Getting the best out of the AI

[Guest Thumper]

Lewis,

Don't like the AI? Simple don't play the game. Just as soon as you do better then maybe people will take you seriously.

[Guest -panzershreck-]

hehehe, this is hilarious!!!

[KiwiJoe]

Close Combat AI is horrid. I quit playing the series once #4 came out because the tank AI was the worst yet, it couldn't attack to save itself. I remember at 1 point I had no AT units and just 3 allied rifle squads left. The AI had 5 panthers and several infantry squads and would NOT attack. After 40 mins I got bored sitting on my victory locations and went and had dinner. I forgot about the game and went out with friends. Didn't come home till 3am. Guess wot - the AI still hadn't attacked! It was there for over 7 hours with 2 tanks doing a tango in the middle of the map and the rest just sitting in the rear.

I promptly uninstalled the game and went to bed.

[WendellM]

<BLOCKQUOTE>quote:</font><HR>the AI still hadn't attacked! It was there for over 7 hours with 2 tanks doing a tango in the middle of the map and the rest just sitting in the rear.<HR></BLOCKQUOTE>

Those must have been the superior PzKw Planierraupe (bulldozer) type the Germans used from time to time.

[Guest MantaRay]

A chess program is really unfair to compare CM to, and to call it artificial inteligence is in a word, wrong. Computer Chess programs are just a series of statistics in which someone has given values and trends to calculate the probabilites for chances at success. More like the calculations CM uses for ballistics.

But Lewis, if you have so many issues with this game, stop playing it and, well go play Chess, as I know 20 chess programs that will kick your arse everytime.

Sadly, you do not seem to realize that games are supposed to be an escape from the realities of everyday life. You seem to expect CM to be as real as the cup of coffee you drink in the morning, or the food you eat at dinner. Not going to ever happen, no matter how many years or technological advances mankind makes.

Is Cm the end all of games on war? No, but if you really step back and take an objective look at it with an open mind...nothing has ever come along that has come anywhere close. The only other game that even has come close to being a great wargame is TOAW, but if your really look at it objectively, CM is better all around.

Ray

[Demangel]

I think the programmers have done an admirable job... I agree the ai could be better, in both human vs computer and human vs human. But I have so far only experienced a handful of things I was po'ed about. Most of which I realized could actually be attributed to some malfunction of my own mind when I planned the manuever.

(damn Panzerschrek team damn sherman... Why did you run at the house when the sherman was sitting there looking someplace else... why when you had a nice shot at it;'s side... why when you where not under fire... why when your morall was high and your experience good... why oh why did you run into the house only to be slaughtered.)

Anyway the game is great, and I'm sure will have sequals and improvement/patches.

It beats trying to get a few friends together who have enough peices let alone nice train scenery to play a nice game of table top wargaming. (at least one oponent and one referie) And even the best game of table top doesn't have that real life animation. But what the two share is this. ever sit down to a table top game and win or lose you find yourself focussing on a particular thing... I mean like say your losing and you have to plot some brilliant move that would require channeling patton himself... And you do your best and when the dice are rolled or the go button is hit you ignore the whole battle just to watch this one thing unfold?!

GREAT JOB BF!

[Guest Big Time Software]

Demangle said:

<BLOCKQUOTE>quote:</font><HR>Anyway the game is great, and I'm sure will have sequals and improvement/patches. <HR></BLOCKQUOTE>

Correct. As we have clearly said, CM's AI is not perfect and thereforet by definition that means there is room for improvement. And improvement there shall be, even if we have to raise the bar on wargame AI for ourselves to rise above We give no promises that the next thing we release will somehow beat the pants off of Guderian if he were still alive, but each new version of CM will bring with it improvements for sure.

Steve

[Sgt.Morgue]

username , i'll play you. i can return several turns a day , it can be almost like head to head. You pick the scenario , i prefer the americans , but will play either side. Consider yourself challanged!

[Lee]

--------------------------------------------

All your complaints are of little value unless you can point to

another game that has a PC-based AI that's any better. And no, Lewis,

CC's comes up laughably short, tank dance and all... Nor does bulldozer

driving AI. LOL! I can see it now! "Bulldozer Death Match 3000"

endorsed by Lewis himself as "the best AI I've seen since, well,

since my 8051 microcontroller days!" LOL!

Charles

--------------------------------------------

ROFL !!!!

That may be the funniest thing I have ever read on this board!

