My oppinion of what the AI is good for...

[Haohmaru]

I can think of only one word to describe Combat Mission's AI: embarassing. I'm not talking about only QBs but the scenarios included on the cd as well. I know some of you will be very offended by this, but just listen to these examples.

People claim that the ai is good on defense. Currently I've been playing the Operation 21st Army Counterattacks as the Russians. The German defenders have nice large amounts of wooded areas that they can fortify themselves in, yet the tactic that they choose to "defend" is to sally forth from these defenses, bum rush my forces and get slaughtered in the process. Some squads will manage to crawl (or limp) back to the woods, only to emerge a couple of turns later. Wash, rinse, repeat. Now to add to the embarassment, several of the computer controlled squads have managed to stumble into their own landmines, even before being targetted by my troops! Therefore they are not doing this out of panic but rather sheer stupidity. They blow themselves up and manage to reveal the locations of these mines in the process, very generous of the ai. I thought it was impossible to plot movement inside mines you know exist anyway? The computer still chooses to do this. I've seen incredible placements of fortifications also, eg. wooden bunkers with their rear facing the enemy's side of the map, sometimes they will be placed right on the edge of a river so they are facing downwards and don't have LOS to anything.

Now lets move to attack, one of the cleverest things I've seen the computer do is allow the rear of their AFVs to face my side of the map, this is particularly suicidal with CMBB's slow turn rates. If you are still not convinced about the lousy ai, try playing a game against it with FOW turned off. You will see that consistantly a large number of squads are out of command, and it has a curious habit of making arty spotters move side to side, back and forth, for no good reason in particular.

People claim that scenarios can be tailor made to "assist" the ai. I cannot comment on that, but I have to say that I've played a large number of scenarios on cd and they have produced consistant results. The very first time I played CMBO (aside from the demo) I got a minor victory, I was a complete newbie at this time. Surely the scenarios originally designed and shipped with CMBO and CMBB were designed with the ai in mind.

Now, to sum up, let me say that I love Combat Mission. It is unequalled for multiplayer enjoyment and I have enjoyed playing it for 3 years now. But I am disappointed that CMBB in my opinion shows no improvement in the computer ai department. The truth is I still enjoy the occasional game against the ai, I give it a minimum 25% increase and +2 exp. bonus. Sometimes it will still do well because tanks are so much more effective with a higher experience level. Overall though I find it disatisfying to have to give the ai a bonus, even if you "lose" you will usually still win on points because you've blasted away so many of its troops.

Again let me finish by saying I have nothing against CM and think it's a great game. I just think that people who try to insist on how good the ai is need to open their eyes. And I apologise to the starter of this thread who just wanted to explain how best to use the ai, this thread has been somewhat hijacked already so I thought I would throw in my comment.

Haoh

[Sirocco]

The basic mistake that people make when thinking about AI, is making what are natural assumptions in their thought process. That's a difficult hurdle to overcome.

[undead reindeer cavalry]

writing a good AI is damn hard. it's not the abstact ideas or rules of thumb that are hard, it's the technical application of the basics that is so hard.

one should start with something simple like coding the pathfinding for a single unit. the basic stuff is pretty easy, a standard A* algorithm should do it. you get an AI that considers both speed and distance. your tank will now have the comprehension to find the fastest route to the target, avoiding driving thru scattered woods and such. but once you start adding things like LOS and known enemy positions, you get screwed big time. add to that cover, other friendly units, issues related to firepower & armor, mission objectives etc etc, you get really really screwed. coding it is HARD.

about game AI for those interested:

http://www.gamedev.net/reference/list.asp?categoryid=18

[Jack Carr]

The AI makes a good attacker when the odds are numerically in it's favor. Bump up the AI enemy troops and equipment and have at it. I especially like to play German defender QB's against a numerically superior Soviet AI attacker giving the Russians a +50% edge. Lots of fun and often times not as easy as you would think. I think I'll try a bigger bonus soon and see how zany it gets.

[MikeyD]

I fear lurkers who read this thread before actually playing the game may get the wrong impression.

