Initial Observation re AI: Patch 1.033

[Tac2i]

I stress this is an initial observation based on the first four turns verus the AI. I'm playing as WA and my observation is limited to North Africa. First turn the Axis AI occupied the central Med with its fleet and landed additional milita in North Africa. I advanced one infantry to meet the Axis milita. I reinforced Suez with an infantry unit, one heavy and one medium bomber and retreated my Med fleet into port (they were not attacked). Next turn the Axis milita disappear. I attack the vacated area and advance the medium bomber and bring in another infantry. No sign of the med Axis subs. Turn three the Axis fleet in the central Med returns to port. A fighter over the fleet did attack and destroy my medium bomber. I invade and capture Sardinia and another North African area. Turn four the Axis fleet remains in port. I notice Silicy is undefended and invade and capture it. I attack an undefended Liyba and now own all of North Africa, Sardinia, and Silicy. Unless the Axis AI defends Itay adequately, Italy is at grave risk of an early surrender. I also noted the Axis subs staging out of Germany are not being agressive. One moved into port in Western France but in four turns, not a single transport has been attacked.

Based on this very limited observation, I'd would expect the Axis AI to [at least] defend Sicily so as to forstall an invasion and not risk an early surrender of Italy. I'm also concerned about the Axis sub fleet. Is the strategy to build up before attacking? Anyone else have any observations on the North African area or the AI Axis sub play?

[aletoledo]

I'm not sure if the point of the patch was to enhance the AI?

I doubt the AI will take advantage of attacking east germany first to eliminate the west german fortress. AI programming can be difficult and time consuming. Gary would have to have no life outside the office to be able to completely overhaul the AI logic in the short time between the last patch and this patch.

I believe the changes were focused upon factors brought out by PBEM human-to-human games, that the AI doesn't necessarily exploit.

[von_Schmidt]

I guess that the AI is subject to the changes, but does not 'know' about them and will not actively pursue the new prerequisites for e.g. Italian surrender or West-German fortress annulment.
Unlike say the rule on Chinese inland provinces activating factories, which the AI does know about and avoids.

Fine by me, since it did not abuse the rules to invade Italy on turn one anyway.

Any observations on the AI evacuating Leningrad without a pressing reason? That always annoyed me.

von Schmidt

[Tac2i]

@aletoledo: Item 18 of the 1.033 Patch ReadMe file: AI Improvements - Many changes were made to improve AI play involving both movement/combat strategy and production/research strategy.

[aletoledo]

@aletoledo: Item 18 of the 1.033 Patch ReadMe file: AI Improvements - Many changes were made to improve AI play involving both movement/combat strategy and production/research strategy.

I get your point, but those changes in the AI could be items that were identified in other areas and not the new aspects of the patch. If I recall they mentioned they were improving the AI to attack west germany or honshu, despite poor odds of actually winning. so perhaps this is the 'improved AI" that is mentioned.

programming can be tedious and time consuming. add on top of that the time needed to test every variable and thats why games take years to produce. the AI is IMO the hardest thing to program in any game. I find it hard to believe that in a matter of two or three weeks they could have rewritten the AI to incorporate defenses against every possible strategy that players have developed since release.

[Joel Billings]

Yes, the AI did get many changes. Now, you'll see it go after Spain sometimes even at the Normal level. Sometimes it will go after the Med. Sometimes it will ignore the Med. I hope the game you are describing is a game where it has decided to ignore the Med. It also uses a different sub strategy. In testing it was better than the old strategy, but it has it's limitations. There were actually many AI improvements, but it is not a human player and don't expect it to play as well as a human player. It still has blind spots and if you are an experienced player you need to be at Challenging at a minimum, if not at Hard to give it a fair shot at you.

The AI does know about the new rules in that we adjusted the AI somewhat to account for them. However, it is still known to have a hard time juggling it's priorities (especially as Germany) and without some help you will find weakpoints. If you want to see it's different strategies in action you will need to play several games. But please do so at the higher play levels if you want to increase the challenge.

We expect to do some more AI work in the future as players and testers identify very weak areas and send us saves showing the poor play (pre and post saves are needed), but we think it will put up a very good fight now at the Hard level, and will give beginnners a run for there money at Normal.

[Joel Billings]

Aletoledo is right that Gary did not rewrite huge sections of the AI. This was more a matter of making minor adjustmetns some AI routines where he saw he could improve them, changing some basic AI production strategies, and making a few adjustments to certain pre-set strategies. In some ways the AI is much better, but in other ways, it hasn't changed at all or is just a marginal improvement. There were also enough changes to the way the combat analyzer was working that we felt the AI should improve from that alone. However, we won't really know how much of a change all of these together will have until we start hearing from those of you on the forum as to how different the game plays. We do want to caution against making broad generalizations after only a few games however as depending on the strategies the AI picks and the play level you are playing at, your experiences from game to game could be quite different.

