AI Suggestions Call - Axis Eastern Front

[Edwin P.]

What can the Axis AI do better on the Eastern Front? What improvements would you make?

In my opinion its attacks during the first half of the year are excellent, but it faulters when the Russian can finally build a line.

This post is an attempt to collect a list of suggested improvements to the Eastern front AI for use in a future Version (SC1.08 or SC2).

--------------------------------------------

1) First, the AI does not seem to concentrate Air Power like a human player. It usually keeps a few air units far from the main front and Air units close to the front are not concentrated. Perhaps this is a function of the AI not being able to concentrate its forces on a single major thrust.

2) Second, the Axis AI needs to do a better job of investing in Tech. I don't know how many tech chits the Axis buys or when but my guess is that it is nowhere near the number that the average human purchases.

3) Am I correct in saying that the AI does not use HQs like a human player? It appears to me that the AI keeps the HQ units too far from the front lines and tends to operate them out too fast.

4) Humans normally buildup a large army on the eastern front. The AI invasion force seems to be much smaller.

[ September 16, 2003, 01:23 AM: Message edited by: Edwin P. ]

[JerseyJohn]

Edwin Important topic, great idea, we'd all like to see the AI play a better game than it presently does.

I think, as a generalization, it can be said the AI plays a very conservative game, missing opportunities and failing to either create or exploit breakthrough opportunities.

[Edwin P.]

I agree, another reason why it fails against an experienced player in Russia is that it lacks the production base of a typical human Axis player.

Example:

1. It never conquers Vichy France, but it attacks it.

2. It never conquers the Nordic countries.

3. It always delays attacking Denmark. Most humans attack Denmark on Turn 1.

4. It never takes Greece in 1 turn.

An improved game where the AI has the capability to conquer the Nordic Norway & Sweden and conquer Vichy France would make the game more interesting for the Allied Player.

Of course, updating the AI is one of the hardest and most time consuming modds that one could ask for.

[ September 16, 2003, 06:44 PM: Message edited by: Edwin P. ]

[pzgndr]

Of course, updating the AI is one of the hardest and most time consuming modes that one could ask for.

Exactly. Considering that none of us except Hubert truly understands how the current AI with it's fuzzy logic and other variables is programmed, it is virtually impossible for any of us to clearly define the problem and propose workable solutions.

We all recognize the limitations of the AI and would all like to see it perform better, smarter, etc. The age-old issue boils down to how to make the computer "think" more like a person. And for a WWII European Theater game with its broad spectrum of strategic, operational, and tactical options, that's no small task. Hubert's first effort at programming an AI was pretty good actually. It needs improvement, and I for one am prepared to be impressed by Hubert's next effort with SC2.

A repeated theme here seems to be requests for more and more programming of predefined options. I'm not convinced how practical that might be, since there's always the risk that a selected option turns out to be a stupid choice given whatever situation actually exists, or that alternate options are selected and deselected frequently enough to render the AI braindead.

It may be best to let the AI "think" for itself. Maybe Hubert can weave some user-defined AI scripting into the fuzzy logic algorithms, and maybe not. I'm just as curious as everyone else to see what can be done.

[Edwin P.]

Hubert's first effort at programming an AI was pretty good actually.

I agree. When I first started playing the game's 1939 Scenario it was beating me soundly as Axis and Allies (it still beats me in the later scenarios!) It concentrated on the most important part of the war - the war in Continental Europe and it uses corps and armor quite well.

The problem with designing an AI is that it is a never ending process. If you do not set any time limits on it to avoid scope creep then the product be it a game or an advanced warehousing management ERP module will never get out the door. As a designer you have to decide what are the most important elements to the AI and focus on those elements. The other items are added to a long list of things to add/improve once the most important items are resolved.

In addition, you have to decide whether to release new features as a free upgrade to the current product, as part of a new product/optional module, or as part of a prepaid maintenance plan.