The AI is not 'broken'. The AI is not 'awful'. Admittedly, after someone's played a gazillion hours on the game you get to know the AI's tricks and quirks pretty well. But how awful can the AI be if people merrily play a gazillion hours against it?

Considering that you're pitting your whits against an appliance about the size of a toaster, it does a pretty darned good job for itself. Especially on first-runs through proper scenarios, and especially during the first four months(!) that you've owned the game.

[Battlefront.com]

JasonC,

Horsefeathers.

Funny, I thought that would be your educated response

It is not processing or program size (modern desktops can run a Mars mission, easily), it is just time spent programming it up, debugging, and refining routines. It is the input time that takes work, not the number crunching.

And are you speaking from experience, or an orafice in your rear area? AI is horribly time consuming, for sure, and that is indeed one of the big problems with it. CM's AI could be a lot better if we did a code freeze on the features and then spent 4-6 months writing nothing but AI. On that count we are in agreement. But you horribly underestimate the problems associated with handling a complex world like CM's on a modern day computer. That Mars mission example you tossed out is a strawman argument since I am sure it in no way shape or form is even remotely comparable with what CM does. A better example would be the computing power that it takes to beat a world class chess champion. Even though Chess is an exponentially simplier system to code AI for, it took decades and hundreds of millions of dollars (and a LOT more than a desktop computer) to stand a chance against the best human player.

It is no surprise to me that some people think that Charles can toss in Deep Blue like challenge in the first place, not to mention having it run on your computer. Oh, and to have it delivered in a reasonable amount of time and for $45 or less. Get real.

CMx2's AI will benefit from what was learned during the development of CMBO/BB for sure. We expect it will be quite a bit better. But a decent player will likely still be able to kick the snot out of it most of the time. And until someone can show me an example of an AI that does better than CM's, in a similarly diverse and chaotic environment, to ask for more is to be unreasonable at best. But to insist that it is easy, without any personal experience or examples to point to, is simply laughable.

Haohmaru

Again let me finish by saying I have nothing against CM and think it's a great game. I just think that people who try to insist on how good the ai is need to open their eyes.

It is you that needs to open his eyes. People who say CM's AI is good are comparing it to other wargames, past and present. They are not comparing it to a Human as you have. Why? Because that is an impossible standard to compare it to and therefore makes such an argument completely irrelevant. It is like me expecting all our customers to be reasonable and rational. We'll be able to write a Human level AI long before that happens

Steve

[Oddball_E8]

holy bananas batman! what have i started?

i only wanted to give my view on how to use the AI to its best advantage, but now it seems to have de-evolved into some kind of "i hate the AI" rant...

well ill have no part in it any longer...

i hereby cordiously resign any responsability on the future evolution of this post.

there... now you cant pin this debacle on me...

[Battlefront.com]

MikeyD,

I fear lurkers who read this thread before actually playing the game may get the wrong impression.

The AI is not 'broken'. The AI is not 'awful'. Admittedly, after someone's played a gazillion hours on the game you get to know the AI's tricks and quirks pretty well. But how awful can the AI be if people merrily play a gazillion hours against it?

Now now Mikey... this thread was going down a perfectly horrible line of irrational and unreasonable bunk... and you have to come in and try and inject sense and reality. Shame on you

Considering that you're pitting your whits against an appliance about the size of a toaster, it does a pretty darned good job for itself. Especially on first-runs through proper scenarios, and especially during the first four months(!) that you've owned the game.

Now you're just rubbing salt into the wound, aren't you?

Seriously though... anybody that wishes to forcefully push a belief that AI, any AI, can even get close to satisfying a Human player as if he is playing a Human should exit this and any future discussion of AI. The reason? Because I for one don't expect it to happen in my lifetime as a game maker, so I'd rather not have to keep hearing the same unrealistic nonsense for the rest of my tenure.

Can CM's AI be better? Sure. Anything can be improved. But can it be improved to the extent that people will be satisfied with it enough to stop whining and complaining? No. Kinda puts us in a tough spot, especially when we have only a limited amount of time to make a game. Is our time better spent making a slightly better, but still woefully inadequately Human, AI at the expense of the game itself? No.

