Does Allied AI think outside the box????

[dr.hal]

Greetings folks,

I'm going to start a game against the Allies who will be captained by the AI. However I've got a concern. I don't want the game to be a walk in the part, I want the AI to put it's best foot forward, HOWEVER, I don't believe that the attack on Pearl was a good move and I do NOT want to do it. Yet, I want the AI not to go bananas, so I've got a question for all you folks that know the AI better than me (which is ALL of you!). If I DON'T attack Pearl and hit Manila instead does this "odd" move destroy the AI as an opponent as it is a move outside what it is programed to respond to? If I have to attack Pearl to have a good game (relatively speaking, as a human opponent is much better, but due to health concerns, I can't commit to a game, hence the AI) so be it. But I thought I would ask if the AI seems to be able to handle the Japanese being non-traditional. Thanks in advance. Hal

[GetAssista]

AI will see it as a Pearl strike where all the bombs and torpedoes just missed.
AI thinks in terms of bases held or lost, assets available, enemy units visible, but not individual strikes. As for your broad question - no, AI does not handle non-traditional Japan well if non-traditional includes very deep invasions, deliberate carrier-hunting, CV fishnets across oceans, mass sub positioning on established sealanes. Also try not capture Karachi =)

[rustysi]

Personally my use of the AI is just as a learning tool. I've had it hitting Milne Bay for over a week now and my carriers are well within striking distance. I just let it go as I know it'll lose. The best thing to do would be to jump in as others have said for two/three days a week at times to 'get it right'. My concern is to see if my designs for the Japanese are at least somewhat feasible. My latest attempt has had me concerned a bit, but things have finally 'convalesced'. A while later than thought, but that's fine. TBH its my first full attempt at the R&D situation. Other games it was only partially addressed. I've come to find just how expensive this area of the game is for Japan.

In answer to your question, I'm sorry to say the Allied AI is rather weak. That's not a knock on the game I find all AI's to be weak unless given some advantages. In addition to what I'm saying here I play stock scen1 so that may have something to do with it. I've seen it try to do some good things, but all I have developed counters to with experience. Some of them are standard tactics in the game. Not bad. As I always maintain (and eventually I could be wrong) computers can not truly 'think' abstractly, humans can. At least some of us (I say some because it does not include me).[:D] The game after all is simply following a script.

Now all that being said, I have yet to play into the 'happy years' for the Allies. As I have said it is a learning tool for me, and I've gone to a certain point to achieve this or that and started over. In my current game I will go further than ever and it should be my last AI game. Then if I can find the time I will commit to a PBEM. I fully expect to get my head handed to me, at that point.[:D]

[dr.hal]

Thanks Rusty, so in truth the Allied AI is weaker than the Japanese? Right now I'm playing the Japanese AI and it is loosing so badly that it is really doing some crazy stuff, like trying to make landings without escorts and getting the "sh#t" stomped out of them. It really is no fun. I was hoping that the Allied AI would be smarter, at least in the beginning, as it has few options (as long as I don't think WAY outside the box, which is what I've done to my live opponents). But I could be wrong. I'm disappointed by the lack of response to this thread, as I thought many folks would have views on the subject, but it would appear I'm wrong. I'm also under the impression that the AI as an opponent only works with the stock scenarios, am I right on that? Has the AI been tweaked on any of the non standard scenarios? Inquiring minds want to know! Thanks again, Hal

[SheperdN7]

In ironman games you'll crack a smile with some of the AI's plans and schemes but in stock I would rather play against myself , more of a challenge.