[ September 16, 2003, 06:48 PM: Message edited by: Edwin P. ]

[Konstantin V. Kotelnikov]

Another thing I'd like to see the AI quit doing is all these little piddly attacks.

It only serves to weaken the AI's units and then you can move up strong units, kill their weak unit and exploit.

It's all well and good that the AI has some agressiveness, but there is a time to park a unit and dig in waiting to be attacked.

Devise some way to have the AI define a point of attack and concentrate on that. Instead of attacking with every unit when it's strong enough to.

Of course there is a time for that to. We just need to teach the AI when to use each kind of attack.

[JerseyJohn]

What's the problem? It's only a matter of teaching the AI to exercise discression, have an intuitive sense and some imagination. The AI needs to see things the way a human player would and then select the proper profound strategy and finally make the best possible moves some human player, like Terif, might choose.

After that the AI should finish college and choose a more challenging project, writing something along the lines of Beethoven's Ninth Symphony, for a start!

Scene from a SciFi Golden Oldie; Colossus, The Forbin Project

[ September 16, 2003, 03:02 PM: Message edited by: JerseyJohn ]

[santabear]

Here are some relatively easy to adjust things I've noticed:

AI's propensity to attack when surrounded in a city. It will often try to bash one of the surrounding units, take damage and then be destroyed easily in the following turn. It should be possible for an entrenched unit in a city to delay an attacker for at least one and possibly several turns with conservative play. Hubert claims to have adjusted this, but I don't notice much difference.

AI too often seems to ignore terrain, especially river/stream hexes when planning its attacks--Leningrad and the southern Russian cities.

AI moves eastward too agressively and triggers the Siberian transfer sooner than most humans would.

Not so easy to fix:

Once east of the marshes, AI seems to lose his strategic focus. Most humans will adopt a "Moscow" or a "head south and drain MPP" strategy. AI tries to advance on too broad a front and becomes easy to stop. (Actually, this is probably a re-wording of Edwin's original observaton).

[Edwin P.]

Santabear - Once east of the marshes, AI seems to lose his strategic focus. Most humans will adopt a "Moscow" or a "head south and drain MPP" strategy. AI tries to advance on too broad a front and becomes easy to stop.

Excellent observation.

Perhaps the AI could evaluate its position and adopt a Moscow or Head South or Divide the Country strategy. It would then reinforce that front with air units, HQs, new units built on occupied cities and units operated in.

If it selected the divide Russia strategy and this was successful it could then decide on a Moscow or Southern strategy. With Russia divided Russia can't build more units in the south.

So:

Past Marshes

--- To Moscow

------- Take Moscow

------- Surround Moscow

--- Head South

------- Concentrate Air on Rostov

------- Concentrate Air the Sevastopol

--- To Voronezh & Divide Russia

------- Then To Moscow

------- Then Head South

--- Attack on a Broad Front (Current AI)

In fact, I would say that a Novice level AI should always attack Russia on a Broad Front. At higher intelligence levels give the AI a greater chance to make an intelligent choice.

[ September 16, 2003, 10:01 PM: Message edited by: Edwin P. ]

[Dan Fenton]

Writing a good AI is extremely hard to do. Although when compared to humans, the AI seems less than bright at times, Hubert did a wonderful job on the AI for programing his first game.

In a nutshell, on offensive, the AI is a little overly aggressive, but fairly good. I have also heard that offensive AI is the most difficult of AI's to program. The real problem is that the AI uses up its army on offense and when circumstances change, the AI really does not redeploy to create a good defense. The AI then often seems to be issuing stand and defend where you are orders. Those units can be encircled and destroyed. Vulnerable HQ's and air units get exposed and can be overrun.

Specific AI flaws in Russia that I have noticed are:

1. Failure to establish a continous defensive line. This is the biggest one in my book.

2. Seldom moving HQ's to the rear when the defensive line breaks and wasting MPP's operationally moving them around too much.

3. Seldom moving air units to the rear when the defensive line breaks and and wasting MPP's operationally moving them around too much.