Reality dicates that we can only do so much in given period of time, so best not to squander large amounts of time for a small gain that will only be complained about by the same people complaining before. Oh, and they will also complain about how long the game took to make and any other flaws perceived. Those who whine the loadest are the ones most ignored because to pay attention to them would mean disaster.

Oddball... don't worry, your intial post was a useful one. Not your fault the thread went down a deadend.

Steve

[ October 27, 2003, 11:36 PM: Message edited by: Battlefront.com ]

[JasonC]

"CM's AI could be a lot better if we did a code freeze on the features and then spent 4-6 months writing nothing but AI."

CM's AI would be better if any significant improvements were made to it, ever. If continual "builds" of the AI engine were a priority equal to researching new vehicles to include, or new art. If getting AI arty use right were considered a matter akin to getting MGs right (as CMBB did), or including dust (as CMAK does). That it isn't has everything to do with programmers' time and nothing to do with computer processing speed or memory.

"you horribly underestimate the problems associated with handling a complex world like CM's on a modern day computer."

Check my job description in my profile sometime.

"A better example would be the computing power that it takes to beat a world class chess champion."

No, that is a strawman again. You only need to give a passable performance against a club player. The environment for a chess program is more easily axiomatized, but the standard one is held to is also immeasurably higher. Not for a "Blue", but for an off the shelf shareware chess program. There are a number of chess engines that fit in less than 250K and will still kick your tail - freeware. It'll play faster than you will on a 386. Let alone the current generation shareware on pentiums.

The problem with something like a CM AI is not to give optimal play. It is simply to give tactically coherent play of any kind. Fog of war blindness, odds, and ever present "friction" will do the rest, and result in challenges - where it wouldn't in chess because those elements are absent.

For people first learning the game, for cases where you must attack against uber armor, or with ordinary ammo limits against 3:1 odds, those factors - and not the AIs coherence - can result in challenges given the present state of the CM AI. But that state is indeed greatly in need of improvement.

"it took decades and hundreds of millions of dollars (and a LOT more than a desktop computer) to stand a chance against the best human player."

Best has nothing to do with it. And it does not take program size or processing speed or execution time, for the above mentioned small chess programs to beat anyone but masters (while playing at about the expert level themselves). What it *did* take was a lot of iteration on the programming side. Instead of whining and moaning about how unappreciated BTS is for what results from its lack of that time, why don't you think about how to get it?

I am not suggesting hiring 4 full time AI tweakers and giving them 6 months to do nothing but. (Although, if the army paid for it with the idea of using the result for training, that'd definitely be worth doing). Instead, tap into the thriving resource you've got here in this body of grumblers.

Instead of coding one way for the AI to do something and spinning the complaints that result, program 10 and put the knobs in the hands of players. In 6 months a thousand control freaks will tweak the thing in ways you couldn't imagine beforehand. Let the AI grow like Mods, and you will get one that can be compared to chess programs.

That is how chess programs got good. Not processing speed, not memory bloat, not millions of dollars, but enthusiatic amateurs all over. They played their programs against each other and watched what worked. The results were then used as a base for showing projects by some large companies out for publicity, sure. But that is not what first got chess programs to master level.

"some people think that Charles can toss in Deep Blue like challenge in the first place, not to mention having it run on your computer."

Who said anything about Charles? All you guys would have to do is put the weightings for a set of nodes into refiner's hands. OK, you'd have to code the thing with a certain top down flexibility so it could be adjusted. That was a good idea in the first place, just for internal revisions. If you did, great. If not, some time, between some version and another, it is worth doing. What you want is an AI on an improving track without further input from you all. Well, that requires some one-off work to make a flexible AI engine, but is perfectly feasible.

"CMx2's AI will benefit from what was learned during the development of CMBO/BB for sure. We expect it will be quite a bit better. But a decent player will likely still be able to kick the snot out of it most of the time."