[Aurorus]

ORIGINAL: dr.hal

Thanks Rusty, so in truth the Allied AI is weaker than the Japanese? Right now I'm playing the Japanese AI and it is loosing so badly that it is really doing some crazy stuff, like trying to make landings without escorts and getting the "sh#t" stomped out of them. It really is no fun. I was hoping that the Allied AI would be smarter, at least in the beginning, as it has few options (as long as I don't think WAY outside the box, which is what I've done to my live opponents). But I could be wrong. I'm disappointed by the lack of response to this thread, as I thought many folks would have views on the subject, but it would appear I'm wrong. I'm also under the impression that the AI as an opponent only works with the stock scenarios, am I right on that? Has the AI been tweaked on any of the non standard scenarios? Inquiring minds want to know! Thanks again, Hal

The only real use for the AI is to learn the intricacies of the game: how to use ports efficiently, how to coordinate air strikes, how the game mechanics handle various aspect of the war, and so forth. You will be disappointed with the AI time and again, even on the ironman scenarios, and even if you do nothing to upset it. The auto-TF system produces exhorbitant loses for AI supply fleets, and the AI editor only allows Andymac to assign objectives, HQs, and time frames. The AI will select assets from the various HQs to allocate, which results in disjointed attacks and improper TF creation.

Another major problem with the AI is that it needs to cheat to do various things. For example, it will route its long-range CAP fighters in certain areas after you have made your strike selections, always appearing where you do. You will learn these tendencies and either use them against the AI (by just fighter sweeping it to death) or decide to deliberately allow yourself to suffer for not doing so, which becomes a silly exercise very quickly.

It will also teleport ships to destinations. Some claim that this is not true, but I have seen it do it. I had a Java invasion TF suddenly intercepted by 2 British BBs, which came from nowhere. When I went back to the save from the turn before and loaded up the game as the allies, I found the 2 BBs in India. The only way that they could get from India to the exact hex of my Java invasion in 24 hours is to teleport. So do not let anyone tell you that the AI does not teleport ships. It does.

My advice is.. fine tune your game with the AI, then find a Play by Email opponent.

[Sauvequipeut]

No doubt at all it teleports. Recently I've done a few run throughs of Andy Macs reworked stock scenarios, playing as Allies up to 1st April 1942. In the last one, I switched to the Japanese side after the last turn to see what it was doing - by then the AI had cleared Singapore and was on with the invasion of Java and Sumatra. The game ran through an end of turn first and a whole bunch of cruisers that I'd had in Columbo and Sydney the day before suddenly appeared in the Java Sea.

So there's no doubt it teleports...what I'm curious to find out is if it can teleport between any bases or if it can only teleport between certain designated key bases. The cruisers I mentioned above could have come from either Batavia or Soerbaja.

[Chickenboy]

Dr. Hal,

I don't have a ton of recent experience with the Allied AI, but I wanted to at least contribute what I could to your thread.

I haven't played as the Japanese against the Allied AI ever. From the "early days" of WiTP:AE, I'd heard from many that the Allied AI was poorly rendered and self-defeating. Yes-it was recognized as being 'worse than the Japanese' AI, for which there was more attention lavished on the appropriate scripts. The only way to make either of them remotely entertaining was to provide them with various levels of 'cheating'. This can involve allowing them ahistoric / other worldly supply LOS, allowing them to do away with many iterations of base levels' or base forces' impact, the aforementioned teleportation, etc. In short, the AI cheats like a mother ******.

For myself, the inability to starve out Japanese troops with the AI set at "very hard" was a deal breaker. You just can't do it. That changes basic gameplay and strategy to the point that I'm just not interested in being surprised by yet another novel way of AI cheating that I was unaware of. If we're suspending the laws of physics, basic mathematics, quantum reality, and conservation of matter I guess I should not be astonished when the AI isn't compelled to follow *other* game rules.

I've recently embarked on a game vs. the AI as the Allies. My goal is to beat the Japanese AI into an ALLIED autovictory by January 1943. It's a tall order, even against a lobotomized and cheating AI.

So far, it's working out well. In March 1942, Singapore, Manila, Ambon, Timor, Java and Sumatra still stand. The latter two have been heavily reinforced and are probably unassailable. Kido Butai has already been shattered-a combination of submarine action, surface warfare and Allied carrier application at a weak point. Japanese unescorted amphibious TFs inbound for the Celebes have been chopped to pieces. Injudiciously considered overloaded amphibious TFs have resulted in upwards of 17,000 IJA casualties when the ships are sunk beneath them.

Reinforced Singapore's defenses have withstood several assaults with decreasing Allied casualties. The Japanese effort to piecemeal attack Singapore will likely be its undoing.

