The AI poll thread patch 0.85+ (NEW)

[subunit]

I think I may have been on the old version for at a couple of the battles, then. I just tried doing my Confederate steamroller (basically just beeline for VP locations ASAP with as many units as possible) against a determined Union AI. The first phase was still a little silly, they sat around plinking at Archer on Oak Ridge with 3:1 numerical superiority without taking it and ignored the fact that I capped Oak Hill with Heth's skirmishers until the last minute, so I got an Epic victory again without much thought, but in the second phase they held all their objectives- last time they were milling around getting slaughtered at the bottom of the screen. That's an improvement, anyway! Next time I'll record my game in case any weird AI behaviour crops up.

 

e: Trying the first phase of the Confed attack again with a Dynamic AI, it's definitely playing much more sensibly than before. Thumbs up, guys, I'm definitely having way more fun with this now.

[subunit]

OK, there are definitely still problems with the AI:

 

This is a Confederate dynamic AI. First phase they played sort of poorly (I got an Epic victory), second phase they played quite well, collapsing my lines at Seminary ridge and in Gettysburg (only a Minor victory, although the drubbing they gave me should have been a loss for me), this was the third phase at Peach Orchard/Round Top. I immediately abandoned Peach Orchard and pulled back to the Round Top ridge line, the AI approached with a few thousand troops, capped Peach Orchard, and then did nothing for the rest of the battle. They kept about 10k troops at the Western edge and made no attempt to attack, giving up a Major victory. The vid shows the state of play well into the "Battle Delayed" phase (no idea why it was delayed, no one was charging and nothing was contested- this is quite boring when the AI decides not to do anything). Subsequent phases the AI was a little more active but not by much. I ended up with a Decisive victory for the whole battle although if the AI had continued to be as aggressive as it was in the second phase I think I probably would not have won.

[Nick Thomadis]

After patch 0.85 the AI got a serious improvement. 

We need now your new opinions in this new thread.

 

Please vote for the best AI settings you like to play against and also for the least liked.

If possible write why you prefer or not some AI opponents. Any issue you notice in AI please write it here so we prioritise to fix.

PS 1.

Obviously the Random AI is not mentioned in the list because it is one of the 9 personalities picked randomly and you do not know which one

 

PS 2.

You can find the old poll here: http://forum.game-labs.net/index.php?/topic/1216-the-ai-poll-patch-083/

[subunit]

I've been playing some more with the Dynamic and Determined AIs. Does the AI reference the balance state when deciding whether to attack? I think the #1 problem I've observed consistently with the AI is that it does not press when it has an advantage. I think if it has a substantial advantage in terms of the numbers/morale/condition bar, the Determined AI, in particular when attacking, should be actively firing or moving into contact with most of its units. I have seen several instances now where the AI will settle into its lines after capping a minor objective despite having massive superiority and clearly being able to take other lightly-defended objectives. The AI also seems not to support attacks well with artillery. I think if the Determined AI could be made to be more aggressive in pressing the advantage when it's got more/better units on the field, and to use artillery a bit more aggressively, it could be pretty challenging.

In any case, I've been having a lot of fun with the game and I really appreciate that you're trying to make a non-cheating AI that has to fight on its own merits- my criticisms are made in the sincere hope that the end product will be a ruthless, brutal opponent!!

[greyheli]

I hope the AI remains the #1 priority at this point.

The gameplay vs. realism tradeoff choices that you've made thus far are brilliant,

and the gameplay is just fun, simply put. 

For example, I love the choice that units are rarely annihilated!

It is a brilliant choice that allows you to stick to slightly historical progression over a 3 day battle!

 

Now its important to focus on the AI, to take advantage of the game

rules that you've chosen.

The world is dying for a game like this with a competent AI. 

The AI is slightly better now in 0.85... however I can still beat the AI every time on

Cunning/Dynamic/Determined settings.

[R.E.B.Blunt]

I have changed my choices from last poll slightly.
 

I went with Dynamic this time because I enjoyed the way the AI both plays defense and offense on a more balanced level. Although the AI is still not that hard and Determined AI does attack better. Overall Dynamic gives the best mix of what you want in an campaign battle.

 

Still though Cautious AI is the least favorite of mine to play against. Its just so boring and they do not move.