If the win is righteous, even that is just fine. What annoys is the completely imbecile, the win that shouldn't have been, the loss of the challenge. Human players like winning. They aren't going to gripe about being able to beat the AI. They do gripe, and gripe rightly, about being able to get away with ridiculous tactics against the AI and having them work without fail. Especially if it is a similar bag of tricks to those that worked in "A Chance Encounter".

[JasonC]

So how would this work? Just as you have a scenario editor now, you'd have an AI editor. You'd put one in a directory, like overwriting default mods today. Few people design massive operations. But plenty of people try their hand at a few scenarios or play small testbeds. And thousands download and use scenarios designed by several different teams of independent builders who do it for the love of it. If you provided an AI editor there would probably be somewhat fewer people who would use it. But if significantly improved AIs resulted, people would use them. A few would even play AIs against each other to refine them.

[General Brock]

A great ol wargame that had excellent A.I was Battles of Napoleon. The A.I. was quite aggressive but one of the great options was the ability to choose the A.I. style. You could have it random or attack-aggressive, or cautiously etc. Sid Meier's Gettysburg had something similar too. These kind of things give play vs computer a whole new lease on life.

[Dschugaschwili]

JasonC, you're not being fair here. The fact alone that CM is not a game with full information (unlike chess) makes programming a CM AI so much harder. I have a master degree in computer science, and I do think that I'm a good programmer, but they would have to pay me lots of money to write an AI for a game as complex as CM with an audience as demanding as CM's.

Mimicking human behaviour is very hard to do for a computer. Chess programs have taken a long time until they could handle concepts like positional advantage. CM's AI was written by a single guy just a few years ago, and I honestly don't think that you can expect it to hold its own against the best human players on this planet, nor should you assune that there are easy improvements that nobody has thought of before (especially since AI threads pop up in fairly regular intervals).

Dschugaschwili

[eichenbaum]

A few years ago I designed some scenario's for the game Force 21 (http://www.redstorm.com/games/force21/).

One big problem was that I needed to program every move and every action for every unit of the AI. This programming was done with some C++ scripting, calling functions with their parameters. For every action done by the player I needed to write a reaction or else the units were only sitting ducks. Every time I wrote some trigger event there was the testing, a very time consuming job when the editor needs to compile the scripts first before implementing them in the game. Try to re-do some AI actions where the player must at least play 15 minutes before reaching that particular trigger event. Writing scripts for a medium scenario took a week (when working on it in the evenings). Writing AI that does respond like a human does sound almost impossible to me.

Combat Mission on the other hand, does not need any scripting. Just place the units with their parameters and you'll have an `instant battle'. This method will spare a lot of time for the scenario designers and even people without any programming experience can create a level for a great game. OK, the AI does not always do the things that it should but when you're playing against the computer that does not make any mistakes it doesn't sound very human to me...

It would help a bit if the scenario designer did had the possibility to set out the waypoints for units like you can in the 1st turn. Together with cover arcs and arty pre-planning, the designer can force the AI a bit to stick to a strategy in the 1st turns. But extremely necessary ? No!

BigTime did a very good job on this one.

[MikeyD]

I remember a BFC post (a looooong time ago) mentioning refinements they were hoping to make to the AI in the next big upgrade. It sounded very clever and very promising, AND they avoided 'scripting' like the plague.

One thought I just had -- A lot of those refinements (including different 'invisible' flag types to influence movements, I recall?) sounded great for scenario designers. I wonder how well they could be implimented for auto-generated QuickBattles, though.

Awh, considering the amount of hard info I have on the subject I could just as well be speculating about how many angels can dance on the head of a pin.

[ October 28, 2003, 10:53 AM: Message edited by: MikeyD ]

[Haohmaru]

Perhaps I was a bit harsh in my wording, and upon reflection the computer ai never seemed bad until I started playing against human opponents. I have to thank them for teaching me how to play much better than I had been, but it did have the unfortunate side effect of making ai opponents boring. But the fact remains I was quite disappointed with the ai of CMBB, because despite being completely green and having had no practise with the new infantry modelling, orders etc I still found it easy to beat both on defense and attack right from the first battle, even with a handicap. The ai was promised to be improved, but personally I can not see any improvement over CMBO other than it being able to use the new commands, which don't really help you when you use poor tactics. I don't know anything about ai coding and so I can accept battlefront's stance that it requires a lot of manpower. But isn't anyone else baffled that the computer can not keep it's units in C & C, at least in the initial setup?? And why does it place a bunker facing the wrong way? The scenario clearly defines which side of the map is hostile, surely it cannot be difficult for the ai to point its units in the right direction?

