Questions on the poor AI

[Jeremy Pritchard]

A lot of people have talked about problems in the AI, and are there particular things that they are poor on?

I know of a few, such as weakly defending the Home Bases (US West Coast, Japanese Home Islands), as well as the North Seas area (i.e., North of Japan).

Are there any other weak areas?

I have already 'fixed' some problems with Japanese units in the South Pacific, notably New Guinea, but are there other bases that are easily attacked as they are poorly if not completely undefended.

Any suggestions of weak areas in Japan and the Allied defensive AI zones that makes attacking too easy?

Unfortunately, a lot of the tactics that I am using to beef up the Japanese and/or Allies (in their respective AI OBC files) is to give them early deployment in some bases (that are never defended, like Truk, Wotje, Maleolap, etc...). However, one might 'explain' this, as these forces would be released by Imperial GHQ due to an early Allied breakthrough of their lines. If the human player encounters the 52nd IJA Division attacking Truk in 1942 (instead of when it really was deployed in 1944), it could be explained that since the Allies are attacking into the defensive zone 2 years earlier, one could believe that the IJA deployed divisions to the Pacific Islands earlier to make up for the early advance. The Japanese stripped their China and Manchurian defenses when the Allies started their breakthrough in 1943-44, so one could propose that shoudl the breakthrough happen sooner, so too would the reinforcements.

So, chances are, when you are fighting the Japanese or Allied AI, you will encounter units that you should not really meet until 1943-44 years early, but remember, that you are also breaking history by attacking into the Japanese homeland sooner.

However, the PBEM OBC files will remain historic, so when you are playing an opponent, you will rely on historic tactics to hold off the Allies until you get the reinforcements in 1944.

Unfortunatley, I would have perfered triggers for the deployment of units instead of a timeplan for reinforcements. When the Marshall Islands and/or Soloman Islands fall, reinforcements for the Marianas and Philippines should be released.

[Denniss]

When fighting in Rangoon the southern Moulmin is usually not defended so as allied player it's easy to land and cut off supply for the entire area north of it .
Maybe change it to land base or let the AI have a division or a smaller unit in this base .
AI never uses Andaman Island as strategical Airbase . It's easy to stop all Routine convoy shipping in this area from this base with one G4M (or Wellington) unit

[Capt. Harlock]

The Japanese AI doesn't seem to do a good job of securing the all-important oil supplies. It is vital to invade Java, both to capture the two oil resource bases on the island itself, and to prevent the Allies from isolating Palembang with AZOCs. Both in V. 2.3 and 3.1 I've noticed the AI is not interested in invading. (In my current game the AI has decided to go for Koepang instead, which amounts to two Japanese divisions voluntarily interning themselves, since I managed to get an American division dug in before they arrived.) Also, the Japanese AI does not march any units from Manokwari to Sorong, which is totally undefended.

Incidentally, I notice that the Oil Resource points have been moved from Rangoon to inland bases. Will the Japanese get the oil points if they capture the bases, since they are not accessible to Routine Convoys?

[Jeremy Pritchard]

The main reason that I switched targets (Ambon and Koepang before Java) was that they were actually secured before Java (as it was important to cut Java off before invading it), as well as the Japanese only occasionally landed troops there after securing Java.

I might toy with a different order of priority of base invasions, to see what happens. However, being unhistorically aggressive as the Allies will always result in the Japanese AI being stumped.

I moved resources from bases in Burma to Rangoon, as I did this with most major ports (had them as the 'collecter' area for export of raw materials). So, there will be a few very large resource bases that are actually funnels for many inland bases and smaller local ports.

I am thinking about removing Moulmein altogether, as it does a lot to stump the Japanese AI.

I have attempted to do a few things about Andaman, and it appears like the IJA AI tends to invade it, but does not defend or build it up. I might have a way that enables them to do this, but, like everything, will get messed up if the Allied player is too aggressive.

[madflava13]

I heven't played Pacwar in awhile, but my strategy was always to hit Timor/Medan/Palembang as the allies. Timor and Medan never seemed to be well defended (1 Bde at the most).

[pasternakski]

Jeremy, I think that the best improvement will be what you mentioned on another thread about designing four or five strategic approaches for the AI and having the game secretly select one of them at the beginning (especially for long campaigns, and particularly for the campaigns beginning in 1941).

Eventually, you will be able to figure out which one the AI is following, but you will be kept guessing in the crucial early period, which, as far as I am concerned, is exactly where you should be. The Japanese AI, in particular, should be much tougher to beat, and that's the side needing the most help.