[von Lucke]

Did someone get up on the cranky side of the futon today? Hard day at the office trying to compile code for that new back-hoe controller? Does somebody need a hug?

Here, try this:

http://www.gameai.com/ai.html

Way more funny than most of this thread (if you don't think too much about the whole ferret thing, that is...).

[mikeadams]

I really don't understand all the complaining: if you don't like the game, then don't play it.

[CrapGame]

Lusername: = Troll?

And that's all I have to say about that.

CrapGame out

[Stef]

I'm playing blind PBEM as the germans everyday and I'm astonished by the AI in the individual unit decisions. Ok not perfect AI, makes some terrible actions with some terrible consequences...but as human intelligence would do in reality in the middle of a battle...in 3 word: "a realistic AI !"

2 examples:

1) I've knocked out one of my opponent Sherman 105 with a schreck during an ambush.But the schreck died in the next second due to the US infantry. My opponent complained against the AI which made the other Sherman 105 (who was following the killed sherman) reverse in a slope instead of sticking to the orders "FAST" to take position near the dead schreck in order to support the ambushed US troops (as the schreck was dead , so no danger anymore). I understand his anger 'caus the alive 105 would have avoided him a lot of casualities in his troops if it had sticked to the "FAST" order...but it's the vision from somebody who knows all the game, all the action unfolding there,...in reality the alive 105 chief who sees the tank in his front being killed without seeing the schreck (he probably even doesn't know it's a schreck who has fired ), who doesn't know what's happening uphill, who doesn't know if there's not an other AT weapon there and who cares for his own life would have done the same action: "REVERSE". A very realistic action from the tank AI who don't have a global vision of all the battle and who didn't know the schreck was dead (unless he has some extra lucid skills) . On the other hand if the sherman would have gone on with the "FAST" order it could have been realistic too from a mad/careless tank chief...

It's really what nice in CM AI, the both reactions of the AI are very human even if one of them doesn't stick to the player decisions ! A good mix of all wise and bad decision as a human being would have!

2)On the road uphill the US side some Shermans stuck in the woods and in each other.'caus of this I've killed one of them with my 75mm pillbox. Anger again from my opponent about the pathfinding system. But again very realistic situation. He thought the tanks could go through that path but they stucked in the woods. Realistic again . How much time hasn't it occured in real war that tankist thought they could drive through this place and were wrong, hence stucking themself and the following tanks and bringging on them the enemy fire ??

Very realistic !

People anger to much about CM AI instead of looking at their own unwise decisions. To well understand the CM AI realism, people who have a global view of the game, should take the place of one troop/tank/vehicle to really understand why they react "this and that way" and that they take decisions first for their own security before thinking about supporting the other troops or fanatically sticking to the orders.

Long live to CM AI and BTS,

Stephan

PS: sorry for my bad english.

[This message has been edited by Stef (edited 05-17-2000).]

[gaffertape]

Anger... Rising...

It never ceases to amaze me how some people seem to get a thrill from aggravating other people.

Interesting isn't it how many of us want to leap to the defence of BTS. You lads sure inspire some loyalty.

Although reading these vaguely 'flamey' threads irks me, I do try and use the information to my advantage: By reading them I get a pretty good idea of who's future opinions to respect and who's to ignore. Who to consider a PBEM with, and who not to.

Thanks for the info, Username.

And congrats BTS on the game so many of us have been waiting for. You have legions more fans than detractors.

GAFF

[Black Sabot]

Don't feed the Troll !!!

[jgdpzr]

Lewis,

Just as on the Panzer Elite board, I see that you have made lots of friends here <vbg>. What is sad, is that I agree with some of your points, but your bedside manner leaves much to be desired.

I believe the AI in CM is solid. Certainly better than I have seen in any other tactical game. About the only game on a similar scale is the CC series, and well, we all know that is no comparison (that horse is a skeleton now, so I won't go there). I agree that the AI in Ardennes Offensive is about as dangerous as I have seen, but it doesn't surprise me that SSG cheats. Also, the dynamics of that game are far simpler than in a tactical game like CM.

My only nits with the AI in CM so far is its propensity to use anti-tank teams, mortars and FO's in an unrealistic manner when attacking. I don't know squat about programming, so I also don't know what is reasonable, but I don't like to see the AI sending these units forward just like they are rifle teams. No insult to BTS intended, but if these units performed their support roles more realistically, the AI would be quite dangerous as it in general does a decent job of coordinating units, even when on the attack.

