Playing Against the AI - Wish List for Changes

[alesha93]

Right off the bat I'll say that I realize that playing against the AI will never be as challenging or enjoyable as playing against a real person. However, the AI is a great way to test new strategies and it is always ready to play - no investing x years of time in a game only to have your human opponent disappear.

There are a couple things that I wish could be changed in the AI's logic routine (or however all that works). I am playing as the Allies in the Ironman Grand Campaign(in June 42 now) so my comments come from that point of view (i.e. the AI is the Japanese):

1. The AI throws away planes on long-term raids. For instance, if I have CAP set over airfield A with a CAP range of 2, the AI will normally not attack that airfield but it will send unescorted bombers time after time against airfield B which has no CAP over it but is 2 hexes from airfield A. It is as if the AI can't "see" that the airfield A CAP can cover airfield B. I can understand not seeing that once or twice but not turn after turn. It would be great to either stop the unescorted bomber attacks or simply escort them.


2. The AI seldom seems to use its fighters as aggressively as it does its bombers. A bunch of aggressive fighter sweeps would have pretty much wiped out most of the Allied fighter opposition early on.


3. The AI also throws away ships trying to take somewhat illogical bases. For example, the AI has tried time and time again to send small amphibious TFs to Geraldton. No moves against any of the northern Australia bases even though they are basically undefended. A CL and 4 DDs out of Perth wipe them out one after the other. I'm not sure why the AI doesn't either go after the northern Australia ports first or simple assign a decent escort to these small amphibious TFs.


4. I'm not sure how the AI sets its strategic objectives. Initially, the AI was very aggressive and took all the logical objectives; however, the beginning of the war for the Japanese is easy to script. I'm not sure there is really any strategic objective that the AI has now (maybe June 42 isn't far enough into the scenario to see it).


There are lots of other changes but those are the big ones. It would be great to have the AI be a great challenge so that I can escape to it when/if my human opponents need a break or fall away.

[Andy Mac]

I am doing a full re write of the AI at present but I am starting with a solid re write of Scen 1 once I get that up and running correctly I will seek to add sensible variants and if I have time move onto Scen2 and the Ironman series but my focus is to get a clean re write of the Scen1 AI done 1st.

As John would say the script is only 50% of the AI the rest is in the exe and thats tougher to deal with so some of the air changes are going to be outside my scope

The Geraldton attacks they are all my fault !!!

[Feltan]

I have played against more than one human opponent that makes the AI look like Clausewitz.
 
Regards,
Feltan
 

[Kull]

ORIGINAL: Andy Mac

I am doing a full re write of the AI at present but I am starting with a solid re write of Scen 1 once I get that up and running correctly I will seek to add sensible variants and if I have time move onto Scen2 and the Ironman series but my focus is to get a clean re write of the Scen1 AI done 1st.

As John would say the script is only 50% of the AI the rest is in the exe and thats tougher to deal with so some of the air changes are going to be outside my scope

The Geraldton attacks they are all my fault !!!

God Bless you Andy Mac! This has to be the best supported game of all-time (and most of that by volunteers)

[HansBolter]

AndyMac,

If you are going to put some work into revamping the AI focus on making it deploy fighters to forward bases.

I have played several different scenarios only to see the same trend. The AI never forward deploys its fighters.

It does a good job of stockpiling LCUs in decisive bases, but ignores air cover. It never bases more than a single fighter squad at Lae and never more than 2 at Rabaul.

You clear the Solomons and New guinea without any air opposition to speak of and then get recon on Truk to discover 500 fighters and 800 bombers. Even a third of those forward deployed to Solomons/NG would have made it a way better fight.

[rustysi]

ORIGINAL: Feltan

I have played against more than one human opponent that makes the AI look like Clausewitz.

Regards,
Feltan

Ouch!!! Hope I don't become one of them.

[Leandros]

oreskovich, I have seen much of the same in my first game against the AI.

And I can only agree heartily with Kull!

Fred

[Andy Mac]

I will take a look at forward deploy fighters

[chemkid]

.

[Andy Mac]

Just to manage expectiations I think I can improve the AI but it will never be as good as a PBEM player I am still using crystal ball to look at what could happen

[obvert]

Thanks Andy for working on this again! Amazing! [&o][&o]

A few things I'd offer for consideration. I'm not at all sure how many of these things are possible. Please excuse my ignorance of what may or may not be possible.

1. Coordination (and preservation) of force.

It's very easy to fight separate elements of the AI forces and defeat them. The AI doesn't seem to mass force in certain areas that are critical to defend (others mention this with moving force forward).