People keep asking to come up with an example of better ai, well how about Medieval: TW? In this game the ai manages to keep its army in well coordinated formations, it probes weaknesses in your defense, tries to flank you with cavalry, tries to occupy the high ground to give its units a bonus. It prepares ambushes within forests, it brings on reinforcements at opportune moments, it tries to crush vulnerable but dangerous units such as archers. It even has unique tactics reflecting different nationalities, for example the muslims will often feign a general retreat, forcing your heavily armored soldiers to charge after them in the hot desert. Then when your forces are tired and scattered they will rally and suddenly strike back hard. The only foolish behaviour I've ever seen is it will sometimes put it's commanding unit into a dangerous situation, giving you a chance to kill it and route the whole army. Now, some people will point out that this setting is easier to simulate, since soldiers will just line up in box formations and kill each other face to face, whereas WW2 is a whole different thing. But it's still a strategy game, and real time too so the computer doesn't even have "time" to think.

As Jason C said, things such as the mg modelling have been improved and fixed. I clearly remember back in the days of CMBO, if anyone complained about machines guns they would be derided by others and there would be explanations such as "the units are not running straight, they are dodging and zigzagging etc" or "it's not just open ground, there are dips and rises in the ground to provide cover." Now we have CMBB and we see a much better depiction of machine guns, people realise it's safe to make fun of CMBO "the land where mgs do not kill." I really hope that CM2 or some future title fixes the ai, then I believe everyone will breath a sigh of relief and laugh about how hopeless the CM ai was before. They can vent their previous frustration without being seen to be saying something negative about battlefront's products.

One final note, the best performance I've had against the ai ia a major (or total, can't remember) victory in the Arnhem- Red Devils Operation in CMBO, twice, once as the allies, the second time axis. On both occasions the ai's assistance? +150% forces were given to it! It was particularly easy as the British, once you've cleared the town and set up along the bridge it is wholesale slaughter, the only concern is piat ammo starts to run dry for the massive amount of armor that pours through. When it turns into a shooting gallery it is not particularly fun...

[JasonC]

"JasonC, you're not being fair here"

Not trying to be. I could care less about fair. I am trying to improve the AI, and thereby these guys' product. They can thank me for it or they can throw spitballs, I don't care a lick. I don't expect them to care whether I thank them or throw spitballs either; if they do that is their problem. The only relevant question is, does it help them get us a better game?

"that CM is not a game with full information makes programming a CM AI so much harder."

Harder to axiomatize, sure. So does more locations from a finer grid. But the standard is also lower. If you move one knight to the wrong square even once in chess you've lost the game. Double blind is much more forgiving.

If you manage to orient the majority of your force in basically the right direction without committing a small set of definite exploitable errors, you will provide a challenge. You may still lose, but that is OK.

"Mimicking human behaviour"

Is completely irrelevant. Nobody needs human behavior here. We just need the AI to send its people here sometimes, there sometimes. Some randomness from the enemy without full information is already challenging.

"Chess programs have taken a long time until they could handle concepts like positional advantage."

Fine, so let's get started instead of sitting around talking about how long it took somebody else. You'll never get there if you don't put the AI on a progressive path. A progressive path requires tapping user time and energy, by an AI editor with controls in their hands. It does not require any straw man perfection. That means something general enough that editor users are manipulating weights given to various things. That isn't trivial to write, but something like it is needed internally anyway.

Notice, I am *not* talking about scripting. I am *not* talking about building a plan into each scenario. I am not even talking about player set SOPs, though that is closer. Instead what is wanted is something like "importance of x". Closer to flags, sighted enemies, staying in command, cover, spreading out, own morale state, fatigue level, etc.