In short, Dr. Hal, the AI is a tool. It's a dumb tool. The Allied AI version is the dumber of the two. It cheats. It doesn't think. It is what it is. The only challenge in my mind is to use some jiu-jitsu and use its stupidity against it. Let it thrash itself.

How bad can you help it thrash itself? That's kind of interesting IMO. How can you take advantage of an opportunity to kill bushel baskets of laden Japanese transports with the Allied early war OOB? Where can you develop a potent air umbrella because of the Japanese inability to see through challenges? Can you kill enough of the enemy to deliver a freakishly early Allied victory?

[Alfred]

1.  The AI does not ever teleport task forces.
 
2.  The AI will generally not teleport individual anchored ships but will do so under a specific condition.
 
 
The script will set out certain circumstances under which a specified type of task force will be deployed by the AI.  The task force will be populated from anchored ships in the "trigger" port.  However, if the requisite type of ships are not available in the "trigger" port, the AI will teleport anchored ships to meet the script requirements.
 
 
Alfred 

[dr.hal]

Thanks Aurorus, I appreciate your input. I do understand that the AI is a learning tool. I've been playing war games for a very long time and learned that a while back. However in some games the AI presents a real challenge, in this game however it is exceedingly weak... which may be a reflection of the complexity of the game itself, of the desire by developers to put more thought into the game mechanics at the expense of the AI counter play... I'm simply trying to probe the waters here (to use a nautical metaphor!).

[dr.hal]

Thanks Chickenboy for taking the time to respond, for a while there, due to a lack of responses, I thought I had raised a subject that no one cared about!!!!

So far in my game I've not noted cheating by the AI just simply dumb moves.... but as I've alluded to, I can't give a human my time at the moment so the AI is it. But it's been a very large disappointment. Yes I see the attraction of an Allied auto victory sometime in May of '43 but in truth I was hoping to get some challenges. Yes the AI has caught me once of twice thus far, but only on a relatively minor scale and no where near enough to offset the losses it has incurred. I AM trying to avoid using obvious traps, such as leaving my subs on used tracks as the AI seems to never notice that ships keep on sinking if they pass down a certain path, but beyond that I'm not sure what actions I can take to make the game a better challenge. I guess I'll just have to get use to the idea that electronic Japs are going to go to the bottom of the sea in large numbers for the foreseeable future!!! Thanks again. Hal

[Chickenboy]

I used to feel 'guilty' about braining an otherwise lobotomized AI. I don't anymore. I use the AI's oversights, lame behavior and inexplicable stupidity to really put the screws to it. An autovictory in May 1943? Not my goal. I want it in January 1943. I'd opt for sooner if it was possible.

This difficult goal makes me use all tricks of the trade to achieve. Rather than feel guilty about exploiting my opponent's mistakes, I relish them as I would against a PBEM opponent.

Don't feel guilty about mistakes your opponent makes. Capitalize on them! The fact that it's an AI script instead of a human opponent shouldn't dissuade you from aggressively exploiting that which is given you.

[dr.hal]

Well that's certainly a strong position Chickenboy, I'm sure glad I'm not your opponent!!!! But I'll take your words under advisement... Thanks again. Hal

[GetAssista]

You can mercilessly beat AI by exploiting its known weaknesses in a time gauntlet (I once conquered the world for Japan by autumn 1943 in stock GC on Hard). Or you can pretend AI is better than it is, not provoke it into blunders, use more sound strategies even if you know AI won't punish you like human player will. Or you can try have as historical game as possible by guiding AI when necessary, and have fun writing a story of your personal War in the Pacific.

Everyone has their own goals for the AI game and that's fine, whatever gives you fun. Some modes might make you overly reckless for the PBEM, others might actually help you train your PBEM skills, then others are just fun and nothing else. I 've found that I constantly keep finding and learning new things about how this game plays even if playing AI only. And this learning is also fun in itself

[Sauvequipeut]

For added fun against the AI, try a limited sub or no sub game. Variants include:

1) No subs at all...they head off-map soon as and stay there.
2) No subs until a certain date...after the 4/42 upgrades or whenever you think the AI has had a fair crack of the whip.
3) Only selected subs...Dutch, British, S-Boats and SST's only...maybe a limited number of US Fleet boats for minelaying or variety...your choice, really.
4) No commerce raiding with subs. Subs only used for minelaying, base defence. ASW (seriously) or as fleet scouts.

