Huge AI concentration/fixation problem

[Berneroy]

I’m playing the campaign scenario which starts at may 1942 as allied vs. japanese AI (historical difficulty).

After a good fight japanese AI took Port Moresby after winning carrier superiority. No problems with AI this far, it concetrated naval forces very well.

After the loss of PM AI invaded my weak spot, Etafe (2 hexes from Luganville and 7 hexes from Noumea) with very strong forces (majority of CVs, BBs etc.) and without strong carrier group I was not able to fight back and I lost the Etafe base.

Here is the problem: In August 1942 AI tries to turn Etafe to major naval base (like Truk). It has stuffed the medium base (port 4, airfield 4) full of ships, i switched sides and checked the situation: In Etafe port (not in TFs) there was nearly 150 ships, including 6 CVs and CVLs, 5 BBs, 10 CAs and huge number of transports and other ships. Air cover is weak, I can bomb the port with 100+ level bombers (Fortresses, Marauders, Mitchells) from Noumea, few turns of bombing and there is no japanese carrier fleet anymore.

That game was of course ruined. I hope this will be noticed in future patches, IF land based bombers in range AND port size < x AND air cover < y THEN do not disband fleet or something…

Anyway, I like this game a lot. Thank you for everyone involved!

[kev_uk]

Id like to see the AI just have a few more routines in it for checks such as these; for instance in my current Japanese campaign game, the AI decided to dock some valuable British CAs in Singapore, right under the noses of my Nells. Consequently set them onto port attack, which disabled the CAs for many months, and in all probability they will still be docked there when I take Singapore.

I suggest when you get used to the game, do PBEM. It is much more fun playing against a human player, the only drawback being the small amount of turns per day.

[vonSchnitter]

Hi !

have seen the same happen in Scen. 4: A huge number of ships reported in Lae.

As far as I can make out, the AI decides to put damaged ships into the next friendly port, preferring those with any kind of garrison - put ships will put into "empty" ports as well.

Would be interesting to know what level of damage sets this routine into motion. Most likely any floatation damage in the "orange" or low "red".

If the player keeps going after them, of course damage will increase, up to 99% system damage and high to very high floatation damage. This way - as things are programmed - the AI prevents loosing the ships on the high sees right away.

On top of the ships beeing harder to hit at anchor, floatation damage may not be converted to system damage -which may be the main reason for the AI to anchor damaged vessels a.s.a.p. Dunno about the repair ratio/probability of ships at anchor versus "docked".

Cheers
vonSchnitter

[kev_uk]

But a check should be made if designated port is under LBA range (and how heavy it is). Also, the AI should be able to remove ships from port if that base is about to be overrun.

[Berneroy]

It is logical to dock/disband damaged ships to nearest port (depending on damage levels), but 95% of those 150 ships in Etafe were undamaged before my first bombing raid... And AI had not tried to escape from certain death.

My conclusion is that in some situations (like this one) it could happen that AI docks capital ships ignoring (more or less) the local air superiority situation.

Yes, PBEM is the way to go... I had my best ever gaming experience with UV PBEM. I think i have to wait for a patch or two before starting a WITP full campaign PBEM.

[nico71]

Berneroy, I assume this is the fictional MO-scenario? I'm playing it at the moment as US, and the Japanese AI (historical) seems to have an affinity for Lae, placing lot's of valuable targets there. It's easy to kill them with the heaps of aircraft that I have at PM!

I have read an interesting article in Game Star, where one of the game designers of Rome - Total War talks about the design of the AI in the game. He says that it is based on Sun-Tzu and that they basically did convert half of his phrases straight into AI-routines. Let's see in September how good it really works! [;)]

BTW, if you like we can play a PBEM of this scenario while we wait for the patch! I'm a noob, but yet I hope that I am the better AI! [:D]

[Berneroy]

ORIGINAL: nico71

Berneroy, I assume this is the fictional MO-scenario? I'm playing it at the moment as US, and the Japanese AI (historical) seems to have an affinity for Lae, placing lot's of valuable targets there. It's easy to kill them with the heaps of aircraft that I have at PM!

I have read an interesting article in Game Star, where one of the game designers of Rome - Total War talks about the design of the AI in the game. He says that it is based on Sun-Tzu and that they basically did convert half of his phrases straight into AI-routines. Let's see in September how good it really works! [;)]

BTW, if you like we can play a PBEM of this scenario while we wait for the patch! I'm a noob, but yet I hope that I am the better AI! [:D]

No, its the historical long campaing starting in may. I saw some consentration in Lae too, but because I lost PM, I couldn't do much about it... In my game there is small concentration of cruises etc. in japanese occupied PM as well, without any air cover... AI uses wrong bases to store its valuable ships...

