AI Pet Peeves

[Edwin P.]

Here are a few pet peeves of mine regarding AI actions in SC1.

- Malta: Axis AI moves 2 transports adjacent to an unoccupied Malta. Seeing this I move a transport so that it is adjacent to one of the AI transports (I lacked the AP to reach the port by 1 hex)and an Italian Battleship. What does the AI do? It retreats both transports and the Battleship back to Italy. What should it have done? Landed one tansport in Malta to capture the undefended island, retreated the other transport and blasted my transport with the Battleship.

- UK: As the Allies I can withdraw all forces from the UK, Canada, and the USA knowing that the Axis will never invade until Russia is defeated.

- Italy: As the Italians I can withdraw all forces from Italy and the Allies will never invade.

- Egypt: UK can abandon Cairo and Italians will never invade Egypt.

- French Corps in Beruit - AI will never move the French Corps in Beruit to Egypt so that it becomes Free French when France Surrenders.

- Axis AI is threatened in France, Italy, Russia and Greece. Instead of concentrating overwhelming force in one area and staging a holding action in the other areas it spreads itself too thin trying to deal with all of the threats at the same time - including sending 2 armies, one armor and one air fleet to the mountains of Greece. I would much rather see the AI do what Humans do - Adopt a defensive posture in Russia, operate units and HQs West and crush the Allies in France, then Crush them in Italy before returning the units to the Eastern front.

- Axis AI will allow large numbers of units to be surrounded and cut off in southern Russia. If a large Russian force is coming down from the south behind their lines it does not know to withdraw its units to safety - in a recent game I isolated and destroyed 4 HQ units, several German and several Italian Armies and Corps using this strategy. A human player would have withdrawn units from the front before they got surrounded or concentrated airpower to destroy the Russian pincer movement.

And as Night says in his post below the Allied AI will fail to do the same to the Germans.

- AI Does excellent job of attacking and destroying isolated units; however, when facing an enemy in a line it fails to concentrate air power in one area of the line. The result is that although the AI knocks a unit down to 2 or 3 the Human player reinforces it back to full strength on his turn. (NOTE: IN SC2 the Human player will not be able to do this as the maximum level of reinforcements is affected by the number of enemy units adjacent to yours).

- AI does not concentrate Air Power to achieve objective. Example: Allies take Sicily and move in 2 air fleets for support. AI attacks with only 1 army or armor unit turn after turn when a concentrated strike force of 4 or 5 air units could decimate the sole land unit defending Sicily. Similary for when a single allied land unit holds Brest in France. The AI should concentrate its air power to eliminate this unit instead of allowing the allies to continually rotate in new units to hold the city and divert Axis forces from the Eastern front.

- UK AI should, at least in some games, follow the tried and true strategy of sheltering their air force from attack in Canada and Northern England and wait to spring it upon the allies when the USA enters the war. Otherwise they lose a lot of MPPS just replacing their losses and have nothing to show for it.

- IT AI does not know how to handle its Naval units. When it is outnumbered it goes out to fight and when it dominates the Med it stays close to port not hunting down the enemy.

IDEA: IT AI should know how many Enemy Naval Ships (not transports) are in the MED. IF AI has 2:1 Advantage then it hunts them down (100% offensive strategy) by moving its naval units as if they were one massive fleet. IF AI has 1:2 disadvantage then its Navy seeks shelter in the Adriatic (100% defensive strategy)

IDEA: IT AI Should build a sub if only that is needed to give it a 2:1 advantage in the Med.

- IT AI does not use its sub to interdict allied merchant shipping in the Med, even when there are not allied ships in the Med.

- IT AI does not use Air Units to spot location of Enemy units in the Med.

- AI needs a better research strategy, except for the Italians which somehow seem to do quite well.