[Tac2i]

@Joel: Good point about the difficulty level. It was set to normal. Being a life long wargamer I should perhaps increase that level. The only big thing in my early observation, to me, was the AI leaving Silicy completely undefended.

[Joel Billings]

It can't be everywhere at once. I don't think it gives any priority to Sicily. Even though this means it can still lose Italy easily, it can afford to lose Sicily but it can't afford to lose an Italian area and have all the infrastructure damaged even if Italy does not surrender. It is trying to defend the mainland, but it sometimes fails at this (it should be better with the patch, but I can see at Normal it will still have a lot of problems doing this all the time).

[Tac2i]

@Joel: Your points are well taken. Let me just say that Gary Grigsby's World at War is one of the funnest computer games I've played in a long time and I'm certain the patch has made it an even better game, especially for the PBEMers. I'm already looking forward to my first PBEM experience post patch 1.033.

[aletoledo]

The only big thing in my early observation, to me, was the AI leaving Silicy completely undefended.

I always leave sicily undefended. Since without placing an AA or artillery there, it would be pointless to leave a sole miltia the island. I feel an extra AA and artillery would be better served on the mainland.

on the other hand, since sicily is now a proper island I will probably start using it a lot more.

[Tac2i]

Update: I've continued this initial game and the Axis did defend Italy well. After some bombing I was able to capture Southern Italy in Spring 42 forcing a surrender. The Axis reacted in a measured way by reinforcing Northern Italy. Two to three turns later the Axis counter-attacked and severly mauled my forces recapturing Southern Italy. At least it gave Russia a little breather. In the Pacific the Japanese are giving the Americans rough time. All in all I'd say the AI is doing much better and I'll have to play more alertly and I'm still at the normal level. Good job on the patch!

P.S. The Axis subs are playing more intelligently also: not losing them piecemeal as before.

[Drax Kramer]

Allied AI is more aggressive as far as invasion of Western France is concerned. Yesterday I witnessed the very first Summer 1944 invasion against German AI. Up for now, AI mostly concentrated on bombing.

What puzzled me was the behaviour of German AI (1942 campaign) which abandoned North Africa in the very first turn, only to send lone airborne unit to reoccupy Cyrenaica few turns later. I also noticed that in 1942 scenario German AI does not attempt to win the war against Soviet AI (I am playing Japanese), but rather tries to maintain the status quo and waits for time to run out.

I'll see how does Allied AI behaves in Pacific. So far, it spectacularly failed to defeat Japanese AI in any of the games I played so far.


Drax

[PJL1973]

Heh - I actually ran an AI game just now, and saw the Allies mount a summer 1943 D-Day, instead of invading Italy.

[a511]

AI problems i observed last nite while playing the allies for the Sp'40 scenario.
in Wn'42, russia did a great counter attack and was going to trap a huge axis army defending Rumania (incl. fresh troops from W.europe) but when i checked the status of the axis army, i found 0 supply around them!! but i also found piles of supply 30+ supply units in West and East Germany[&:] why didnt the AI moved a couple of supply units (together with the fresh troops from W.europe) to the starving axis army in Rumania??

i also noticed that the whole axis fleet in the port of W. Germany had made it a bad habit to waste supply by taking a "boat ride" to Finland and come back to the port of W. Germany w/o firing a shot. with 2 hvy and 2 lgt fleets, the "boat ride" is really expensive in terms of supply.

[Joel Billings]

If you saw a fleet move and then return in the same turn, it is quite likely that the fleet actually undid it's move and didn't spend supply. Did you see supply symbols on the fleet that made you think it had actually done the move as opposed to just undoing a potential move? You will see the AI's undone moves when it changes it's mind. It's programmed to "test" moves and then undo them if it decides against it.

[a511]

Am I right to say that if u undo all moves and end up in the original region, the fleets should not use up any supplies and should show no supply icon?

The reason why I thought the fleets did make the "boat ride" was that I did see the supply icon w/ red arrows, which IMO, the fleets did take an extend movement during the last turn and did spend supplies. I could be wrong, and I will double check.

AN

[Joel Billings]

If you undo you should not spend supplies and not be tagged as being in supply or fueled. If you see that behavior and can confirm moves that are costing supplies, please send us a save game.

[a511]

i will try to send u the save game file after work today.
thanks for ur prompt response. [:)]

AN