Instead of telling the tac AI to always pop smoke and back up when facing an AFV that can kill you, write 5 routines - shoot first, smoke and back up, continue order, etc. Then have weights for each. It does a random one according to the weights. Put the weights on a control panel in an AI editor. If a player-designer thinks smoke and back up is always right, he put it to 1 and the others to 0. If he thinks that is too predictable but usually correct, he puts it at .75 and continue order at .25.

"CM's AI was written by a single guy just a few years ago"

So let's change that and get lots of people working on it instead of just one. They are right here.

"don't think that you can expect it to hold its own against the best human players"

Who is talking about the best players? I am talking about anybody but a newbie. Who is talking about holding its own? I am talking about giving a satisfying challenge even if it generally loses, instead of losing to unrealistic tactics that exploit the same holes it has had for years. If holes remain, so long as they change and the player doesn't know what they are that would still be an improvement.

"nor should you assune that there are easy improvements that nobody has thought of"

Of course there are easy improvements. I've mentioned dozens of them. Whether others have thought of them is a different question - many probably have. But they can't do anything about them. Only a few people inside BTS can. And their time is limited. So the value-added thing they can do, is use their time in a way that taps into other people's time. Because it is programming time, testing time, that is the scarce input that keeps the AI weak. And that state of affairs can be changed.

Look, if Charles had to write every scenario, do you think this game would be as successful as it is? If there were no scenario depot, no disks full of Rune creations, no Stalingrad packs - and no Quick Battle generator - how much replay do you think CM would get? By providing an editor they leveraged Charles and company's programming time a thousand fold.

[Battlefront.com]

Not trying to be. I could care less about fair.

Then please go somewhere else to be an ignorant pain in the arse. Whoops... gee, perhaps I am not being fair? I could care less about being fair.

I am trying to improve the AI, and thereby these guys' product.

I want car makers to produce a 4x4 truck that gets 100 mpg, is safer than an armored truck in a collision, and can stop in 10 feet while driving 65mph. I can come up with all sorts of fanciful suggestions about how this can be done. But since I don't know what the Hell I am talking about, they don't mean a damned in any way shape or form.

As usual... the response to your posts is directly proportional to your ego's presentation. Condescending posts like you are well know for dictate the "spitballs" tossed back at you. If you were humble, even if illinformed (like Haohmaru), you would not garner such negative attention. As you have so many times in the past.

They can thank me for it or they can throw spitballs, I don't care a lick. I don't expect them to care whether I thank them or throw spitballs either; if they do that is their problem. The only relevant question is, does it help them get us a better game?

Correct. And your naive, uninformed, uneducated, insulting rantings are not in the least bit helpful. You really think your grand ideas are news to us? Ideas are a dime a dozen, and all of your ideas have been covered before in one shape or form. But see... there is a thing called reality. If you have never experienced it first hand, as you have certainly not in this context, there is no possible way you can understand how clueless you are.

Is completely irrelevant. Nobody needs human behavior here. We just need the AI to send its people here sometimes, there sometimes. Some randomness from the enemy without full information is already challenging.

Yeah, its that simple. In fact, that is exactly what CM does now. It just doesn't do it with the type of intelligence you say it doesn't need (i.e. Human level intelligence).

A progressive path requires tapping user time and energy, by an AI editor with controls in their hands. It does not require any straw man perfection.

No, it requires a development effort that is probably greater than the sum of the entire rest of the game. Again, you don't have the slightest clue what you are talking about when you suggest some sort of end user window into the AI's mind. I know... you'll come up with all sorts of baseless, uninformed reasons detailing how exactly we can do this, but I can come up with as many informed reasons about why GMC is simply lazy in getting me my super truck.

Notice, I am *not* talking about scripting. I am *not* talking about building a plan into each scenario. I am not even talking about player set SOPs, though that is closer. Instead what is wanted is something like "importance of x". Closer to flags, sighted enemies, staying in command, cover, spreading out, own morale state, fatigue level, etc.