Sun Tzu stuff sounds great! Maybe they should use Macchiavelli for diplomacy routines...

Small PBEM game before the great one sound great, but I'm without internet connection for few months... Home in one city and work in other...[:@]

[nico71]

ORIGINAL: Berneroy

Sun Tzu stuff sounds great! Maybe they should use Macchiavelli for diplomacy routines...

"The Prince"? Will get it later and see if it is useful. I had recently read Sun Tzu, and I would say that 90% of the stuff in that book is useful for wargames in general and WITP in particular. Both for players and for coders!

At the moment I read Clausewitz. Still very philosophical stuff, but also useful as far as I can tell for now.

The problem is how all of this can be translated into C++ or whatever they use here for WITP! I don't think that I have seen any huge leaps in development of AI over the past 15-20 years! At least AI-scripts are somewhat rare by now in most genres. It is all about graphics and sound today. But instead of AI development the game developers rely on multiplayer! I mean all of them in almost all genres, not just here. That's a shame! My own programming skills are virtually non-existant, otherwise I would give it a try!

Small PBEM game before the great one sound great, but I'm without internet connection for few months... Home in one city and work in other...[:@]

That's too bad!
[:(]

[ZOOMIE1980]

ORIGINAL: nico71

ORIGINAL: Berneroy

Sun Tzu stuff sounds great! Maybe they should use Macchiavelli for diplomacy routines...

"The Prince"? Will get it later and see if it is useful. I had recently read Sun Tzu, and I would say that 90% of the stuff in that book is useful for wargames in general and WITP in particular. Both for players and for coders!

At the moment I read Clausewitz. Still very philosophical stuff, but also useful as far as I can tell for now.

The problem is how all of this can be translated into C++ or whatever they use here for WITP! I don't think that I have seen any huge leaps in development of AI over the past 15-20 years! At least AI-scripts are somewhat rare by now in most genres. It is all about graphics and sound today. But instead of AI development the game developers rely on multiplayer! I mean all of them in almost all genres, not just here. That's a shame! My own programming skills are virtually non-existant, otherwise I would give it a try!

Small PBEM game before the great one sound great, but I'm without internet connection for few months... Home in one city and work in other...[:@]

That's too bad!
[:(]

Well AI programming has been a major pet peve of mine for some time. I have also been quoted on this forum as stating the same thing, the art of AI programming is no better today than 20 years ago. It hasn't really advanced much at all. The reason for that is economic. It needs to be only "playable" for wargames, as the assumption is most will play the AI until proficient, then move on to PBEM games. Problem is, I strongly feel that is a FALSE assumption. Just because a lot of long time posters here may be PBEM types, most of the purchasers are decidedly NOT and will play the AI until the game becomes a bore an from then on, the game collects dust. But at least with this game, it will take a while to get to that point.

Have not played long enough to make a judgement about this AI, yet. It's at a part of the game that requires a heavy degree of scripting, and at a point I have not been able to counter the AI very well (playing the allies right now in the main campaign scenario). I'll have a better feel for the AI once I start to oppose it strongly in the Solomons. We'll see then how this one reacts.

[Cmdrcain]

The Original Pacific war by Grigsby also had the problem of the AI
filling up a port it gets fixed on with ships.

Often was the time I'd hit a Port in that game and Boom Take it over and poof went like 250 mcs

[ZOOMIE1980]

AI's, at the most basic level, come it two varieties, the "chess" AI, where the computer strives to evaluate and weight every possible move the other side could make, every turn, and the "scripted" AI, used mostly in computer gaming, especially wargames. In games like WitP they usually have a lot of subroutines on the order of "Take Rabaul from Truk", or "Take Singapore/Malaysia from Saigon/IndoChina", or "Defend Lunga" etc.... And those scripts are fairly rigid. More sophitisticated AI's in more involved games probably deploy a series of scripts of scripts. They usually employ at least some "sanity checks" and/or have various "success threasholds" to evaluate whether or not to abandon the script at some point. The more checks coded into each script and more alternate scripts it can branch into based on various checks, the better. But at some time, that begins to evolve/devolve into the chess game AI and slows down the turn resolution.

