon|12|80
Heavy Crossbowman|12|200
Arquebusier|15|200
City Guard|19|210
Musketman|20|260
Rifleman|26|370
 
Dancing Hoskuld said:
Back when, ie classic era and earlier, most slaves were owned by the peasants, just because there were a lot more peasants than there were rich people. There was not much difference between the living conditions of the slaves and the peasants that owned them.
Click to expand...

You are right. Funny thing btw, there is a new show out there called Plebs, that exactly shows that. It's really worth a watch (if you like "life of brian" style of humor) and I think other civ players will enjoy it as well.

4 episodes have allready been released -(it airs on english ITV I think),
the torrents are on the bottom of the page I linked

Cheers!
 
Just a question on the subject of TW units. The change to Civilian from Melee... did this actually remove them from qualifying for the xp given by the Great Military Instructors? I thought I'd read that it had. Or was it just that the change eliminated the added xp from those sources giving xp to melee units?

I'm wondering if this actually helps the situation. Our early city raider specialist units such as the Stone Macemen for example, have a fair bonus against Melee but nothing grants a bonus against Civilian. As a melee, the TW may get more xp, but as a Civilian they're immune to a lot of effects that could put them at a disadvantage. I'm wondering which is a more severe effect. Perhaps the answer would be to offer the Macemen units a bonus against Civilians equivalent to the bonus they have against melee and to give melee and riding units an access to a promotion line that gives a combat modifier VS Civilians (like shock does against melee.)

@Koshling: Would the AI realize what these changes would mean to a proper evaluation of the benefit of a TW? Would such changes get them back into the AI view that they're role is best 'for controlling crime' over the role of 'Primary City Defender'?
 
No civilian units still get exp from military instructors - I noticed it when the story tellers were starting with more exp than expected.
 
Yeah, I thought they might've been from what I'm seeing in my game here but I hadn't done the calculations to see. I don't think that's really a bad thing at this time anyhow.
 
strategyonly said:
What if we put the base cost of the TW double?
Click to expand...

No leave it. No knee-jerk reactions here please until we see how adding AI for properties on specialists works out (that should make the AI less fixated on slave specialists, which are causing a lot of its crime currently)
 
You must log in or register to reply here.

Similar threads

Slyceth
  • Locked
(7-VT) Pledging fealty to an AI player to cure constant frustratingly petty war declarations.
Replies
5
Views
268
axatin
A
PDX Firefly
Developer Diary | Art of the Ages
Replies
3
Views
659
Prester John 2
Prester John 2
Yakostovian
Invitation to the Mafia Champion - Season 11
Replies
0
Views
313
Yakostovian
Yakostovian
M
AI Declaring War
Replies
6
Views
596
martindo
M
Xaviarlol
[DLL] (7-57) Bombing cities beyond their HP has a chance to destroy Buildings
2
Replies
20
Views
564
Xaviarlol
Xaviarlol
Top Bottom