AI, North and South

[Spharv2]

#2. See what happens if there is a revamp in VP (to expand on Clovis' tests before asking for it to be included in an official patch). Transportation hubs and important rail lines should have VP values to encourage the AI to take and secure them, as well as to show their importance to the military cause (even some non-urbanized centres could use some VPs).

Yeah, the reason the Northern armies get surrounded so easily is that we as humans recognize the importance of Manassas Junction, and the defense as the South is usually centered there. The AI refuses to recognize the importance and sees an easier and more direct path south towards Fredricksburg and takes it, usually leaving a pretty strong CS army in it's rear.

I like the idea of an additional variable to the setup menu to add a modifier to the AI's activation rolls, that would definitely help the AI out a good bit.

[PBBoeye]

I am going to pre-answer my own question here, and if I'm incorrect someone can point it out.

But I am guessing that it's impossible to have events change the VP value of any given region. Is that correct?

One thing I enjoyed about HoI/HoI2 was that you could install AI files by means of event scripts that would change weight values all over the map for provinces (regions). This way the AI could be led by the nose if certain circumstances were in play (say, France conquered, USSR attacks Germany, etc).

Without that ability, the game looks at everything from a static viewpoint, so there is no progression that can be scripted.

Now I'd love to be corrected here, so fire away.

[McNaughton]

I am going to pre-answer my own question here, and if I'm incorrect someone can point it out.

But I am guessing that it's impossible to have events change the VP value of any given region. Is that correct?

One thing I enjoyed about HoI/HoI2 was that you could install AI files by means of event scripts that would change weight values all over the map for provinces (regions). This way the AI could be led by the nose if certain circumstances were in play (say, France conquered, USSR attacks Germany, etc).

Without that ability, the game looks at everything from a static viewpoint, so there is no progression that can be scripted.

Now I'd love to be corrected here, so fire away.

I currently have not found any information via events regarding affecting cities, but I am guessing that it is possible. I am thinking of the event that causes the destruction of the railway near Baltimore, as it directly affected a specific province.

So, you can target a specific province, and affect the rail line. We know this for sure. It may be possible to affect city victory points, but we don't have any info because there isn't anything like this currently in the event files. Something to ask the developers...

[PBBoeye]

Only negative I can foist at the team is the lack of documentation for events and so forth. I think that is a pretty important piece of info to have.

[Zoetermeer]

I tried setting the options to the following:

Low aggressiveness
Small FOW bonus
No activation check

This made a huge difference with the Union AI, as others have already reported. I think this is most likely because of the activation rule being off. This is honestly the first game I've played where I actually feel like there's a war on - the Union is acting like the Union, like it actually wants to invade and conquer. The AI organized two large armies in Alexandria - one under Fremont with good commanders (Sheridan, Hooker, etc.), and one under Butler with a corps under Sumner. It sent Butler down to Fredericksburg, where I had Lee entrenched, and tried to cross the Rappahannock to the East of me and hit me in the flank. It also sent Fremont to attack Jackson's corps at Harper's Ferry. For some reason, the AI doesn't seem to have any interest in taking Manassas, even though I've pretty much surrendered it to him in favor of keeping my forces concentrated south of the Rappahannock.

I noticed one other thing here - the AI in the Eastern theater seems to place such a high priority on keeping control of Harper's Ferry, I almost think it's a good idea to use a house rule to not try and take Harper's Ferry unless you are trying to invade Maryland as the CSA player. That way the AI won't divert critical manpower from the Army of the Potomac in trying to retake this position. It's probably not very "historically realistic" to try and take HF as the CSA anyway if you are playing a defensive game.

The AI also did some things I didn't think it was capable of doing, like landing sizable forces near the Hatteras Inlet to (presumably) disrupt things there and maybe interrupt the flow of supplies north to Richmond. Another thing it did that was a pleasant surprise was actually settle its forces into a reasonable position and sit out the harsh winter months. When snow started to come down in the east, it consolidated its armies around Alexandria and kept them there for the duration of the winter months - and likewise in the west.

Most of the places I'd taken or was laying siege to that the AI had been indifferent to (most notably Fortress Monroe) were immediately addressed by the AI when I changed these settings (I changed them mid-game).

[Jacek]

Yes, Zoetermeer, these settings really CHANGE a lot and have produced my most challenging campaign opening so far. Another thing that Union AI does under these settings is imoving its commanders across theaters. You have seen Fremont move to Alexandria, I have seen Banks lead the assault of Prestonburg, KY.