In WitP, there are already a lot of complaints concerning the time it takes for turns to resolve. The more checking, the more AI script branching, and so on, the longer the turn resolutions will be. It's a tradeoff. If they think the majority of buyers have PIV 3.0Ghz, 1GB RAM machines, you can code a lot more "intelligence" into an AI, but if they feel most have less than half that power, it restricts what they can do and still be "playable" to the majority.

I am playing on an ancient PIII 650Mhz machine, and turns take about 12-15 minutes to resolve with the messaging most shut off using ctrl-z. And for me, that's long enough......

[nico71]

And this game is turn based without any time constraints! I wonder how bad things are in sims, where the PC has to keep up enough horsepower to run the game with 30+ fps while it has to deal with the AI of 16+ opponents!

The less scripted variants I was refering to above are usually in RTS and RPG (to a degree) games where the field of play is limited and decisions don't go too far into the future (tactical level). BTW, I wonder how the Combat Mission AI works!

OTOH I can't imagine that the AI of a game of the sort and size of WITP will ever be capable to develop complicated attacks or defenses on its own! In 'Rome' things are different. Even though the number of units looks impressive, they are usually bound into huge formations. Those formations are limited in number and size (I think 20 or so formations at any one time). The strategy level is even less demanding because it's turn based. In this way they can afford to use a chess-like AI.

But that doesn't change the fact that the WITP-AI is weaker than it could be! Maybe it's not the number of scripts, but the way it is scripted. I mean, a check whether a unit moves into the range of an enemy unit would certainly not bog the system down!

[The Gnome]

I'd take better AI and sacrifice speed any day, sorry Zoomie... but if the results are bad what am I rushing to see anyway? The AI needs to evaluate it's aircover better, plain and simple.

As to those suggesting "just PBEM", many of us don't like to. I just want to play at my own pace and maybe replay a turn a couple of times. PBEM is not an option for me, sorry. I knopw this group has a very strong PBEM culture but again this game has AI for those of us that don't want to do it. That being said I hope we could maybe sticky this thread for AI improvements so they don't get lost. I don't want to see the AI treated like the red headed step child.

[ZOOMIE1980]

ORIGINAL: nico71

And this game is turn based without any time constraints! I wonder how bad things are in sims, where the PC has to keep up enough horsepower to run the game with 30+ fps while it has to deal with the AI of 16+ opponents!

The less scripted variants I was refering to above are usually in RTS and RPG (to a degree) games where the field of play is limited and decisions don't go too far into the future (tactical level). BTW, I wonder how the Combat Mission AI works!

OTOH I can't imagine that the AI of a game of the sort and size of WITP will ever be capable to develop complicated attacks or defenses on its own! In 'Rome' things are different. Even though the number of units looks impressive, they are usually bound into huge formations. Those formations are limited in number and size (I think 20 or so formations at any one time). The strategy level is even less demanding because it's turn based. In this way they can afford to use a chess-like AI.

But that doesn't change the fact that the WITP-AI is weaker than it could be! Maybe it's not the number of scripts, but the way it is scripted. I mean, a check whether a unit moves into the range of an enemy unit would certainly not bog the system down!

The next big leap, IMHO, for AI development in turn based gaming, will come from using the disk and an RDBMS. Programmers have always coded everything to cached up into memory because disk access has always been so slow, and there has never been a good, fast FREE RDBMS around. You can't expect people to buy and Oracle or SQL Server license everytime they buy a game!

But nowadays with lots of very good, OpenSource, very FAST RDBMS, and extremely fast disk access with Ultra DMA 133 and better out there, it begins to open up the possiblity of using an SQL database to make some massive enhancements to AI's. Add to that, using much more object oriented design, means lot less stored in memory and a lot more stored on disk without a great deal of slow-down. And with games even like WitP eating up 200MB+ of memory, you getting a lot of disk paging anyway so in reality, the programmers are accessing the disk a lot whether or not they intended to!

[The Gnome]

The next big leap, IMHO, for AI development in turn based gaming, will come from using the disk and an RDBMS. Programmers have always coded everything to cached up into memory because disk access has always been so slow, and there has never been a good, fast FREE RDBMS around. You can't expect people to buy and Oracle or SQL Server license everytime they buy a game!

But nowadays with lots of very good, OpenSource, very FAST RDBMS, and extremely fast disk access with Ultra DMA 133 and better out there, it begins to open up the possiblity of using an SQL database to make some massive enhancements to AI's. Add to that, using much more object oriented design, means lot less stored in memory and a lot more stored on disk without a great deal of slow-down. And with games even like WitP eating up 200MB+ of memory, you getting a lot of disk paging anyway so in reality, the programmers are accessing the disk a lot whether or not they intended to!