Maybe this is to ensure some defense is present in all areas. Perhaps a hub system could be used regionally and the AI could simulate how the Japanese used Rabaul for instance as the center of activity for a whole region. Soerabaja, Rangoon, Singers, Truk, Guam and Ambon could be hubs where you would know the AI would have significant force. Maybe in different scripts this would vary in each theatre.

If ships were also tied to these centers and told only to attack when "x" number of Allied ships were in the area "y" distance away it would keep them under air cover until an actual invasion force was near.

2. Air Combat TFs

Although smaller concentrations of force were used when it was the only option, in general Japanese doctrine would have sent the full KB against the Allies if at all possible. The decisive battle and all. I find the AI use of CVs inneffective in that they work in small TFs and present themselves in an area for too long, making them easy targets. If they were all together (must have "x" number (6?) CVs to move forward into contested airspace) and only remained for a short duration (must stay only 2 days) then this would make it much more difficult to move forward as the Allies. Especially if there is a way to have the AI react against large concentrations of Allied ships.

3. Fighter sweeps

This is mentioned briefly above, but the AI seems not to sweep much, if at all. Sweeps at max height for the airframe used would make the AI much tougher.

4. CAP settings

The AI tends to use CAP settings that aren't effective. I've mostly seen it use CAP set to a wide distance (max normal range for the airframe used) at 20-30% CAP. This not only makes the CAP unable to respond easily to threats in its own base but also makes the airframes and pilots wear out quickly. It seems also to set CAP too low to be effective against sweeps. Some CAP at max height would be very effective set to 0 hex range. Not sure if it's possible, but most players dedicate a group to CAP and another group to escort (not CAPing) to deal with this problem, and the CAP group is set to 50-70% CAP at 0 hex range.

5. TF composition

I've noticed the AI uses many different types and speeds of ships in TFs. If there were some rule about combat ships especially to normalize TFs around speed this would make them more competitive.

[wegman58]

For suggestions -

'No Go' Zones. IJN battlewagons near Pearl in late 1942 without the KB? Raiders make sense - within air range of Pearl - NOT SO MUCH.

IJN subs near the West Coast of Oz. Unless there is an attempt to take India or Ceylon I can see the Persian Gulf - Karachi route being really far, but Capetown to Perth shouldn't be safe.

AND more effort on ASW for the IJN.

[urtel]

first many tanks to Andy for unending support, you are the MAN!

i maybe have something which can help you, just finishing game as US against jap AI and have save games 1 per month for whole game (~40 saves all in different slots) so i can somehow send them to you so you can see how AI was doing and what trigger it used on strategy level...It is DBB(iron-c) scen 30 game but i think AI file is same u use...

so far i see couple of things which i think are more related to DBB than AI:
- AI was send not small number of TKs to forward bases they have nothing to do so (no port, no ships) i think at last 1 type of JAP TKs in DBB AI use in wrong manner, it maybe some conversion type...
- AI use all type of panes in attack missions including recon and transport planes some of those are kamikaze but some really try to attack ships even they not poses ordinace (dihans, myrts and betty-l from memory)

other than that AI doing fine script for closing PM was fantastic it almost force me to abandon PM....

i'm against that AI use plane max height for sweeps or at last put those in some variants we can turn off, no one was fighting at 42.000 feets in pacific so leave it on historical heights please...

[bush]

Thank you Andy! I only play against the AI so any tweaks are more than welcomed!

[GetAssista]

Andy, many thanks for the support! this is such a rare occurence in the gaming, we are fortunate [&o]

Some notes for landbased AI, hope something will be useful:
- AI is too aggressive when in strategic defence, easily lured out of fortified positions by small units. I once aggroed all Chungking garrison divisions with one Recon armor, which led to an easy capture later, while divisions marched on the plains trying to catch the uncatchable. Oh, and attacks strong opposition out of fortified bases, making sieges easier.
- AI does not prioritize land targets very well wrt distances involved and overall front situation. Sending several divisions to capture a remote city deep in the enemy territory -> long time spent -> encirclement -> death w/o supplies. Admittingly, this is happening mostly in India and Australia, not China. Maybe a system of land threat levels ~enemy force/distance? When some base nearby is captured by a significant enemy force, it is usually not ok to set course to some remote enemy city, you have problem at hand already.
- AI does not reconsider ground movement plans much. This is connected with the previous point. I saw divisions stubbornly marching all the way from Karachi to Calcutta on foot while Bombay was contested.
- I don't ever see AI using strategic RR movement, would've helped it in India/Australia
- AI is very passive after units arrive to a hex near their destination and decide that the enemy is too strong for them to enter. Mostly sit in place, presenting an easy target for encirclement. And generally I'm not sure if AI ever considers retreat if units appear running out of supply.

[Andy Mac]

Woooooaaaaahhhh

Wow I guess the AI players are alive and well [:D]

I need to manage expectations I am looking at the scripts I dont have coding support for this mini project.

