AI US Civil War 4.05

[Alan Sharif]

I am seeing a BIG improvement in AI. Much better than what I am used to. Thanks for the continuing development, and an extra special thank you to those who worked on the AI. It shows a lot of work was done on this.

[rahamy]

Thanks for the comments Alan, glad you are liking what you are seeing!

[Alan Sharif]

Simply giving credit where credit is due. It's well deserved.

[Quaama]

I am seeing a BIG improvement in AI. Much better than what I am used to. Thanks for the continuing development, and an extra special thank you to those who worked on the AI. It shows a lot of work was done on this.

You got my hopes up there. Unfortunately, I did not notice much improvement.
I played the historical AI Challenge Bull Run (Antietam #144) and the AI received the usual severe walloping. By Turn 24 I had CSA minor Victory; by Turn 26 I had Major Victory; by Turn 32 I had three times Major Victory so did not proceed any further. It was going to get a lot worse for the Union and there was a good chance their entire army would have been isolated and eliminated before the time limit.
One odd thing I noticed was that when I went to capture Centreville, Miles Division was not there. No-one was there. In the normal Historical Bull Run, Miles Division is fixed there for the whole battle (unless approached by the CSA). Perhaps in the AI Challenge they were there but free to move as I noticed Col. Blenker was among the Union Leader Casualties.

The only main difference I noticed was that the Opportunity Fire seemed above the norm. Many defending units fired multiple times which had previously been a rare event. It may have been triggered by the larger than normal units on the CSA side but it did seem to occur more often. As this operated for both sides there was no advantage there although it probably did increase the toll on the Union as they were generally moved forward in large stacks and launched one frontal attack after another. Consequently, Union casualties were heavy.
The only other difference I noticed was that the AI now seems more careful with its leaders although I suspect this is more to do with the recent change to the games regarding lone leaders and the proximity of enemy units.

The main problems with the AI seem to be that it seems too focused upon various locations on the battlefield and fails to react to large bodies of men flanking it. An example of a game where the AI does this better is Gary Grigsby's War in the East (WitE). The AI in WitE does consider locations but also reacts to where your units are and what they are doing. WitE will also try to isolate your units every chance it gets which is something that the WDS AI never seems to do despite the obvious advantages in isolating the enemy.
The AI in the games also seems to have problems in other major areas:
it seems incapable of coordinating an attack and just comes straight at you from front-on;
crossing rivers is a big challenge for it; and
the AI remains extremely careless with artillery.

[Alan Sharif]

Oh drat. I must just be a worse player than I imagined. I had a hard time playing the Union in the Red River Campaign. Whilst the AI is not perfect, and ultimately I still won, I found it a much tougher game than previously, having played several times before. A step in the right direction, for me at least.

[Alan Sharif]

On the other hand perhaps it is performing better at some scenarios than others. Without knowing what has changed it's difficult to know what effect it will have had. That said, an actual 'AI Challenge' should show it at it's best.

[rahamy]

I would not expect "positive" comments from anyone who has been doing PBEM for a long time - as stated in our blog post, it's not a human replacement and never will be. Placing expectations on it as such is unrealistic.

[LarkinVB]

I did not expect positive comments at all. So thanks to Mr. Sharif for his kind words.

[Quaama]

Oh drat. I must just be a worse player than I imagined. I had a hard time playing the Union in the Red River Campaign. Whilst the AI is not perfect, and ultimately I still won, I found it a much tougher game than previously, having played several times before. A step in the right direction, for me at least.

It's more a case of many people doing things they shouldn't do and failing to do other things. The AI may have had some small adjustments made to take advantage of those things. Here's a few examples.

1. I said before that the AI "seems too focused upon various locations on the battlefield and fails to react to large bodies of men flanking it". Why defend a location just because there is a flag plonked upon it?
In that scenario the AI seemed determined to get the three Objectives closest to Blackburn's Ford. After a brief defence I let them have all three because they were not essential to my plan. I defended from the higher ground several hexes to the south.
Bonham's Brigade crossed to north of Mitchell's Ford and dug in. Early's Brigade assisted Longstreet defending south of Blackburn's Ford. All three were within range of Beauregard. Most of Bonham's Brigade eventually routed (I moved Beauregard away to help rally some other units) but that didn't matter as that area was soon supported by others (see 3. below).
2. Watch Command and Control. If your command and control is poor your boys will run and you'll have a hard time getting them back into the fight (i.e. they will fail their morale tests).
Also, did you notice that when you click on either of the two Army Commanders to show 'Command Range' the game shows the expected Army Range (28 hexes). However, from the first CSA move the three brigades within Beauregard's range all became 'detached'. Therefore, as soon as I could I moved Beauregard to a position where those brigades were again within his range. The range for the Army Commanders was that of a Division Commander (6 hexes). This made historical sense to me because it adequately reflected the confusion within the battle by reducing the ability of the Army Commanders to exert control over a large area. [Note: The CSA did not have divisions at that time, just Brigades and other formations reporting direct to the Army Commanders.]
3. Always look for an opportunity to attack. It's wasteful to defend something you don't need to defend. I could never hope to defend such a big area so the Objective near Stone Bridge was soon vacated. The two Brigades in the north-west were soon formed into an attack led by most of the Army of the Shenandoah who marched to Island Ford then headed east to flank the Union. The cavalry (all of them) were sent north to then turn east in search of the enemy and eventually walk into the undefended Centreville. Evan's Brigade wasn't too far away by that stage so could have helped if needed.
Once that attack approached the enemy it was 'game over'. The AI never stood a chance from the moment that force joined the fray.

[Marc Brillon]

To "help" the AI, I'm using a rather straightforward method to exploit its strategic weaknesses. When I decide to adjust my tactics by repositioning multiple brigades to different areas for either attacking or defending, I utilize a standard six-sided die. If the result is 3 or less, the unit remains stationary. Additionally, I introduce bonuses and penalties based on several criteria: whether the brigade is within the chief general's command radius (imposing a -1 penalty if it is not and a +1 bonus if it is), as well as the quality of the chief generals and brigades (+1 for exceptional, -1 for mediocre). This approach significantly complicates the modification of ongoing plans, as well as the coordination of offensive movements and even the complete restructuring of defense.