[CynicAl]

One big problem I've noticed is that the AI doesn't garrison captured bases well, or at all. Malaya and Burma have been mentioned, but I've seen the same happen in New Guinea and across the Central Pacific. The AI (especially as Japan) just doesn't keep anything in its rear areas. So the forces it does deploy get bypassed and cut off.

On the offense, the problem is target fixation. The AI isn't flexible enough to say, "Okay, that target is too hard. Time to back off and try again somewhere else." Instead, the AI just keeps throwing ships and troops into the meat grinder until it runs completely out of APs. In a 2.2 game I massively reinforced Timor, and beat back one penny-packet landing attempt after another until they finally stopped in about October 1942, IIRC. I do recall that after the fighting off Timor ended, I never saw another IJN transport in the game. Which, granted, wasn't too much longer after that - but still!

[Jeremy Pritchard]

Target fixation is part of the game. Without it, the AI will just skip over 'tough' bases and never invade them, leaving places like Bataan as Allied control throughout the game. Realistically, the AI only gets stumped when unhistoric things are done (like sending a US Army Division to Timor).

A way that I have helped limit this was to change arrival times for many US Divisions. In version 3.2, many US Divisions appear later (for some reason, the 31st Division appeared in 1942, when it should appear in 1944). What I did for the special OBC41 to REALLY help the Japanese are three of the following.

#1. Placed many historic defensive units well ahead of schedule. Virtually every Japanese base in 1941 has a ground unit that was there in 1944. This was done because, as CynicAl said, the Japanese never defend certain Islands. This way these islands will be defended, and will even be defended should the Allied player attack in 1942, 1943, 1944 or 1945. It may be unrealistic, but pretend that they are placed there after a certain trigger is done, like if the Marshall Islands were invaded, they would appear.

#2. Increased the time of arrival for many early US Army Divisions. In the regular versions of the OBC files, there are 6 US Army Divisions out ready to be deployed in 1941. However, 4 of these were used as West Coast garrison units for the first few months of the war, however, the player knows full well that the AI will never attack the West Coast and sends these troops off to Timor. I changed all US Army Division arrival times to match the time that they were freed from serving on the West Coast. So most have a 2+ month delay, which will hopefully give the Japanese AI some time to consolodate the East Indies before US troops arrive.

#3. Added time triggers for auto deployment of certain units. Some bases cannot be guarded by deploying troops there in 1941 like my solution #1, primarily because in 1941 these bases are controlled by the Allies! So, on certain dates, IJA units are 'automatically transported' to these bases, and after much testing, tend to remain there. I have only really done these to bases that would usually be under the control of the Japanese, or totally undefended. Certain bases, such as New Guinea and the Central Philippine Islands (the Islands between Luzon and Mindanao) rarely ever got troops on them, but in January 1942, an auto trigger puts 3 Divisions on New Guinea (who were supposed to only arrive in 1943, but the Allies will always be there before then, but don't worry, they tend to just defend and not attack), and another trigger in 1943 to move and place IJA Divisions and Brigades on the Central Philippine Islands, plus a few other minor unit placements. This was difficult to do, because I am fully aware that many of these bases might be under Allied control on these dates, and would feel kind of weary having a division appear behind enemy lines on a certain date, but with the reductions of US Army divisions in the field, hopefully holding off the Japanese will be a little trickier.

A lot of what is done can be explained away by using 'fake triggers', that being, just pretend that these units are not there until you attack certain bases (i.e., as the Allies, you know that these bases are defended, as should you break through the barrier in 1942, the Japanese sent troops out to these islands to beef up their defenses, rather then doing this in 1944).

I tested to see what will happen with option #3 in a PBEM game, and it seems like these 'time triggers' do not occur in PBEM games, as I used these triggers in the same way as HQ transfer triggers are done, and the HQ triggers do not seem to work in PBEM games. So, when you play a PBEM game, you will not find any of the above in the exe or scenario files. The Japanese will be left with the historic 1941 OOB to take the Pacific, and the US will be given all 6 divisions to do whatever they want with (but will be way more cautious for fears of daring Japanese attacks).

[screamer]

relax i forgot to garrison rear new guiney bases too exept for 2divs at lae. cost me dearly now my jap forces are fightin to their deaths in the first succesfull allied offensive. december 43. 3000tojo's in pbem but its a version even earlyer then 2.3

[Jeremy Pritchard]

I also noticed that the use of battalions guarding certian bases, notably Borneo, really slowed down the AI. As in order for the AI to take these bases, they had to plan massive invasions, taking up a lot of transport and units that could be better spent taking those more heavily defended islands, like Timor, Ambon and Sumatra. Since these units really offered very little resistence historically, I removed them as they cause more trouble to the AI then they actually are worth, and don't deter a human player either. So, I removed all batallions that were not reinforced, or less then 1000 men (such as the IJN SNLF battalions, USMC Defense Battalions, AUS Reinforced battalions in the East Indies, these formations had 1000, and usually 1500 men in them).