I will do my level best to clean it up and fix what I can but some of the wish list above is out of my ability to fix I am looking at the scripts which are about 50% of the AI and yes its easier because of the changes Michael has made to understand whats going on but I am not a miracle worker !!!!

Lets see how we get on fixing one script for scen 1 and doing that full review then we shall see what we can do about some of the other stuff.

Andy

[Andy Mac]

I think I can fix the PH no go zone

IJN subs I can take a look at

ASW is tricky because of how the Ai forms TF but will do my best

ORIGINAL: wegman58

For suggestions -

'No Go' Zones. IJN battlewagons near Pearl in late 1942 without the KB? Raiders make sense - within air range of Pearl - NOT SO MUCH.

IJN subs near the West Coast of Oz. Unless there is an attempt to take India or Ceylon I can see the Persian Gulf - Karachi route being really far, but Capetown to Perth shouldn't be safe.

AND more effort on ASW for the IJN.

[Andy Mac]

Yes please a.mcphie@btinternet.com

Try and pick any that you trhink the AI was doing something moronic in

ORIGINAL: urtel


first many tanks to Andy for unending support, you are the MAN!

i maybe have something which can help you, just finishing game as US against jap AI and have save games 1 per month for whole game (~40 saves all in different slots) so i can somehow send them to you so you can see how AI was doing and what trigger it used on strategy level...It is DBB(iron-c) scen 30 game but i think AI file is same u use...

so far i see couple of things which i think are more related to DBB than AI:
- AI was send not small number of TKs to forward bases they have nothing to do so (no port, no ships) i think at last 1 type of JAP TKs in DBB AI use in wrong manner, it maybe some conversion type...
- AI use all type of panes in attack missions including recon and transport planes some of those are kamikaze but some really try to attack ships even they not poses ordinace (dihans, myrts and betty-l from memory)

other than that AI doing fine script for closing PM was fantastic it almost force me to abandon PM....

i'm against that AI use plane max height for sweeps or at last put those in some variants we can turn off, no one was fighting at 42.000 feets in pacific so leave it on historical heights please...

[Andy Mac]

Most of these are so ingrained in the AI that it will be tricky to resoilve will do my best

ORIGINAL: obvert

Thanks Andy for working on this again! Amazing! [&o][&o]

A few things I'd offer for consideration. I'm not at all sure how many of these things are possible. Please excuse my ignorance of what may or may not be possible.

1. Coordination (and preservation) of force.

It's very easy to fight separate elements of the AI forces and defeat them. The AI doesn't seem to mass force in certain areas that are critical to defend (others mention this with moving force forward).

Maybe this is to ensure some defense is present in all areas. Perhaps a hub system could be used regionally and the AI could simulate how the Japanese used Rabaul for instance as the center of activity for a whole region. Soerabaja, Rangoon, Singers, Truk, Guam and Ambon could be hubs where you would know the AI would have significant force. Maybe in different scripts this would vary in each theatre.

If ships were also tied to these centers and told only to attack when "x" number of Allied ships were in the area "y" distance away it would keep them under air cover until an actual invasion force was near.

2. Air Combat TFs

Although smaller concentrations of force were used when it was the only option, in general Japanese doctrine would have sent the full KB against the Allies if at all possible. The decisive battle and all. I find the AI use of CVs inneffective in that they work in small TFs and present themselves in an area for too long, making them easy targets. If they were all together (must have "x" number (6?) CVs to move forward into contested airspace) and only remained for a short duration (must stay only 2 days) then this would make it much more difficult to move forward as the Allies. Especially if there is a way to have the AI react against large concentrations of Allied ships.

3. Fighter sweeps

This is mentioned briefly above, but the AI seems not to sweep much, if at all. Sweeps at max height for the airframe used would make the AI much tougher.

4. CAP settings

The AI tends to use CAP settings that aren't effective. I've mostly seen it use CAP set to a wide distance (max normal range for the airframe used) at 20-30% CAP. This not only makes the CAP unable to respond easily to threats in its own base but also makes the airframes and pilots wear out quickly. It seems also to set CAP too low to be effective against sweeps. Some CAP at max height would be very effective set to 0 hex range. Not sure if it's possible, but most players dedicate a group to CAP and another group to escort (not CAPing) to deal with this problem, and the CAP group is set to 50-70% CAP at 0 hex range.

5. TF composition

I've noticed the AI uses many different types and speeds of ships in TFs. If there were some rule about combat ships especially to normalize TFs around speed this would make them more competitive.

[Admiral Scott]

I havent played this game since 2010 because the AI was so terrible. I would like to try again after Andy makes his changes to the A.I.

I see there were patches back in 2012, and March of this year, how is the A.I. right now after these patches?