4. Putting too many Italian units in Russia, reducing vital Italian defenses to below minimum.

5. Not returning Italian units to Italy when Bari or Rome are threatened.

6. Not using enough corps units to garrison against partisans.

7. Moving into places where most humans would know that an enemy unit should be, suffering unneeded surprise attacks.

8. When on defense, repairing air fleets with scarce MPP's when those MPP's would be better used on repairing ground units or building new defensive ground units, like corps.

[Edwin P.]

Just played a game at Intermediate with Romania neutral and in Russia I cut off five HQ units - 3 German and 2 Italian which were supporting 6 German and 3 Italian armies.

The AI needs to allocate HQ units better.

There was no need for 2 Italian HQs adjacent to each other on that Front. At least one of the HQs should have been operated west to oppose the Allied invasion.

For Italy, I believe the AI should allocate to a front (Western Front, Eastern Front, Italian Front) or sector (Norther Russia, Southern Russia) at most 1 HQ unit per 5 units on a front/sector with priority for fronts with the most enemy units.

Example: There are 5 Italian Units in the Russian Front (Russia, Finland, Romania), 5 Italian units in the West (France, Spain) and 5 Units in the Italian Front, and there are 2 Italian HQ units available. The AI sees 2 enemy land units in France, 20 units in Russia and 5 units in the Italian Front.

The AI should allocate 1 unit to the Eastern Front and 1 unit to the Italian Front.

Proposed Italian HQ AI Rules

Rule1: 1 HQ unit per five land/air units per front. If ratio > 0.20 then operate HQ units to new front/sector until ratio is reached.

Rule2: Priority to fronts with most visible enemy land/transport units and most Italian units.

Rule3: If ratio of HQ unit per Italian units >=0.5 then operate/move HQ to Italian Front.

(Why - not worth it risking 1 HQ unit to support 2 combat units.)

[ September 16, 2003, 10:46 PM: Message edited by: Edwin P. ]

[Edwin P.]

Dan Fenton

Excellent listing of the major areas for improvement.

[ September 16, 2003, 10:49 PM: Message edited by: Edwin P. ]

[Night]

The AI def. needs major improvments. When attacking on a D-day for example, it never brings it's HQs over in the first turn, or even two turns! this leaves it's armies easy to destroy and often cost's the game for the A.I.

Also, it doesn't do enough to launch planed attack's to destroy large number of forces, and concentrates solely on taking/defending cities, no matter what the cost it seems.

[umple_day]

Hi all, nice to meet you..

Ive been playing SC for a wee while now. One thing i have noticed is the dramatic improvement in challenge by playing with FOW off - it is a slight annoyance being able to survey all deployments but the improvement in AI capability is noticable. It would be excellent if there was a feature that enabled the AI to play with FOW off and the player to have it on

Not an improvement to the AI short falls commented on above (some spot on observations by the way), but an improvement to challenge non the less

[Blashy]

Anyone notice the worse thing it does?

Not concentrate on eliminating units but damage as many as possible.

I'll have a unit surrounded by 3 of his at times and with 2 attacks my unit will be down to 1-4 and with its 3rd unit it will attack another target instead of finish off my barely alive unit.

This I see over and over and that is why I always win, I end up not spending as much MPP for new units as that AI does.

Until that is changed, no matter how different it attacks or where, you'll always win if you make it spend MPP on new units while it keeps yours alive and you just replenish, which is much cheaper.

[santabear]

Umple Day: The best results I've found are to play with FOW ON, but to turn it OFF for two turns every 8-10 turns. It helps AI quite a bit without letting the human know everything about what he's facing.

Blashy: Your observation is accurate if you're playing at +0 experience. But if you play with +1 or +2 experience bonus, the AI strategy actually works well for it. Play at beginner or intermediate level, but set the experience bonus to +2. You'll get significantly more challenging games.

[J P Wagner]

BUMP....