[subunit]

Just lost my first campaign to the AI, a Cunning Union. The Cunning AI is much more difficult to drag out of position than Dynamic or Determined AIs, it's quite good on defense. Really happy to see that the AI can win once in a while, even at this stage!

[FGarage]

Even though I voted Determined, I find it squander's the morale of it's troops too much when it's handling the Union. If the Confederates aggressively defend against them for long enough, they simply cannot hold against a mass assault because their brigades are routing or getting ready to rout.

[David Fair]

Determined is fine for a single Phase - but it continues to attack very aggressively long after it takes heavy casualties.

Over the course of the battle Determined makes the balance of the increasingly easy.

 

It is almost like the AI needs to be able to "switch" between AI characteristics (e.g., from Determined to Cunning) after X casualties or Y VPs have been attained.

 

I don't play on Determined any longer because it resulted in shorter games with very heavy casualties for the AI.

The last time I ran a test with the CSA AI on Determined the AI suffered 40,000 casualties.

 

My vote is for Cunning.

[tornsage]

I win every time as confeds - I move armies up and then spend time to hold and defend before moving on to attack positions or take ground. Within 3-4 iterations of the game I beat the union every time! It's just too easy to win where the cons should have a monumental challenge. It also does not matter which AI I play.  The AI needs to be a lot tougher!

 

 

[David Fair]

Yep, the Triumphant Victory thing gets a bid stale.

 

IMO the victory metrics at the end of the game favor the contribution of the VPs way too much.

 

Gettysburg was about critically injuring/destroying one of the armies and ending the war.

VPs weighted so heavily disrupts the historical reality and the motivation and perspectives in play at the time of the battle in 1863.

 

If the CSA suffers more casualties the best they should be able to do is a "marginal victory".

[Husserl]

The AI needs to be a lot tougher!

 

It will be!

 

IMO the victory metrics at the end of the game favor the contribution of the VPs way too much.

 

After the OoB and the statistics (probably with the next patch) and some other additions related with the battle results (that is a surprise that Nick is working on), the outcome calculation will be reworked.  

[long5hot]

Playing as Union versus any of the most difficult AI settings - the first battle stage their 2 initial brigades can be distracted and forced to rout by dancing around them with the cavalry and skirmishers attacking from multiple sides.  That not only messes up the morale of those brigades, it buys you considerable time, more than enough to reinforce McPherson's ridge, secure Oak Hill and get artillery in place on Oak Ridge and Seminary Ridge for defending later on.

 

As Confeds in the same battle stage, its again pretty easy to force charge the opposing infantry back (the AI does _not_ dance them around but defends between you and the objectives) and use Heth's skirmishers to take and hold Oak Hill.  The Union rarely places artillery where it can really inflict damage when assaulting Oak Ridge and Seminary, so you can then patiently build up your forces, repelling any attacks, then launch charges at those objectives late in the battle, marching in with some fresh units to hold them and take the win.

 

Those are just two examples.  In general I find that the AI does not place artillery well in anticipation of attacks - where there should be a row of batteries ready to smash you with shell and canister if you try to attack the key points, with their status set to Hold, the artillery are usually evenly spaced (not concentrated) and set back from the line where they'll only hit from long range.  If a battery is being annoying, attacking with infantry forces it to retreat immediately - indicating that hold is not being used. 

 

The AI rarely seems to use Hold for infantry either.  Its all too easy to get their units to disrupt and move, and this is key to winning those important spots in the battle.  Conversely if you're playing on defense you can set your best infantry to Hold on those wooded hillsides (with "held" artillery concentrated beside/in front) and rarely ever will they be routed - instead the enemy formations approach in turn each dissolving away soon after entering rifle range.

 

I also don't see the enemy flanking too often.  They'll occasionally try it with a few units, but usually its just a predictable pound up the middle driving for the objectives.  The AI will rarely try seriously to turn your flank.  Neither will it recognize when you've turned the approach to an objective into an absolute death trap and focus the attack elsewhere - instead it'll just keep marching brigade after brigade into the meat grinder.

 

What the AI does very well is make units fall back, reform, then attack again - its reluctant to let units get so badly routed that they flee the battlefield.