Lewis, I also would like to second some of the others' thoughts here regarding assessing the AI. Namely, it is really only fair to judge it the first time a scenario is played AND you must look at more than just a couple of scenarios. As in real warfare, luck plays a big part. For example, when I played VoT the first time (as US) I got a major victory mostly just because my avenues of attack happened to sheild me from much of the AI's firepower. Had it deployed differently, my plan may not have been nearly as effective. So play 10 or 15 fresh scenarios before passing judgment. But in the end, don't expect to get the same test you would against a quality human opponent.

Finally, Lewis, do you have to be so rude and condescending to the good folks at BTS? Wings, and my fellow Panzer Elite players, also got a fair share of your venom. I will say this, at least you are consistent.

[Spook]

By & large, I have to agree with many people here, and with BTS, in that generalized criticisms of the AI don’t provide any good feedback to BTS.

But now I have to state this opinion emphatically to the AI “boosters” AND to BTS: telling me that the AI is better than that for other tactical wargames is not really telling me enough either. Those of you who’ve related actual details on WHAT you like/dislike about the AI behavior are being a bit more helpful.

So, BTS, telling me that your AI is better than for Close Combat (CC) isn’t that big an issue for me than you would think. Yeah, I used to play CC, and yeah, I know full well about the CC AI “tank dance.” But I don’t play the CC series anymore (no surprise? ), and I don’t plan to get back into SP, not even SP:WaW. It’s your game that I’ve preordered, and it’s your game that I hope to be satisfied with. Citing the performance of other games will lose weight with me over time.

Now, to further clear the air with you, BTS, I will also state here that I can’t “ask for the moon” either and expect human-like AI behavior from CM. It’s just as you’ve said---there are far too many intangibles to tactical combat that involve a large variety of troop types & weapons. It’s not easily broken down into “rock, scissors, paper.” For example, If I were to have five Panthers against five AI Shermans, is it expected that the AI should be programmed to have those Shermans run away for the hills? Or to have the AI assess that it should seek out my flanks? No easy answers, of course.

Certainly, it is with PBEM that I expect to get the most satisfaction with CM. The open question to me is whether or not the AI is “good enough” (not “perfect”) to give me a satisfactory game. We should also recognize that some gamers out there would rather play only against the AI, for whatever reasons, and they will stick with CM if they think the AI is “satisfactory.”

Now some of my recent observations on the strategic & tactical AI’s for CM.

1) At the present, I think the strategic AI is performing well, but for the Gold Demo, I’ve only allowed the AI to operate as defender so far. When defending objectives, the AI troops don’t “flitter around” or shift often and will wait to see where my attack directions are. The AI in “free setups” also will frequently seek defensive cover for initial infantry deployments, so to maximize ambush chances. And when I approached the main town in the VoT game and the AI could see where my forces had concentrated (after the smoke had lifted), it shifted several infantry units over to face me in response, rather than to sit “frozen” to its initial positions.

2) The tactical AI is pretty effective at times, but I’ve seen some questionable behavior too. On one “plus” occasion in Chance Encounter (with the AI as Germans), I watched how the AI tried to maneuver its remaining Stug along the north ridge’s treeline for a flank shot after I had knocked out the other two Stug’s that moved out in the open. Nice “adaptability” there. But as I’ve pointed out in the other recent “tactical A/I” thread, I’m a bit bothered at the AI’s present habit to frequently drop smoke in its defending role. It will use HE barrages too (and it DOES try to use them on concentrated troops!), but I openly wonder now if the AI should be tweaked to limit its usage of smoke, at least in a “defending” role. Another thing is that far too many times, the tactical AI seems willing to have troops fire & fire at spotted enemy troops that are beyond “effective range.” I don’t have an easy answer to this one, but perhaps the “game options” might have an AI “fire discipline” switch to limit (but not prohibit) the AI’s “range of engagement” for its various troops & weapons?

Or perhaps instead, the spotting routines could be adjusted so that “covered troops” (not in open ground) are a TAD less easy to spot at longer ranges? Somehow, I have this general impression that I’m spotting covered troops easier than I should be, but that’s all it is now----an impression.

On an overall view, am I satisfied with the AI? It’s too early for me to have an informed judgement, but I DO think that the CM prospects are good so far. “Tweaking”, though, might still be needed.