Great in theory, but fails in practice like all your other ideas. Take "staying in command" for example. In order for this to work as you envision we would first have to make the AI keep units in perfect formation (i.e. in command) under "combat" conditions and any other conditions for that matter. THAT is the tough part. Having the weights be available to the player, in at least rough form, is a piece of piss to do. It's just that there is NO sense to it if the AI is inherently incapable of doing it in the first place. Currently CM's TacAI does a decent job in some situations some of the time, but has a lot of problems in others during other times. We could spend a month just trying to get the AI to modestly improve its performance. And there is absolutely nothing the end user could do to hurry this along.

Instead of telling the tac AI to always pop smoke and back up when facing an AFV that can kill you, write 5 routines - shoot first, smoke and back up, continue order, etc. Then have weights for each. It does a random one according to the weights. Put the weights on a control panel in an AI editor.

You picked an example of exaclty how the AI works right now.

If a player-designer thinks smoke and back up is always right, he put it to 1 and the others to 0. If he thinks that is too predictable but usually correct, he puts it at .75 and continue order at .25.

Such tweaks by the player are useless since the use of the behavior can only be determined "correct" or "incorrect" based on the exact situation it is displayed in. To get the desired behavior you need to first code up all the behaviors to perform flawlessly (or there abouts). In simple laymans terms, the AI can never be any better than it is programmed to be. Variable tweaking is not programming, it is simply variable tweaking. It won't make a hill of beans difference unless the variable defaults are carelessly set. And I assure you that every variable has been carefully tweaked over the last 5 years to get the most optimal results from the code and its successive improvements.

Like "staying in command" the real work is defining HOW to use the behaviors under a very specific circumstance. The weights assigned to the behaviors could be done by a monkey with bad case of hiccups. There is no magic to be found there, and that means a major effort to let people tweak something that is basically irrelevant

So let's change that and get lots of people working on it instead of just one. They are right here.

First of all, to open up any part of the game to user tweaks would involve a lot of time and effort. And for what? So a few egomaniacs can argue about how much better a "1.5554" is here instead of another guy's "1.673"? Hardly seems to be a good use of our time when the end result will be, for the most part, identical and not in the least bit more challenging in any meaningful sense of the word.

Who is talking about the best players? I am talking about anybody but a newbie. Who is talking about holding its own? I am talking about giving a satisfying challenge even if it generally loses, instead of losing to unrealistic tactics that exploit the same holes it has had for years.

Realistic Tactics and beahavior changes only come as a result of painfully long and involved coding efforts, not from tweaking a number up or down. Which gets me right back to the main point... you are demanding something that is not practical.

If holes remain, so long as they change and the player doesn't know what they are that would still be an improvement.

Improvements come at a development cost. Money doesn't grow on trees, and neither does our time. We can do only so much and to demand more in the insulting and childish tone as you have displayed here (and in many other threads I might add) is completely counter productive.

Of course there are easy improvements. I've mentioned dozens of them.

Hehe... I am going to make this my sig line

Steve

[Battlefront.com]

Haohmaru

I really hope that CM2 or some future title fixes the ai, then I believe everyone will breath a sigh of relief and laugh about how hopeless the CM ai was before.

Never going to happen If we spent 10 years on the AI alone, and required 10 linked computers to process the turns, I can assure you that Jason and others would find a reason to complain. It is one of the Laws of Wargaming... if someone can complain, he will. In fact, it is the First Law of Wargame Development ("Build it and they shall complain").

On a slightly more serious note (because I was being serious above), there is a quantum difference between changing the behavior of something like MGs and getting the AI to be overall better. It is like fixing the hinge on a door to stop squeaking compared to making the door vanish into thin air when a button is pushed. To you, the gamer, this might not appear to be the case. But that is completely irrelevant (and beyond Jason's ability to understand, apparently) since reality does not bend to a consumer's incorrect assessment of how coding works.

And BTW, the "tweak" to MGs required a very fundamental overhaul which probably took a month of programming and several months of testing. What looks easy to you is an illusion that is carefully crafted by us. We pride ourselves on making the complicated look easy, for that is the hallmark of a good game developer. But we can't make the impossible look easy since we can't do it in the first place.