- The Axis AI should know how many Allied ships have entered the Med as Axis spies were active in Spain, Egypt and Turkey, not to mention Greece, Cyprus, Malta, Beruit, Corsica and Algeria. This would maximize the ability of its fuzzy logic to adopt the correct strategy and tactics in the Med Theater of Operations.

a. IF AXIS Naval Superiority >= 2:1 Then 100% switch to an Offensive Naval Strategy in the Med.

b. IF AXIS Naval Superiority <= 1:2 Then 100% Switch to a Defensive Strategy in the Med.

c. IF AXIS Naval Superiorty >= 1.5:1 Then A) Build More Naval Units or Build Naval Air Units or C) Adopt Defensive Strategy or d) Research Gun Laying Radar

- Axis AI never makes serious effort to take Malta

[ May 26, 2004, 10:00 PM: Message edited by: Edwin P. ]

[Night]

I agree on all points and have said it before that the AI's main problem is what you described in your 2nd to last point but in reverse, the AI lacks the strategic foresight to try large scale maneuvers to destroy enemy forces, this is esp. bad for the AI when it is basicly the best/fastest way to win the war in russia from ethier side, and more-so for the Russians. I conduct operations like the one you describe on the AI often and with devestating results, making the game very unbalanced.

The AI should realize that destroying mass enemy units is often more important then just trying to take/defend cities and resources.

[Retributar]

Excellent observations 'Edwin P.'and great analyzation of Edwins post 'Night'...this is truly the crux of the heart-of-the-game!.

Now...how can this problem be overcome SO....as TO say on Star-Trek...'MAKE IT SO!'.

[Raven25]

Please note that as human players we have advantages. For example, we can make predictions as to what the AI will do. The AI cannot. It can only make actions that are programmed into it, and movements and attacks based on the human player's movements and attacks.

[James B.]

I agree with Raven. We are asking for critical thinking skills from the AI sometimes. However, several of Edwins points are entirely plausible. The AI should be able to tell that a transport cannot attack, as in his first suggestion. Also, there is no way for the computer to know that all of the forces in Italy are gone if FOW is on. It's just not logical. And if you don't have FOW on, don't be so picky about things like this, you aren't being realistic anyway. Anyway, just be careful about asking the AI to think too far into the future. It is near impossible for the AI to think so far as to say "Wait until USA joins" without scripting that randomly in. Otherwise it's a good idea. Mass troop movements are not the specialty of a machine whose basic function is to cruch numbers. It analyzes things based on formulas and their effects. Strategy for the AI will never be able to compete with a human's, but you are right that it can be improved. I believe Hubert has mentioned that "several algorithms have been identified [and fixed]". But lets not expect too much, after all, it is a machine.

[Edwin P.]

Also, there is no way for the computer to know that all of the forces in Italy are gone if FOW is on.

Not quite, if you look at the Reports screen you can see how many Land units that Italy has. Now if that many units are spotted by your units in other areas of the map and you assume that three or four are in Yugoslavia and Albania you will know when Italy is undefended or is defended by only one or two corps and ripe for attack. (Human players rarely assign German units to Guard Italy)

You can also use the reports screen to determine how well the UK is guarded by taking into account how many UK units you spotted elsewhere and to track how many allied units are in transports. In fact when playing a game as the Axis I keep a record of how many Allied Naval Units I have sunk to help me determine my naval strategy (as the Allies will never, almost never, build new warships). Total UK Ships Less Warships Remaining = Number of Transports at Sea.

The key problem is that the AI does not analyze all available data and use that to influence what action to take.

Drat - I just gave away two of my tricks for securing victory. :mad: However, no one will remember to use the reports screen in this manner or to track the number of ships sunk.

In my opinion the best way to help the Italian Naval AI in the Med would be to allow it to know how many Allied warships are in each section of the Med - East and West of Malta. The AI would not know exactly where enemy ships are and it would require a search for the AI to find them, but at least the AI can adopt its strategy and tactics to the level of the perceived balance of forces in each section of the Med Ocean.

Similarly if the Axis AI knows how many Allied warships are in the Med it should be able to deduce how many warships remain in the Atlantic and be able to use this information to guide its operations there.

Taking this perhaps a step too far, perhaps in the map editor players could create sea zones. The AI (and only the AI) would know how many warships are in each zone. This would allow the AI to adopt its naval strategy to different maps.

[ May 27, 2004, 02:58 PM: Message edited by: Edwin P. ]

[Retributar]

Edwin P. those are good ideas for victory... but i prefer myself not to do that...(keeping track of enemy units)as i want to have that extra uncertainty in order to increase the tension in the game.

Infact...i would like to be able to give the AI varying degrees of EI(Espionage Intelligence)...so that it could on different occasions see exactly what i have where...so that it would know exactly what i had left in England or wherever.

Maybey then it would launch a surprise invasion on me?.

[Edwin P.]