[Jeff Pattison]

I agree with everything that has been said in defense of the AI. It is the best AI in a computerized wargame that I have seen on the market to date(not counting Chess). But, it is not the best AI I have seen in a wargame, and I wouldn't be a grognard if I didn't say so. I think some of the solitare board games like Tokyo Express, Carrier War, Raid on St. Nazaire, and SASL have been a lot more challenging. With exception of SASL, these were game systems that were heavily simplified to make them AI friendly. SASL on the other hand required the player to do a little fudging with the oponent's pieces on the Tactical Level.

Where this game's AI continues to have major problems is at the Strategic Level on the attack and when left to its own devices on defense. On attack, it sometimes still attacks on a broad front which makes its chances of success slim against an opponent with the same order of battle. On defense, it fails to guard the most likely axes of approach adequately. I recognize from discussions about the AI on previous threads teaching the computer about things like "key terrain", "lines of departure", and "assembly areas" are nearly impossible, because they rely so heavily on human judgement. There is a partial solution that could be implemented in future modification. IMHO, that solution is incorporating a detailed scripting language in the scenario editor which is pretty typical these days.

That scripting language would allow the scenario designer to store a plan of attack or defense at the platoon level including initial setup, lines of march, and intermediate objectives with supporting fires. Several of these frag plans could be stored with each scenario with the computer selecting randomly the plan to use. Also, the language could have commands that could be used tailor the AI behavior for say the agressiveness of the employment of supporting arms(air, arty, armor), the willingness to stray from the plan, and engagement range, etc.

Of course, another solution would to be use that portion of the SASL rules that cover the movement to contact portion of a scenario and

rely on the current computer AI for the rest of it. That would make this game a real bitch. But, that has numerous technical and legal hurdles.

As far as using a "brute force" chess type engine on CM, just think about it. To play at the grandmaster level in the middle game, it takes 15-20 minutes for 600Mhz PIII to search 15 million positions with simple rules on 8x8 board with 64 squares. Just think about 2km x 2km board with 50m squares. That's 1600 squares. That's 2 orders of magnitude, 100 times, more complex than chess. Add a third dimension to the board, 1 order of magnitude mininmum. 1000 times.

Add complex rules, another order of magnitude minimum. 10000 times. So, if my calculations are correct which I doubt, I would be waiting for about 2-2/3 years between moves for the computer to think about it which also slightly longer than the rate at which todays computers become obsolete.

Also, if you consider that it took 50 years for the computer to beat the best grandmaster at chess and that computer had about 700 years of chess experience programmed into it, I think BTS is doing remarkably well for the 4 years that this game has been in development.

Jeff Pattison

[Spook]

Now THAT'S the kind of AI assessments I can appreciate, Jeff. I suspect that it is as you say----that the AI will be more challenged in the attack than on the defense, due to the fact that the concept of a "combined arms schwerpunkt" isn't easily grasped yet by computers. (This isn't saying that the CM AI HASN'T ever applied combined arms before, but I don't keep high expectations yet.) I also like the "chess comments" that help to provide perspective as to the difficulty it it would take to create a human-like AI for CM.

Actually, the CM AI does a reasonable job, IMO, as a defender. It often seeks cover to deploy the footsloggers & guns, and will try to keep the objective sites defended. The unknown to me is how well can the AI do for a "defense in depth" with multiple defense lines, or to know when to fall back to a new single defense line.

[Guest tom w]

Sgt.Morgue :

I salute your bravery

Clearly challenging such a wise and devious opponent as that Famous tacitician whose name has been mentioned in the same breath as Rommel's must make you a very brave and daring opponent.

I hope Lewis will accept your challenge as it seems so many others here are so intimitated by his reputation against the CM AI that almost none dare go head to head with this military genius via the PBEM.

I do hope the challenge will be accepted and the AAR's will flow forth so all the rest of us mere mortals can learn from such a great tactician as the now infamous Lewis as he uses all his cyber-military might to carve your digital units into mere shadows of their former selves!

(Tongue Firmly in cheek, for those who haven't guessed already!)

Oh please....

Someone play Lewis PBEM just to keep him busy so he won't keeping posting more of his "opinions" here.

-tom w

Mr. Peng you can Pile On any time ...

<BLOCKQUOTE>quote:</font><HR>Originally posted by Sgt.Morgue:

username , i'll play you. i can return several turns a day , it can be almost like head to head. You pick the scenario , i prefer the americans , but will play either side. Consider yourself challanged! <HR></BLOCKQUOTE>

[Moon]

Spook - you said:

"Now, to further clear the air with you, BTS, I will also state here that I can’t “ask for the moon” either..."