[Zoetermeer]

I noticed that also - the AI moved Grant to the east sometime during the winter between 1862 and 1863. I'm just about to start into the campaign season of '63, and it looks like the AI has used the winter months to really build up some heavy-hitters to come after me. Pretty exciting!

[Pocus]

AI moving leaders trans-theaters has been majorly upgraded in 1.06, this should not be affected by the settings.

So do you think this is the activation rule which is the biggest problem to AI operations? I can check specifically how activation influence AI behavior, but my time is short these days.

You can change VP values of regions on the fly, the script is SetVP within an event.

[PBBoeye]

It would seem that AI activation is a real issue for the ability of the AI to respond. 'Respond' would be the word, as it cannot plan very far ahead, as we human players do. So activation becomes key as the AI must go 'NOW!'. Explains why so often human CSA players can surround and press McDowell and destroy him so easily - well, it is a part of the answer.

I think the suggestions that McNaughton and I made in this post from the thread probably are the best angle to keep the integrity of the game design in play:
http://www.ageod-forum.com/showpost.php?p=42450&postcount=30

[veji1]

There are obviously many problems with the AI, and faced with the level of complexity in the game, we might have to make it "cheat" to some extent...

There are 2 changes I would like to see in the way the AI does things, but they may prove to complicated and then I guess only a clever an limited way of "cheating" would make a difference :

1/ some sort of code routine that would go on from mid-december to mid-february that would see the AI go on the full defensive mode and do a general reshuffling of its generals according to there abilities : It would check all 3* commanding armies an without a command, and put the best ones in, then same for the 2* and then for the 1*.. Obviously to do that a "relocate generals" button would be necessary, and it would need to span several turns to avoid army disbanding, etc...

The way it would go the USA army would act that way
- end of early december turn : order all forces to go on defensive posture near or in cities.
- late december turn : check generals without command and generals in command, relocate good generals without command to replace the worse generals in command (say the has 4 armies, and has now 2 spare 3* generals : it would check qualities of the army leaders and if an army leader is worse then one of the free gens : relocate the free gen for it to replace the bad one).
- repeat as long as neede so that the AI ends up with an "optimal" organisation with the best generals in command.

If this needs to be then in several turns (one turn to relocate, then another one to change command) this would need to happen only in winter when the game is somewhat stalled.

But if the whole manoeuver coud be done in 1 turn only, the ideal would be for such checks to occur at the beginning of each turn. Say on turn "early april 1862" a bunch of 2 and 1* generals appear, then the first things the AI does is to sort out which are the "worst" 1* and 2* in command and replacing them with the new ones... A great improvement.

Coupling that with the fact that the AI should promote the generals whenever possible and integrate that promotion into the command relocation process, it would lead to regular improvement to the AI command structure, therefore activation chances, therefore manoeuver capabilities...

Say Grant wins a few battles as a 2* and becomes promotable to 3* stars. the AI should promote him automatically (the routine would be for the AI to promote whenever possible, and if several generals can be promoted, sort them by seniority order), and, as a 3* commanding a corps, tag him as "to be replaced as a corps commander asap", next generals checks, the best 2* general available is put in command of his corps releasing him as a free 3*.. Next generals check, it would replace the worse 3* in command...etc...

The main issue with this routine is how to make it sleak enough that it doesn't break the AI ? I think this would require some "cheating", ie for the AI relocating generals should be instantaneous. the way I see, every and any routine that forces the AI to execute a task spanning several turns is going to be very very hard to implement, with the possible exception of winter turns with very low activity.

So the AI should be able to do all this in one turn, with relocation and promotion being instantenaous, and possibly toggles to activate free division formation and disable seniority and political cost penalties as well. The typical routine would be :
Check promotable leaders--> promote by sorting according to seniority
Check leaders in command
Check leades without a command
Replace "worst" leader in command (a newly promoted leader would be at the top of that list) by best free leader.
Recheck leaders in command and free leader, if some free leaders are better than some in command, repeat the replace task, etc.. until the AI finds that all commanding army/corps/division generals in command are better than those not in command.

This would do away for the AI with seniority and political costs, or would be a lot more complicated to do, but it would at least help alleviate the AIs command problems.