Infact...i would like to be able to give the AI varying degrees of EI(Espionage Intelligence)...so that it could on different occasions see exactly what i have where...so that it would know exactly what i had left in England or wherever

Excellent idea, as you are aware of the human tendency to lure the AI into ambushes by leaving cities undefended, thus partial intelligence might be worse than no intelligence.

[Les the Sarge 9-1]

A lot of very fine comments, but this line spoke volumes to me.

"I would much rather see the AI do what Humans do"

The trouble is, I am a human and I have an organic brain inclined to do random actions which might be clever as much as dumb.

The AI will never have stayed up all night drinking, nor will it cheat. It won't rush it's turn because another human told it it was dinner time and to get off the bloody computer.

Basically the AI CAN'T do anything I can do at all.

All it can do is respond to mathematically arrived at conclusions.

If this happens do this.

If given these choices, do them in this order of priority with these peercentage chances for each selection.

That is NOT thinking.

I can't process numbers in my head as fast as a calculator, but that in no way makes me feel inferior.

A computer AI will never do something and then when found out to have done something dumb claim "I meant to do that eh, really".

[Hubert Cater]

AI implementation (for me and for SC) is tricky, and while the option in SC1 was available to me to have it do many of the items you've requested, the problem is that there is no guarantee that it would do them necessarily well. Problems like you've mentioned, (under FoW) like baiting or luring the AI into ambushes by leaving cities undefended and many others left me with the idea of playing it safe and sticking with what I knew the AI could handle. IMO it was better for the AI to stick with what it could do (decently) well and not with a list of items that it could do but in the end do poorly.

Just some background but at the same time it's not to say that there is no room for improvement.

[Night]

Im glad you responded on this Thread HC since the buck stops at you on what the AI can and cannot do.

I feel that if you can make the AI do some or all of the things we mentioned, then it should be given a shot. If it can only do it decently, then let's beat up on it til we can figure out how to make it do it well. It can always be changed. However I think we all agree the AI for SC1 atleast is far less then what we need in terms of strategic planing and execution.

Of course we will run into the problem where people will become familer with the AI's complicated planing and can reconize the signs, thus avoiding them or turning the situation into the Humans favor, making even more problems for the AI.

[SeaMonkey]

Likewise, thanks HC for the orientation. Being past the AI now and only into PBEM I haven't considered AI capability for SC2. But what I think is important is the scripting of the AI using the editor. This will allow the AI to be choreographed for a custom scenario and although it will still be an easy victory eventually for the human, the first time an unexpected player takes on the AI he will be in for some surprises. It will be up to us SC patrons to concoct some custom campaigns for each other to enjoy. This will allow us(the ones into scenario making)to inject a little of our playing style into our custom campaign quiver for the rest to test their SC prowess.

[ May 27, 2004, 08:02 PM: Message edited by: SeaMonkey ]

[Edwin P.]

Many Thanks for the post HC

Along with the others I think that the key point is what HC said related to focusing on what the the AI can do well

At the strategic level I think that in SC2 (and correct me if I am wrong) with Amphibious transports that the AI should be able to execute the following strategic moves well with special scripts:

1. Conquest of Norway - if no allied ships block access to the coast.

2. Conquest of Sweden - either before taking Norway or after taking Norway

3. Conquest of Greece - if Italian Navy has superiority in the Med

4. Conquest of Egypt if Allies leave it weakly undefended and if successful a conquest of Iraq.

5. Defeat of a weak Allied Naval Force in the Med.

Example: As I said in an earlier post, if the AI has a 2:1 advantage in a sea Zone it should build be free to hunt down and destroy the enemy naval forces in that Sea zone.

Anything wrong with this logic? Is this something that the AI should be able to do well?

6. Interdicting UK merchant shipping in the Med if No UK Naval Forces in the Med or Air units in range to attack the sub.

Example: IF No Air unit in Malta and Libya is Axis controlled and Algeria is Axis Controlled or Vichy Controlled AND no Allied Naval Units in Eastern Med then Move Italian Sub to randomly determined merchant ship interdiction tile close to Malta.

Now, what is wrong with the logic in this script? To me it appears to be simple and something that the AI can do well.

7. Bombard Malta to build experience. If Italian Navy has 2:1 Advantage in the Med and a LAND unit is in Malta THEN the Naval units should surround it and continually bombard it to gain experience until Experience Level 4 or LAND unit is destroyed.

What is wrong with this script. Perhaps this is another task that the Italian Navy can do well?