 

Edit: To illustrate the consequences of the current AI's shortcomings, below is the endgame from the most recent campaign I've played as Union against Determined - I'm on the verge of destroying their entire army.  Not only did I prevent the Confeds capturing any objectives in the leadup battles, in this last one I've chosen the counter-attack-from-the-northern-ridges option, turned the Confed's right flank with ease and swept to Herr's Ridge unopposed while the AI spent all its best infantry units in repeated assaults on Seminary & Oak where I'd created a murder zone.  I've only played 33 hours and feel like the game is lacking a challenge right now, but I know AI is very difficult to develop and I keenly await future versions...

 

[long5hot]

Next I'll look at the opposite high difficulty mode - Cunning. 

 

This is of course far less aggressive than Determined and as the name implies, it seems more intelligent.  However it too has its drawbacks:-

1. Its all too easy to distract one or more enemy units.  For instance in the 2nd phase playing as Union I left a single artillery on hold at Blocher's Knoll, and after the Confederate reinforcements arrived on that wing one of their strongest brigades spent most of the rest of the battle sitting under the hill, unable to fire at the artillery (or vice versa) but seemingly unwilling to charge.  Similarly, playing as Confederate I could often weaken Culp's Hill's defenses by flanking around the edge with Heth's Skirmishers - invariably one of the defending brigades would leave their post to try and track the skirmisher's down, getting themselves disorganized and low on morale, and eventually giving up the chase and never returning back to Culp's.
    
2. The use of Hold for artillery and infantry seems to be the same as with the other AI settings - it doesn't seem to do it.
    
3. This next one for me is very important.  With Determined, the green battle timer fill-bar scrolls all the way across, usually by which time the armies are beginning to seriously come to grips and major moves are underway.  Then it says "Battle Delayed" (whatever that means) and files the entire bar again with red, with the actual end of battle sometimes waiting quite a while until combat ceases, usually with a 5 minute warning and the "both armies exhausted" message.
   
   In the several battles I've fought against Cunning, the second red fillbar begins then abruptly the battle ends with no warning at all.  Often I have big moves in motion just beginning to pay dividends but far from completion, so I feel robbed - why did the battle end when its clearly just getting underway?  Of course this AI mode will win if its on the defensive and the battle duration is far shorter (for no apparent reason). This alone has deterred me playing further against Cunning which is a pity, and something I really hope does not carry over to 0.9.
   
   Edit: Just played a game attacking the Cemetary as Confeds versus Cunning, and it went the full time distance to the "both armies exhausted" point.  Captured the Cemetary and Cemetary Ridge in a flanking push that the AI did not recognize or respond to - instead I baited it into attacking a few brigades I left in the town and north of Culps, and then pursuing them further north and thus leaving the Cemetary exposed.  Would have gotten Culps too had Union reinforcements not arrived.  Speaking of those reinforcements, I'm pretty sure if they'd charged me they could have won back both of the objectives I'd secured as many of my brigades were behind the lines recovering morale - but the Union chose to stand off, exchange volleys and eat artillery shells until the battle ended.
   
   I've seen someone else suggest that the AI would be better served if it recognized when to switch between Cunning, Dynamic and Determined depending on the circumstances.  That could be a new mode perhaps rather than replacing the three, and potentially could be used in the lower difficulty levels as well.

What I do like with Cunning is the way it cleverly withdraws units in a delaying tactic while building up strong points, and especially its use of cavalry to charge disrupted infantry which strangely the other modes don't seem to do as much, probably because they have cavalry constantly attacking fresh infantry brigades.  I haven't seen any evidence though that Cunning tries to implement a master battle plan, just as the other modes don't, especially attempts to turn a flank.  Its good at individual unit management (ignoring the failure to use Hold) but does not seem to have a big picture view of the battle or an overall strategy it tries to implement.  Like the other AI modes it just goes directly for the objectives and that's it.

 

Don't get me wrong I'm very impressed by this AI, but its not there yet.  If UGG is going to be restricted to just Gettysburg, the game either needs a finer level of detail and control (artillery limbering, units entrenching, separating regiments, setting formations, ammo supply points etc) OR it needs an AI capable of pushing human players to the limit.  Preferably both.  Preferably both plus a full civil war campaign - for that I would pay a lot more than the current $10.

[Nick Thomadis]

New poll will be made afterwards. Thank you all for the participation on this one. It helped.