Is the AI functioning properly?

[rocketman71]

A while ago I posted about what I felt was weird behaviour by the AI in the Bulge scenario, in which both the Axis and Allies AI withdrew from valuable hexes for no apparent reason. Like the Allied AI vacated Aachen on turn one despite being under no serious threat.
I have now started the Westwall scenario and have had problems breaking through. One thing I noticed on AI (Axis) turn one that was odd was that the Axis withdrew from all hexes needed to block Antwerp from being a valid port. Ok, I could see some reason in reacting to Market Garden. But now it just left Metz which is a victory objective for both sides, it wasn't under a serious threat and basically hadn't been attacked. That makes no sense unless it is scripted. See screenshot. It would have been ok with any old hex, but an objective? Wasn't Metz even considered a "fortress" town? It isn't in the screenshot, but it was fortified to level 3.
I have only played the shorter scenarios so far to learn the game before embarking on a full campaign, but if the AI will keep on doing stuff like this I'm having second thoughts for sure (sadly). I have no time to commit to playing a human opponent at this time. Otherwise the AI seems good, but sometimes makes really weird choices. Something I wouldn't expect from a high quality game like this.

[loki100]

I think it has particular problems in some of the scenarios, but often setting it to 110 morale avoids the worst of it.

Across a 1943 campaign, the Axis AI (again 110 on morale) will give you a good game, where you have to work hard for a draw and an outright win is unlikely. Its an AI, it will be sub-optimal in some respects but its one of the best vs AI gaming experiences I know of.

Less sure about the Allied AI but I've never played against it.

In part, no great interest in the axis side but also from the allied perspective vs AI is a series of interesting puzzles to solve as you are the side with agency.

I know in developing WiTE2 that Gary worked hard on the AI but its not especially terrain aware and that is apparently very hard to code in. WiTE2 tends to make it garrison certain locations and has lots of regional aspects to its behaviour. So I'd not be surprised if it just hands over something like Metz, that sort of 'hold because on the margins I think this will work out' decision making is too subtle and too circumstantial?

[rocketman71]

Thank you for the detailed answer. The game is played under "challenging" setting so morale is definitely on the AI's side. I understand that AI consideration to terrain can be hard to program as complex patterns can arise, but a single objective hex surely must be easier to code for? It is good to hear that more work was done on the AI for WITE2 as I intend to buy it shortly. From what I have seen on Youtube the AI seems very good and that's perhaps why I thought it would be roughly the same in WITW. So maybe in WITW2 some day

[MarkShot]

At a certain morale level, the AI has no movement limitations such that it can always form a front line; so I hear.

I bought both too, but I am enjoying this. WITE2 is still being balanced as they (community) argue about issues of command and control, air losses, and now tank losses.

It is very hard to balance such complex games; especially that allow to play from either side.

---

The funny thing is you think of the East as huge, but in terms of ground counters. WITW is as big or bigger. It's no slouch.

---

I tried HOI4 earlier. What a disappointment. These two games and associated documentation are master pieces!

[loki100]

ORIGINAL: rocketman71

Thank you for the detailed answer. The game is played under "challenging" setting so morale is definitely on the AI's side. I understand that AI consideration to terrain can be hard to program as complex patterns can arise, but a single objective hex surely must be easier to code for? It is good to hear that more work was done on the AI for WITE2 as I intend to buy it shortly. From what I have seen on Youtube the AI seems very good and that's perhaps why I thought it would be roughly the same in WITW. So maybe in WITW2 some day

one significant change is I think its less panicky if in a salient than the WiTW AI. Its still a problem for it to evaluate but it is told to evaluate the value of the held hex more highly.

This came out of trying to teach the German AI to cope with the Red God of War set up, it did the sensible thing and got out of Rzhev etc on T1 - sensible but scenario breaking.

I think Gary gave it some rules to make holding a logical choice and those rules were then added to the generic AI. He's doing similar tweaking with some of the scenarios in testing as they have lower counter density and/or discontinuous front lines (and both these its struggled to evaluate in the past).

[MarkShot]

Loki,

Both players and betas (used to beta) like small scenarios. In huge scenarios rarely can a single unit and decision really change everything, but in smaller scenarios just a company in dense woods set up for an ambush can change everything. They may not be able to hold, but first they cause a halt, then a formation redeployment, then an attack task force to dislodge them, then redeployment into march. That could be six hours already which is enough time to set up a defense further down the line or to make sure the sun sets and their tank main guns are useless in the ensuing fight.

So, small scenarios have a certain charm, but I agree they push an AI to the limits of analysis and action.

[loki100]

oh I agree, I love small scenarios where precision really matters. In WiTW the various N African scenarios are gems in this regard.

With a regular MP partner, we've gone back to Flashpoint: Red Storm, there I think the best scenarios are those that pit a recon company against a small advance guard.

Pavel, one of the main designers, has put forward a view that the 1943 GC should have been locked till a player had completed a number of the smaller scenarios. I certainly think you need to play the breakout scenario (learn about how to apply tactical airpower) and one of west wall or Arnhem (learn to manage a stressed supply system) plus one of the strategic air war scenarios (good place to experiment and test out ideas) before starting a campaign.

[Seminole]

I know in developing WiTE2 that Gary worked hard on the AI but its not especially terrain aware and that is apparently very hard to code in.

I spend 99% of my time playing people for a host of reasons, but that is a remarkable insight.

I consider terrain to be the most important feature in the game.

If your planned axis of advance isn't taking it into account...

I would love to see how a 'learning AI' like OpenAI Five would eventually approach playing the campaigns.

To me most of the fun is wrapped up in planning and foresight, and trying to make those plans a reality in the teeth of your opponent's plans. I realize the AI, as it stands, can only ever really react.

[rocketman71]

I usually view playing against the AI a means to learn the game fully before taking on a human opponent. Sadly, real life constraints make it near impossible to take on large scale complex games long term with a human player. I just don't want to be that guy who bails on a game. I hate when that happens to me when you have so much invested in the battle. I really hope that WITE2 some day down the road will incorporate co-operative play as was hinted to when the game was released. Then larger campaign games would be much more feasible.