What will it take to make sure the AI does these tasks well? What should the conditions be for the AI to consider executing these tasks? Any reason why these proposed strategic moves should not be part of the AI's library? Any comments? any ideas for making these proposed tactics better?

Another Item to consider is the Italian Build and Research Strategy.

-- Say if Italy has 5 Ships in the Med and UK 3 then perhaps the Fuzzy logic should consider building a sub or a Bomber after Italy Takes Greece. This would give it 2:1 Superiority and a force strong enough to sink the Brits without any losses if no UK airpower is in range.

-- If Italy has 4+ surface ships then it should consider research Gun Laying Radar with 1 or 2 chits to give it a chance to destroy any allied Naval Force that comes into the Med.

[ May 27, 2004, 11:12 PM: Message edited by: Edwin P. ]

[SeaMonkey]

Damn it Edwin now you have done it. I had to tear apart the office looking for my TOAW manual, specifically the Event Editor. Why? Because Norm Koger wrote a most amazing engine that scripts the AI to do a number of tasks, ..well. Here is how I believe the generic event editor should be structured. From TOAW the event has the following elements:

1. a trigger

2. an effect

3. a location

4. a value or radius

5. a chance of occurrence

6. a turn range

7. a news bulletin for the report screen when it occurs

8. a status

Simply by filling in the blanks(with the scripted choices) for these parameters will set in motion the things you have been asking for. The events are numbered in sequence and one event can activate or cancel another. They are designed to come off chronologically so event #150 could supercede event #10 if time is the trigger but not necessarily needed to be the trigger. Among other things IMO this was the most precedent setting feature in computer wargaming up to this point and is very powerful, but somewhat complicated, although not necessary for scenario design, completely optional.

[Edwin P.]

Thanks SeaMonkey, it will be most interesting to see how HC implements scripting and to what extent. The TOAW options you dug are are most detailed and comprehensive and are giving me much food for thought and contemplation.

I would like to be able to write scripts that vary the stategy the AI uses to defend France. For example 50% Standard, 25% Scrap French Navy for MPPS, 20% Scrap French Navy and French Air Unit for MPPS and 5% advance into Germany(!). In general copy the strategies of the various top players.

It would be interesting to know if HC has several general strategies for each AI player's units that can be triggered. For example I would like to see the Italian Navy, at the simplest and most basic level be able to switch between a defensive posture, an aggressive defensive posture and an offensive posture at the appropiate trigger, related as I suggested earlier to my suggestion for the AI knowing how many enemy warships are in each sea zone (Western Med, Eastern Med, South Atlantic, North Atlantic, Baltic, Black Sea) and the accuracy of this being related to the AI intelligence level (0% Novice to 100% Expert)..

Think of these AI intelligence zones as being a variant form of the exisiting user defined weather zone but they allow the AI to know the number of enemy forces by type (Air, Land & transports, Naval) in the zone with varying degrees of accuracy. (ie it uses some of HC's already existing code in a new way)

Just this one change would allow the AI controlled Italian Navy to play a more active role in the game, do it well and make scripting a much more powerful tool. Why? Then the triggering events can be based on the relative strength of the two forces in a sea zone and this can be used to change the combat posture that the AI takes.

Example: The Italian Navy has superiorty in Sea Zone X then it is more likely to adopt an offensive posture in Zone X. If the UK forces are superior in Sea Zone Y then it will likely switch to a defensive posture in this sea zone or to move its forces to another sea zone.

Moreover, these routines can be used by Allied AI naval units to enhance their ability to wage the battle for the Atlantic, and I really want to see a more interesting AI directed battle for the Atlantic.

Or imagine this, the Axis AI believing that the Allied Naval Forces are weak in the Eastern Med execute a campaign to capture Egypt. An Italian Corps sails past the Allied Naval Units in the dead of night to seize an unguarded Suez canal in a daring strike while Axis naval and air units foces attack the Allied fleets trapped in the Eastern Med. After sinking them Axis Units attack Egypt and seize Cairo. Then do they turn their attention West to take Malta and then Vichy France, do they head East to take Iraq or do they return to Europe to prepare for the Allied invasion of France?

[ May 29, 2004, 10:19 AM: Message edited by: Edwin P. ]

[Retributar]

I agree...'TOAW' had a good AI Program...not perfect, but good enough to make the game a worthwhile effort!.