Add that routine to an enhanced division forming routine (that still needs a lot of enhancing) and the AI would go a long way toward a better organisation.

2/ the use of army and corp structure by the AI is poor. Some basic principles of operationnal war need to be implemented : The AI needs to use the Army/Corps structure properly, ie on the field it needs to advance toward its objective with a "winged" structure depending on the number of corps it can field, the best number of corps being 3 :

- if 2 corps : the army advances in the middle, winged by the 2 corps :
corps--Army--corps

- if 3 corps : the Army advances in the middle of the 3 corps :
Corps-Army-Corps
------Corps-------

- if more corps, the army should keep the 3 corps formation and either use the other corps for rear cover or to extend the line...

Sorry for this long and incoherent post.. I wanted to post a few ideas.

[Zoetermeer]

I'm not the expert, but I would say at least in my experience the activation rule is by far the biggest hindrance to the AI. The difference between it being on and off is like night and day when the AI is playing as the Union.

[richfed]

I just completed a '61 campaign as the CSA using these settings [as recommended above]:

Low aggressiveness
Small FOW bonus
No activation check

I also allowed the AI more time.

It was a great game, but the results were the same: I shattered the Union morale in late March, 1864.

Without lower-Mississippi invasion capability, I fear it is practically impossible for the Union AI to defeat a competent human player. I even made fairly large garrisons in Mobile, New Orleans, and Vicksburg - knowing I didn't need them.

Still, it was a heckava good ride!!

I am going to try with normal aggressiveness next.

[SirMaru]

In the real Civil War General McClellan never attacked vigorously. Lincoln bided his time and then replaced him after months of BOREDOM.

It looks like the AI IS MCLELLAN.

[PBBoeye]

Obviously the AI needs work, but I'll tell you this just from this RR overhaul project - playing the CSA is going to be a lot tougher.

[Zoetermeer]

Without lower-Mississippi invasion capability, I fear it is practically impossible for the Union AI to defeat a competent human player.

I definitely agree with this, but I guess this is a larger issue. I don't think AI's are generally very good at planning long-term strategies like that.

[PBBoeye]

Not sure if there has been any commercial wargame AI that has ever done something like that.

[DirkX]

T2/ the use of army and corp structure by the AI is poor. Some basic principles of operationnal war need to be implemented : The AI needs to use the Army/Corps structure properly, ie on the field it needs to advance toward its objective with a "winged" structure depending on the number of corps it can field, the best number of corps being 3 :

- if 2 corps : the army advances in the middle, winged by the 2 corps :
corps--Army--corps

- if 3 corps : the Army advances in the middle of the 3 corps :
Corps-Army-Corps
------Corps-------

- if more corps, the army should keep the 3 corps formation and either use the other corps for rear cover or to extend the line...

Sorry for this long and incoherent post.. I wanted to post a few ideas.

absolutely agree on this, the weird "army setup" makes it easy to win ANY battle for the human opponent, and i see it as largest drawback of the AI, which often enough DOES the right decision (attack where and when) but performs it poorly and loses the battles because of its wrong army setup.
if that coul dbe overworked and armies would get a good OOB the AI would be a much tougher opponent.

[PBBoeye]

Hell of a point. I think that needs consideration.

[General Quarters]

The basic problem may be this: A complex game gives the human player many tools to work with in the selection of units, organizing his forces, and so forth; but it is very hard to design an AI that knows how to properly use these tools. The extra elements that make a game more historical and more interesting to the human player may by themselves make the AI relatively less challenging.

[wyrmm]

I would like to see (in the long run) a game where over the AI (in at least historical scenarios) is a script with the actual plans and TO&E of each side in it. If you gave the current AI a force structure to use and the Anaconda plan to execute, I think it would do a credible 'historic' job. (With the activation rule it may be too historical )

[richfed]

I was thinking along these lines, too ... Maybe scripted reinforcements in one scenario [a variant of the 1861 campaign] - with some variables from game to game.

[JB Hood]

i added (1 st turn)

the union 400.000 dollars + 300 Conscription
the CSA 200.000 dollars + 100 Conscription

no activation rule; playing csa

that was a good game

union invasion in north carolina & peninsula & from Washington in summer 62. target -> richmond

but still the problem that the ai doesn´t care about her communications.

the union force (much bigger then mine) took richmond, but then was defeated in detail (surrounded, cause johnston was in manassas, jackson in winchester and lee in richmond).