The simple fact is that if we could put in 2-4 months and improve CM's AI we would. But 2-4 months would put our company at risk (we need money to exist, and time is money) and only result in a marginally better overall AI experience. Sure, it would probably kick butt some times where now it fails, but it would still do stupid things frequently and fail to present a significant challenge to a decent player quite often. From a development standpoint, this is a very poor investment to make (i.e. it has poor bang for the buck potential).

This is reality. Argue all you want, but reality doesn't change. Even we can't change reality.

Steve

[ October 29, 2003, 12:18 AM: Message edited by: Battlefront.com ]

[Michael Dorosh]

[ October 30, 2003, 08:05 PM: Message edited by: Michael Dorosh ]

[Battlefront.com]

MikeyD,

I remember a BFC post (a looooong time ago) mentioning refinements they were hoping to make to the AI in the next big upgrade. It sounded very clever and very promising, AND they avoided 'scripting' like the plague.

Good memory! We are planning on including a host of user implemented stuff on a scenario by scenario basis that will not only help direct the AI but also to help shape the scenario as well. For example, defining the sceanrio to be a "hold at all costs" one vs. "take this specific hill and do so only with minimal losses". This is, in a way, a form of scripting. But since it will be simply a guideline, in most cases, in practice it isn't if done right. We intend on doing it right!

One thought I just had -- A lot of those refinements (including different 'invisible' flag types to influence movements, I recall?) sounded great for scenario designers. I wonder how well they could be implimented for auto-generated QuickBattles, though.

This is always a problem with what we call "trail of crumbs" or "hidden flags" approaches. They really do require Human placement to be effective. But to not implement them means user created scenarios (which we should all agree are the more interesting to play) won't have the benefit of such systems. It is our philosophy for CMx2 to give the scenario designer such tools and implement them for Quick Battles in whatever practical manner possible.

Awh, considering the amount of hard info I have on the subject I could just as well be speculating about how many angels can dance on the head of a pin.

Well, I'm sure Jason can give you an answer for the age old angel question if you ask him real nice

Steve

[General Brock]

Jason C is absolutely right about some of the simple ideas for improving A.I. (not just in CM but in games in general) Along the lines of my Battles of Napoleon post earlier, good A.I. in wargames does'nt mean HAL9000. It usually means some creativity. bit of randomness, and choices of style of A.I that keeps the game fresh.

[General Brock]

Steve Wrote

As usual... the response to your posts is directly proportional to your ego's presentation. Condescending posts like you are well know for dictate the "spitballs" tossed back at you. If you were humble, even if illinformed (like Haohmaru), you would not garner such negative attention. As you have so many times in the past.

Steve, I have to say that this post sounds much more like your own personality than Jason's. With all due respect to you and Battlefront, I don't belive humility is a concept high on your list of priorities.

[Wol]

Perhaps the administrators (as the administrators) ought not be so touchy...

Regardless it would be useful to compile the methods of getting the best out of the AI. This is something that BTS could include in their manuals (especially for CMAK). At least the most important: If the AI is attacking, omit man -portable support weapons they will try to bayonet you with that Light Mortar. - give them SP weapons. Never more than one flag in an operation. etc etc.

[General Brock]

I'm as guilty as the next guy for getting into arguments on here when a heated topic is going. I think CM since it is a finished product would be pretty hard to radically change it's A.I. with mods in the way graphics are. Frankly, I never had a problem with it's A.I. to begin with. I've seen much worse (John Tiller's games come to mind offhand) On a different Battlefront game, Strategic Command to my mind seems to have phenomnal A.I. Is it cheating??? If it is'nt, I have to salute the designer on a great job.

[Oddball_E8]

FIST FIGHT! FIST FIGHT! FIST FIGHT!

oh and if anyone wants to know...

the number of angels that can balance on the head of a pin is 3.

just so you know

now SHUT IT!

the AI is how it is and thats not going to change before CMX and if it "sucks" then you may complain...

but right now there is no use in complaining...

it just aint gonna change people!!!