The Pacific War without subs seems odd, but having to go after the Japanese Merchant Marine with old-fashioned surface commerce raiders is far more fun than endless patrol zone clickety-click. And you do spare your fingers a fair amount of clickety-click.

Haven't tried no-sub/limited sub as Japanese (I've only got as far as playing around with various openings vs Allied AI), but I guess a reasonable limited sub variants would include Glen-carriers only as commerce raiders, others only fleet scout, minelaying or midget sub missions.

[Grfin Zeppelin]

I am playing against the Allied AI in Andys Ironman nasty scenario. I must say I am very content with my game now. The AI wrestled Guadalcanal and Tulagi from me in late 42. It wasnt even funny how much it threw at me to the point I wasnt willing to lose more stuff for this place. Sure I sank alot, much more than against a human player but stil was overwhelemd at the end.
Just dont play into the AIs weaknesses.

[Macclan5]

As posted in another thread in the main forum....

I take the opposite view.

The AI needs no apologies.

The AI is linear... as computers think in linear terms. Function A >> Function B >> Function C

In my opinion : A "fun game" against the AI is not so much grounded "cheats dues to difficulty level" or "re-scripting of its aggressiveness" .... but rather in the restraints of historic / chain of command limitations.

Its possible and fun.

I recognize that others will disagree.

Disclaimer : I have reloaded at least one bad turn (or two... or three) and never apologized to the AI [8D]

[Macclan5]

@ Dr Hal...

I guess however my response invalidates your specific question.. sorry. [8D]

The "historic imperative" to attack Pearl was founded in the basis that the Japanese Empire desired a "early knock out punch" (avoid a war of attrition) verses the USN to prevent it from interfering in its plan of expansion i.e. to gather and conquer access to Oil and resources which had been cut off due to economic sanctions.

I do not suppose I needed to even state that for you... you were more likely very well aware.

My response therefore would be this

Play against the Allied AI but be guided by the requirement to plan and execute your own Kantai Kessen in 1941 or 1942 .... somehow - somewhere.

Ultimately your challenge is to lure sufficient assets into the mix to do so.

Perhaps Midway (reversal?)

[dr.hal]

Thanks Macclan and yes, I know of the rationale as to why it was done, the famous but elusive "Decisive Battle" concept. Given that the Japanese didn't know the CVs were NOT there and given that there was hope of getting them, I can see the attack at Pearl as "doable". However WE know that the CVs are at sea and thus not "gettable" and thus the attack is less then what might have been. I would like to see a mod that allows for the CVs to BE AT PEARL and thus "gettable" from the beginning!!!! But that's a whole other conversation.

But Manila was certainly a valid target and before Yamato's plan, WAS the target. Given the number of subs and support ships there, I think it was still a challenger to the Pearl plan. Here too historical knowledge impacts the play as we KNOW the US's modern torpedoes are "cr#p" and thus the subs less of a threat than was known to the Japanese at the time. But if one thinks in the "long" view of history, the subs DO become a real threat later in the war.... so it's good to take them out NOW!!! But to my point, I don't want to screw up the AI's play too much by using "thinking outside the box"!!!

[Chickenboy]

See, I take the 'other other' side of the coin on this one. It's my *job* to screw up the AI as much as I possibly can to force as lopsided a result as possible in my favor. If I can use my opponent's proclivities and predispositions against 'itself', why wouldn't I do that? Particularly when it cheats like a mother****** in many other ways?

If I can beat a cheating algorithm senseless in short order by providing it enough rope to hang itself with, I'd be remiss in not taking this tact. How many arms do I have to tie behind my back before I feel as though I've beat the AI 'enough' or that it gave me a 'good game'? How much do I have to let it 'cheat' before I can feel vindicated in crushing it (versus like I took candy from a baby)?

The AI is what it is. It's designed to be crushed without mercy. In my opinion, NOT doing this is not playing to the capabilities of the engine or oneself.