[Capt. Harlock]

Just to defend my side, I think that putting a US Army division in Koepang is entirely legitimate. First of all, it became a SW Pacific base--not my decision, but done by the computer when I put ANZAC and SW Pacific under computer control. Second, it was mid-April 1942 when my troops arrived; I managed to buy some time at the cost of two CA's, six CL's, and Adm. Doormann. Third, and possibly of greatest interest to Jeremy, the West Coast is more than adequately defended by the units attached to the 4th Army. There is ZERO threat there because the Japanese AI has not moved East. The CV's are operating in the Southern Phillipines, and the Allies still have Wake.

BTW, you might want to talk to Mika or someone else familiar with the source code of Pac.exe. I think I read that Gary Grigsby programmed three or four AI strategies into the original file. It may not be necessary to write multiple copies of the Pac.exe file.

[Mika Väliviita]

First, I have not seen the source code, I've been hacking the exe.

About the multiple strategies, I haven't figured out how it works. There is a strategy variable that can have way more than 4 values, so it looks like there are substrategies also. But this is really hard to understand via hacking. And as for the part I do understand, it's really hard to see the 'why and how' of it. There are lists of target bases for many (not all) HQs, some overlapping depending on the situation, but then there is code that overrides these lists based on some other situations.

[Jeremy Pritchard]

I was not attacking your strategy, just trying to figure out a way in defeating it. As it does appear to be a very common one.

We have not looked at the source code, and can only really see a bit into things. I am looking into creating a program outside of PacWar that will enable you to choose a few different EXE files, either by choice or by a set perameter of random possibilities of choice. Hopefully I can get the program to choose the required EXE file, which will then rename if from something like PacIJA.exe to Pac.exe and then run the execute (as Mika informed me that the execute must be called Pac.exe). I think that this would be a lot easier to do then to go looking for it in the exe.

The main reason why I want to delay the arrival of some of these forces is priarily because they were deployed on the West Coast as defense, and remained there until they were replaced and could be sent out. Realistically you don't have to worry about any attack, so you will probably send these out, which is ok, except that it gives you a lot of extra manpower to use against a pretty deficient AI. Against a human you will be more cautious, and therefore these divisions will be at the West Coast in December 1941.

I sort of see these divisions like the 52nd, 53rd and 54th IJA Divisions sitting on the Home Islands. Realistically, there is not much of a threat for the AI to attack the Home Islands until 1945, so why have these 3 divisions sit around there for 3 years? If the US can move their home garrison units to the Pacific, would it not be fair for Japan to do the same? These 3 Divisions deploy to the Pacific on set dates, all that I want to do is to do the same for the US divisions for a game against the AI.

I think that it will delay and possibly eliminate the possibility of securing Timor/Ambon (without using British/Indian/Australian or one of the very few US divisions) primarily because the first two divisions recieved as reinforcements arrive in April 1942. Plus I have put these two bases on the Combined Fleet's "to get" list, so should they cause significant trouble, expect to see the IJN pounding on your door. I have also lowered the level 4 airbase of Ambon to a level 2, as the RAAF has gone through a major facelift and the bombers are no longer there.

I really want to do things without cheating too much. I don't want to give the AI experience bonus, or greater production, as that would not be 'fair'. I just want to make up for deficiencies, which is primarily caused due to deployment issues, as well as getting REALLY stumped should they encounter strong resistence early in the war. I am trying to base all of my restrictions off of historic possibility, as well as probability. Historically, these units were not deployed from the West Coast until these new release dates. It would be just like me having them attached to 4th Army, but have automatic HQ changes on set dates (like the 3 IJA Divisions attached to Imperial GHQ).

I was not knocking your strategy, just trying to figure out how to stop or weaken its potential as it really messes up the AI, and I want you to REALLY, REALLY have to work in order to get it going (not just REALLY have to work )

[Capt. Harlock]

Sometimes the simplest solutions are the best. My humble recommendation would be to just invade Java first. The sooner the Japanese can get the oil flowing, the better. Delaying the arrival of the US troops will not stop a clever Allied player--he'll scrape together something, possibly from the Phillippines or even Singapore. (After all, ABDA involved British and Australian units.) Likewise, be careful about lowering the size of the airbases. By April 1942 the Allies have several engineer units at their disposal, so a base that is small need not remain that way. (I have a unit in Dili which is helping me considerably in the struggel for Koepang.)

