Soviet GC vs. AI - WiTE Version 1.02 beta

[Redmarkus5]

The Eastern Front, Spring 1943, before the launch of Operation Tunguska.

[Redmarkus5]

The Soviet and enemy OOBs on the eve of Tunguska.

[Redmarkus5]

Units destroyed between 22 June '41 and the launch of Operation Tunguska in Spring 1943

[Redmarkus5]

Ground losses 22 June '41 to Spring '43...

[Redmarkus5]

Air losses for the same period

[Redmarkus5]

Air recon confirms an interesting Axis AI deployment plan, with a large concentration of troops are air units in the south, between Kishinev and Odessa, and very little by way of reserves south of the Pripyat.

This suggests that a successful drive on Lvov really does offer the chance to split AGS in two, driving a substantial part into Romania, where it will cease to have much strategic significance.

[Redmarkus5]

Turn 102 - OPERATION TUNGUSKA

Hopefully, the months of planning and preparation will now pay off.

Phase 1 of the Operation involves the creation of several large gaps in the enemy's fortified defensive zone through which armour can exploit.

This appears successful, as shown below.

[Redmarkus5]

However, the AI reacts with alacrity, pulling back troops in the south to avoid encirclement and rushing reserves up to the front at all points.

[Redmarkus5]

Turn 103 - OPERATION TUNGUSKA

The AI does not defend Odessa which falls on turn 2 of the operation.

12 Axis divisions (including 1 Pz and 1 SS) are pocketed this turn, taking a good chunk out of AGS defensive capabilities.

[Redmarkus5]

Turn 104 - OPERATION TUNGUSKA

This is a mud turn during which I eliminate the pocketed Axis units from turn 103 and build additional Soviet support units at Front level. No advances are attempted along the line.

[Redmarkus5]

I think I just lost the desire to continue this AAR. I read another post where the player took Berlin in '43 vs the AI (tm.asp?m=2674668) and it basically answered the questions I have been testing here.

My conclusions:

1. The game is fun and mentally challenging, but as a historically reliable representation of the war in Russia, it still needs a considerable amount of work before it is close to being complete.

2. Key flaws in the game at present are the AI's failure to attack or defend key cities and the ability of the Soviet side to easily evade being pocketed in 1941, leading to a situation where the balance of forces is quite a-historical by Spring 1942.

3. The AI is programmed for the offensive in '42, regardless of the VP or casualty situation/balance of forces on the OOB. This leads to disaster.

4. The Air side of the game needs a major rework.

5. The Admin screens are difficult to navigate and make the player's task much harder than it should be.

6. After summer '42 it seems unlikely that the AI is capable of mounting any kind of limited counter-offensive along the lines of Kursk, etc. It just sits there and awaits its fate.

I will give PBEM a try once the basic issues are fixed, and might try an Axis game vs. the AI to see how things play from that side under release 1.02.

This AAR ends here.

[MrLongleg]

Good Job - thank you for sharing your experience.

I think it is basically impossible to create a really good AI for a game of that scale. The complexity is too high and nobody wants t wait hours until a turn is completed. But I still think that Gary and friends did a decent job. The real fun part here is to play against a human opponent, knowing well that the engine is not perfect, but still very good.

[jjdenver]

Interesting AAR, and I think my reading and experience w/ the game leads me to agree w/ your conclusions. I think playing vs German AI would require a tougher difficulty level than you used. Playing vs Soviet AI however should offer some more challenge - although another AAR describes Leningrad 42, Moscow 43, Stalingrad 44 (potentially) as a winner for German vs Soviet (normal?) AI. Still - vs AI w/ some bumped difficulty (100 for Germans, 110 for Sovs maybe) would probably offer a challenge.

The air and swamp issues seem to be priority for the next patch to me, along w/ an AI improvement in handling cities.

As for Soviet pockets, I've found in PBEM that after about turn 3 it's probably tough for the Germans to get any big pockets because the Sovs can delay pretty well using spaced out units. And because the Sov commander can see the big picture and react freely - something that Stavka/Stalin didn't always allow (pullbacks).

This is not to say that "no" pockets can be achieved by the Wermacht after the first 3 turns. You can see from some of the AAR threads that pockets of 5 or fewer divisions are being achieved in some cases in PBEM in August/September.

Of course the lack of historical sized pockets in July/August/September 41 for Germans might be reversed by the lack of similar pockets for Soviets in 43/44. There's no Hitler in place to prevent withdrawals from untenable positions so the Germans can feel free to conduct a mobile defense.

One way to handle this might be the way that the French C&C problems were handled in "Decisive Campaigns: The Blitzkrieg from Warsaw to Paris". Basically French units would randomly lose part of their movement allowance each turn (by command/corp) until Gamelin was replaced. Perhaps something like this that gradually lessened from summer to fall to winter and finally going away by winter would cause some irregular and unpredictable Soviet unit movement abilities and represent Stavka/Stalin interference to let the Germans get more pocketed troops. Just an idea.

(FYI something like this variable movement is built into the game already but the variance would have to be above and beyond this to have the effect desired - an extra movement loss during the summer/fall of 41)

[Redmarkus5]

Thanks for the comments.