Yeap! I can't wait, there are a lot of fun possibilites and I always think to myself "wow they could use an RDBMS for this". Something like WiTP is an ideal candidate for a database app. Just thinking of the reports alone you could write gives me goosebumps! :>

[siRkid]

ORIGINAL: Berneroy

I’m playing the campaign scenario which starts at may 1942 as allied vs. japanese AI (historical difficulty).

After a good fight japanese AI took Port Moresby after winning carrier superiority. No problems with AI this far, it concetrated naval forces very well.

After the loss of PM AI invaded my weak spot, Etafe (2 hexes from Luganville and 7 hexes from Noumea) with very strong forces (majority of CVs, BBs etc.) and without strong carrier group I was not able to fight back and I lost the Etafe base.

Here is the problem: In August 1942 AI tries to turn Etafe to major naval base (like Truk). It has stuffed the medium base (port 4, airfield 4) full of ships, i switched sides and checked the situation: In Etafe port (not in TFs) there was nearly 150 ships, including 6 CVs and CVLs, 5 BBs, 10 CAs and huge number of transports and other ships. Air cover is weak, I can bomb the port with 100+ level bombers (Fortresses, Marauders, Mitchells) from Noumea, few turns of bombing and there is no japanese carrier fleet anymore.

That game was of course ruined. I hope this will be noticed in future patches, IF land based bombers in range AND port size < x AND air cover < y THEN do not disband fleet or something…

Anyway, I like this game a lot. Thank you for everyone involved!

If you have a save before you bombed the port, send it to me along with a description of the problem and step-by-step instructions on how to view it. I get a lot of e-mails and without the info in the e-mail I will not have the time to track it down.

sirkid@cfl.rr.com

[kaleun]

I'd take better AI and sacrifice speed any day, sorry Zoomie... but if the results are bad what am I rushing to see anyway? The AI needs to evaluate it's aircover better, plain and simple.

As to those suggesting "just PBEM", many of us don't like to. I just want to play at my own pace and maybe replay a turn a couple of times. PBEM is not an option for me, sorry. I knopw this group has a very strong PBEM culture but again this game has AI for those of us that don't want to do it. That being said I hope we could maybe sticky this thread for AI improvements so they don't get lost. I don't want to see the AI treated like the red headed step child.

I will have to agree with all of this. Yes the PBEM is the best way to do it. No AI (to date) will equal the resourcefulness and imagination of a human player; but yes I do want a good AI that can give me a game; and the answer is not to let the AI cheat; that is just a cheating way to bypass the problem. Maybe WITP 21st century edition, in a few years time.[8|]

[Scott_USN]

I hope they do not assume everyone will play PBEM... I do not care about PBEM and it is boring and frustrating.

[ZOOMIE1980]

ORIGINAL: Scott_USN

I hope they do not assume everyone will play PBEM... I do not care about PBEM and it is boring and frustrating.

Unfortunately, I think they do, to a degree. Almost ALL the beta testers are PBEM guys and it appears even many of the Matrix admin staff that play this game are! No surprise, that is where the focus is. And I do believe that taking the tact that most players will eventually be PBEM players is completely misguided....

I know that once I have the AI figured out in any game, it hits the dust bin. No amount of scenario editting or setting a higher level, can make up for a marginal AI. The saving grace about WitP is that it is so big, by the time I have the AI on each side figured out, I'll be tired of it anyway....

[The Gnome]

ORIGINAL: ZOOMIE1980

ORIGINAL: Scott_USN

I hope they do not assume everyone will play PBEM... I do not care about PBEM and it is boring and frustrating.

Unfortunately, I think they do, to a degree. Almost ALL the beta testers are PBEM guys and it appears even many of the Matrix admin staff that play this game are! No surprise, that is where the focus is. And I do believe that taking the tact that most players will eventually be PBEM players is completely misguided....

I know that once I have the AI figured out in any game, it hits the dust bin. No amount of scenario editting or setting a higher level, can make up for a marginal AI. The saving grace about WitP is that it is so big, by the time I have the AI on each side figured out, I'll be tired of it anyway....

Agree with this completely, sadly enough. WiTP could be something I'd play on and off for a decade if we see some AI enhancements. I agree too that right now the holes in the AI haven't exposed themselves in my game yet, but as soon as my airforces reconstitute it's only a matter of time.

The only criticism I have for Matrix is their assumption that we'll just love PBEM if we give it a chance. Please don't take that as a slam against PBEM, I'm glad so many people are having a great time with it - but it's not something I enjoy.