[CynicAl]

Persistence can be a virtue, but the AI carries it to ridiculous extremes. Certainly it did in the Timor Game I mentioned earlier - my second game ever. (I finally turned up the notes I made as I was playing that game.)

I rushed two Engineer units to Timor right about the start of January 1942. By April, when the AI finally got down to that end of the DEI, Timor was a size 8 airfield, hosting ABDA HQ and a large LBA force. In support, I had Asiatic Fleet in Darwin with most of Pacific Fleet's combat assets.

The AI tried twice to sneak small landing forces into Flores in April, with no luck. In early May Kido Butai showed up with 11 CV and CVL in 3 TFs; 6 flattops and many escorts were lost to a combination of Timor-based LBA and CV TFs Reacting from Darwin. One month later the remaining 4 CV and Hosho came back for another round, and were dispatched handily. Two weeks after that, the IJN battleline tried to slip in and bombard, taking heavy losses and accomplishing nothing.

For the rest of 1942, every time the AI got a new carrier it immediately formed up a small Air Combat TF and sent its one operational flattop on a deathride to Flores. All this time, the AI kept sending landing forces to Flores every two weeks, only to be repulsed by LBA and surface combatants every single time; the slaughter finally ended when the AI ran out of APs. And everything that followed the first CV battle in May was just pointless and stupid. A "persistent" opponent would back off, build up, and come back when he had a chance of success. Not continue to squander his remaining assets against an overwhelmingly strong position until only a shortage of LCUs and transports on my part prolonged the game into 1944.

(EDIT)And I did it without moving a US Army division to Timor. In fact, since my LBA kept the AI from ever successfully landing troops on Flores, much less Timor, apparently the only LCUs I needed there at all were the two Engineers. (/EDIT)

Otherwise, I like the ideas being discussed here. I especially like the elimination of the useless small units on Borneo and elsewhere - even more especially if the slots vacated can be used to beef up the permanent defenses of the Japanese Home Islands.

[Jeremy Pritchard]

Along with the Divisions appearing later, so too will many of the US Engineers. I find that there are just way too many US Engineers out there that they can start building up airfields everywhere in January 1942 (like as you said). However, they did not quite have such a large airfield and port expansion ability until later in 1942.

[Capt. Harlock]

One more "Stupid AI trick" that I have just been reminded of: the Allied AI has an amazing habit of sending Transport and Cargo TF's into harm's way without proper escort. In my latest game, the Computer-controlled South Pacific HQ has just loaded an infantry division from San Francisco and is sending it to Darwin. (DARWIN??) The TF consists of a single "stack" of Henderson AP's -- there is not a single escort ship.

[LargeSlowTarget]

Originally posted by Jeremy Pritchard

I am thinking about removing Moulmein altogether, as it does a lot to stump the Japanese AI.

True, in all my recent games (set up as 'both human' with the Japanese HQs all under computer control) the Japanese LCUs never marched from Bangkok to Moulmein on their own accord. They always had to be ordered to march to Moulmein and on to Rangoon by the 'AI combat advisor' - myself. Reaching Rangoon, the units went off to capture the rest of Burma on their own, including Myitkyina (I'm using my own modified obc) and Dimapur. I have moved the base slot for Moulmein all over the editor, to no avail. Consequently, I have eliminated this base from my obc, and have established a base at Cooktown instead (hoping to bring Rabaul into B-17 strike range).


(Edited for spelling & grammar)

[Aussie]

. In my latest game, the Computer-controlled South Pacific HQ has just loaded an infantry division from San Francisco and is sending it to Darwin. (DARWIN??)


Actually Darwin is a quite useful base, both early and late into the war - worth defending. Still, I woudn't send a full US division there!

BTW its the capital city of the Northern Territory. You may or may not know this, but Darwin experienced 56 Jap air raids during the war, inlcuding a Feb 19 1942 carrier attack from the TF that raided PH under Nagumo.

Dan

[Ranger-75]

The problem that Capt Harlock notes is that the transport TF with NO escort is likely travelling STRAIGHT to Darwin. That's a NO-GO.

The proper way would have the TF's "destination" Samoa or Togga or somwhere like that with a "home port" of Dawrin. That way it skirts around the Japanese held areas. But, the AI is too stupid to do this.

Also proper, necessary, actually, would be to have an ESCORT.