While I have always agreed that a perfect AI is an impossible dream, many of the issues reported can be fixed and (as jjdenver points out) have been effectively addressed before in other games.

Some key questions are:

1. Why can Soviet units in WiTE retreat as fast as the Axis can advance in June/July '41? (Solution - more effects from 'shock' please).

2. Why does the AI ignore city hexes for much of the game when VPs are calculated at the rate of 3 for a minor urban and 5 for a major urban? (Solution - make VP locations central to AI planning).

3. How come the AI can be programmed to set up a proper defence line in 1943 as the Axis but not in 1941 as the Soviets when it creates a 'carpet' defence? (Solution - whatever Gary did to code the Axis in '43 needs to be examined for lessons with the Soviet '41 coding).

4. Is it really acceptable that a game that represents WW2 in Russia can't model a Kiev pocket or a battle of Stalingrad within the scope of its 'grand campaign'? (Solution - when holding an area around a major urban centre like Moscow, Kiev or Stalingrad, holding ground should outweigh the avoidance of pocketing a given % of the time. This can be a random factor, hidden from human players).

The main issue here is actually with some vocal individuals failing to accept that improvements are needed AND are possible - there is a problem of denial. There are many things that CAN be done to get this AI to where it needs to be and, given the price tag, purchasers are entitled to demand that they ARE done.

Playing a human opponent is a different challenge and one I will come to - I only play WiTP PBEM now, for example. The purpose of an AI opponent should be to give you a challenge in learning the game as well as providing an educational approximation of the 'historical' type of responses (levels of aggression, operational priorities, commitment of reserves, timing of/triggers for major actions, etc.) that occurred during the actual campaign. House rules or other controls can be employed to keep players honest.

Against a human who knows why his side lost or won historically, a game like this typically deviates fundamentally from history from turn 1, making it an entirely different proposition.

[GBS]

Redmarkus, my skill level and knowledge of the subject matter is about a 1 on a 10 scale compared to yours. I don't always agree with the strident nature of your objections but I certainly can't doubt your passion for the game and wargaming in general. In this case I like a lot of what you are suggesting and I for one hope the devs give this due consideration. Thanks for the AAR.

[Redmarkus5]

Thanks GBS. The objections only became strident after the original observations about the game were stridently rebuffed

Sorry if it's been tough to put up with...

[MrLongleg]

Just to clarify - my post was not meant as a rebuff. Certainly the developers should and hopefully will implement improvements. Your points are all valid. But for me playing the AI is only a way to learn the game because it is not too difficult to outsmart it, even if all your requests would be implemented. Computers are still having a hard time to read a map. But of course everybody dreams of a decent strategic and tactical AI, that plays like a human player. Until then I will stick with PBEM (although I probably will get beaten badly in my first game)

[Grotius]

Ah, too bad to see this enjoyable AAR end.

I haven't played nearly as far along as you, redmarkus, but I see the glass as half-full rather than half-empty. The AI isn't perfect, but I've never seen a better AI in an operational wargame, and I've played many of them. This AI is certainly far better than WITP's. Do you know of a better AI for a game of this scale? I'm really impressed that it actually reacts to my moves, that it knows how to encircle, that it generally avoids stupid head-on assaults, that it forms a coherent line, etc. In fact, some of Hitler's decisions were more stupid than anything I've seen the AI do. You mention that you want a historical battle of Stalingrad, but wouldn't that require Gary to code an irrational German AI that insists always on attacking, that refuses the let the 6th army break out, etc.?

I generally expect attacking AIs to be weaker than defensive AIs, but I've been pleasantly surprised with the attacking AI in WITE. I'm playing as the Sovs, at Week 12, and the Germans are within 50 miles of Leningrad and 90 miles of Moscow. They've lagged some in the south, but only because they took the time to liquidate an enormous pocket -- one I was over-confident about escaping. I'm pretty sure I'll hold out, but it sure "feels" like the war to me -- I'm retreating constantly, playing for time, throwing ill-equipped, ill-trained units at the front, battling for my two most important cities, looking forward to winter, the whole shebang.

As for cities, I haven't played Germany yet, and my Russians are still backpedaling, so I haven't seen the behavior you describe. But I can say that I abandon cities all the time, because the alternative -- being encircled -- is worse. I've fought a pitched battle for only one city, Smolensk, and I wish I'd retreated instead.

Obviously I can't order you to enjoy or play the game, but if you do play more, perhaps you should try Germany on a slightly higher difficulty setting? Getting those 290 VPs looks tougher to me than playing as the USSR, which of course won a "Major" victory historically (in WITE terms).

Anyway, I really was enjoying your AAR, but we play games to have fun, and if it's not enjoyable for you, obviously you should move on (or take a break).

[randallw]

We tend to complain about an AI not being as smart as a human, but the limits may also prevent it from making historical mistakes that humans have actually made! [:D]

[MrLongleg]

ORIGINAL: randallw

We tend to complain about an AI not being as smart as a human, but the limits may also prevent it from making historical mistakes that humans have actually made! [:D]

Couldn't agree more ....