Well, actually...

[Guest Big Time Software]

I'll just restate this clearly (again) since I think the two or three times we said it in this thread might have been missed

CM's AI is the best out there for any war or strategy game. Yet it is certainly not perfect, and certainly can be improved. And it certainly will over time. No ifs ands or buts about it.

I would have made a specific statement about Moldy Oldies that ran on 48k systems if I thought we would be compared to them Simple fact is that the old games were very, very simplistic compared to anything done in the last 10 years even. It is like saying we have the fastest car and someone saying that a Model T in its day was the fastest around They really can not be compared to at all. Two entirely different things that just happen to share a common heritage.

As for AI scripting... I can tell you it isn't going to happen. The code support needed for this is a HUGE task. It could easily consume 4-6 months of solid programming to make it actually have a hope of working. This is doable in theory only, as we have to make a living and spending such huge blocks of time on the system is not going to do it for us. Plus, there is an old train of thought about AI programming efforts. It goes something like this...

So long as the AI doesn't barf, people will still buy the game anyway. And no matter what you do, if the AI falls short of a really good human in terms of skill (and it will for a long time to come), people will always find plenty of things to complain about. So why risk going out of business if it doesn't increase sales and isn't going to "satisfy" the people complaining? If you have ever wondered why game AIs on the whole (all genres) stink, this is your answer.

And this is very true. Don't believe me? Look at the progress of the AI in something like Close Combat. Look at its sales. Do you honestly think that NOT improving core AI problems hurt their sales for the first 3-4 games? I don't. And this is not to single out CC as I can think of a few other long runing series that can say the same thing.

So in the end we can only do so much, even if in theory we could in fact do more. Theories have a way of falling short when reality asks for games to be shipped and bills to be paid But we are planning on spending additional time on improving and tweaking the AI as we go along. We already have our own list of things that should be improved, and so far nothing anybody has pointed out is news to us

Steve

[This message has been edited by Big Time Software (edited 05-17-2000).]

[Guest Ol' Blood & Guts]

So this is what that "other" thread was about. This is the first time I have read this thread.

Yep, Lewis has surely outdone himself on this one. Almost reminds me of the time SS_PanzerLeader and I went round and round on the "Auto-Surrender???" thread.

Well, I won't get involved much with this argument, except just one little remark.

To me, to get a real sense of the AI, you have to play a scenario many times, just like we did in the Beta Demo, Because it seemed that I never played the same game twice. It was especially apparent in LD. The Americans never deployed the same and I never knew where the bastards were. And THAT is a major deal with the AI. The fact that it deploys different every time always keep you on your toes, if you're attacking, that is. And even if your defending, you also really never know where the brunt of the attack is going to come from. Hell, I keep setting up minefields and ambush sites and the Americans haven't yet once attcked through them.

Well, I have a few other points along the same lines, but I won't bore you all any longer. So I'm done here.

------------------

"Why don't we say that we took this one chance, and fought!"

"Stupid humans. Hahahahahahaha!"

--from the film Battlefield Earth

[Miles Teg]

From a team of 2 and for $50, the AI is mighty good. [Anyone want my copy of Electronic Art's Fleet Command?]

Further, BTS responsiveness to customers seems unparallalled. A few examples include: fast move command for schrecks, rifle grenades, smoke graphics, etc.

Gawd, what other game company would do this? The majority seem more preoccupied meeting compressed horizon due dates and fulfilling ROI goals (of their publishers), with little regard to customer desires or AI.

There is no doubt in my mind to the continued improvement of CM/AI. [Although, they could just say the hell with it, and goodness forbid, alternatively, develop ERP software or bulldozer logic.]

[Guest Big Time Software]

Thanks! Having enough knowledge already about ERP (if you don't know, DO NOT ASK FOR THE LOVE OF GOD ) to know that fearing us going and developing that stuff can be laid to rest. I'd rather be burried up to my neck and have someone drive over my head with my 5,000 pound Weasel I had enough problems working with corporate gaming development to know that ERP development is NOT for me (even if it pays a lot better!).

Steve

P.S. ERP = Enterprise Resource Planning. Big money in it for a few companies, but some appear to have problems with the "Planning" bit of their software

P.P.S. My wife reports on ERP for a living, so that is my excuse for knowing this trivia.