On scripted enemy's AIs and the nature of war

[Chelco]

Good morning gentlemen,
I am a big fan of war games. Maybe is because I am a frustrated soldier who got to be a scientist (I chose to be a molecular biologist) or just because war games offer an intellectual challenge that surpasses chess, I don't know. I recall that like 10 years ago, as a graduate student I was frequently teased about my military admiration by my fellow students. "Military men do not need their brain because their spine is enough to march", was a common joke among them. I really despised these pseudo-pacifists guys, who at the very least ignored the sacrifices every soldier goes through, even during peace times, just to protect the "freedom of speech" they abuse. At that time, that was just all I could thought about the topic.
Time passed and I came to the wonderful USA to perfect my scientist side. Better wages (love this country which gave me so many opportunities I didn't have before) gave me access to a decent computer and (the most desirable thing for a scientist) LOTS of books. To my complete delight, there is a military science and there are scientists among the soldiers. Dupuy, Leonhard, Boyd, Lind... woah! How much I loved reading their books. I feel overwhelmed in joy every time I fire up a war game and think that there is a theory being built to explain what happens in the battlefield.
War games is and will be my only experience with military stuff. I don't have any chances of being a military man. I feel really honored when I can engage in chats or threads in which professionals speaks their mind about war games. Is really unique to talk about a simulated war with guys who know the real war from experience.
I bought ATF the last year. I felt overwhelmed with its complexity, which I loved. I am a terrible tactician and I never could get a grip on any scenario. But that's not my complaint today. The point on which I want to have feedback on is on the enemy AI in ATF.
Let me begin quoting Clausewitz:
"The essential difference is that war is not an exercise of the will directed at inanimate matter, as is the case with the mechanical arts, or at matter which is animate but passive and yielding, as is the case with the human mind and emotions in the fine arts. In war, the will is directed at an animate object that reacts. (Clausewitz, On War, Book One, page 139)."
Even when some of Clausewitz's points are being challenged today, I think this one is still totally valid. Indeed, is this point what makes war unpredictable and extremely fluid.
In ATF, the enemy AI is scripted. The scenario designer pre-loads a combat plan to the AI side and prepares it for a series of problems the AI could face. In this way, the enemy AI has the ability to react to your inputs but in a limited way. If the scenario designer didn't care about a particular situation, the AI would stay there rather immobile, just hoping to defeat your pressure and continue with it's pre-loaded plan. In this way, the enemy AI is more like one of those old electric football games in which you give each player a path and hope the path will be not perturbed because if so, there is not an alternative option (like an evasive maneuver). Maybe I am wrong, but if I am right, the ATF game could be ignoring a fundamental principle of war.
I am totally aware that non-scripted AIs are not easy to create, that there is no agreement on how to do create them, and that after a huge effort they could end doing completely stupid things.
I really will appreciate your feedback about this point.

[kbluck]

I think the "scripting" is not entirely unrealistic with regard to the higher-level command. Indeed, large-unit commanders typically develop a fairly rigid, rather scripted operation. Not infrequently, if the real situation fails to match the planning assumptions or the enemy behaves in an unexpected fashion, the wheels fall off and the whole operation goes to hell. Just like in ATF. Now, theoretically they *shouldn't* make that sort of plan, but in practice they usually do.

Where ATF could use some work, in my opinion, is in the "small-unit" scale. The individual units don't show a lot of initiative, nor do they show a lot of human emotions. In real-life, higher commanders have only a tenuous influence on what their subordinates are going to do. Sometimes they exceed expectations; sometimes they do stupid things that might have seemed obvious at the local scale but screw up the bigger plan. Sometimes they simply freeze ineffectively, or run away altogether.

In ATF, when unexpected situations occur, units mostly just follow orders blindly and either plow ahead to their deaths or sit foolishly waiting for input. They ought to do *something* when they're being shot at, even if that "something" isn't what you might have hoped they would do. SOPs are a start at this sort of semi-autonomy, but in their present state are rather blunt tools.

Probably the best example of this sort of "independent thought" was seen in the "Close Combat" series of games. As long as all was well, units would meekly comply with your every order. As soon as they took some fire, they'd rapidly develop minds of their own, and not infrequently beleaguered troops would flat-out refuse to obey you. Also, sometimes, if you ignored them long enough, they'd take matters into their own hands and head off on their own adventure.

This sort of thing is very difficult to simulate in a playable game, both in terms of engine and interface. What is more, there is little incentive for authors to do so, since most players are very unhappy if they don't have microscopic control over their minions. Even military clients, who should know better, often demand unrealistic levels of control over their simulated troops.

--- Kevin

[CPangracs]

Scripting should also be kept to a bare minimum in ATF. There is no way to plan for every conceivable action by the player of a scenario. Just look at how many times you stop and change orders/routes/etc. during the game!

KISS it! Keep It Simple and Stupid. Vary routes for different units, add stop nodes along routes for various lengths of time, specify a time to commit forces rather than a specific event trigger, as often the case is that the trigger never gets initiated for one reason or another, and reserves never enter tha fight.

The problem with trying to enter a plan with a large number of "What-If's" is that it slows the game to a crawl. Fewer AI orders means smoother game play. The above is much easier and much more effective when Red is in the defense, but can also be used effectively to throw an unconvential plan at the Blue!

Missions should be built with replayability in mind. Keeping things simple and using saved battle plans are just two ways to do this.

Curt

[Pat Proctor]

In ATF, when unexpected situations occur, units mostly just follow orders blindly and either plow ahead to their deaths or sit foolishly waiting for input. They ought to do *something* when they're being shot at, even if that "something" isn't what you might have hoped they would do. SOPs are a start at this sort of semi-autonomy, but in their present state are rather blunt tools.

All I would say in response is that men, under stress, often revert to "herd" mentality. Here at the NTC, they call it, "drive to the lights" syndrome. Whole armored companies will plow forward, to their simulated deaths, even after seeing the first half of the company destroyed. One can argue that this would never happen in real life, but I would disagree.

As for the original question, I would just say that "scripting" is does not, neccesarily, invalidate the modeled enemy. I freely admit that computer AI will not, for a very long time, be as competent as a human opponent, but computer AI CAN give a man a run for his money.

Military tactics are a limited domain. While there are many more variables than in a simpler game, such as chess, the domain IS limited. When you further limit the domain by abstracting the infinte complexities of human vehicle crews to the neccessary subset of unit behaviors in ATF, you further limit the domain. So it IS possible, just as a computer can be programmed to beat a chess master, to create a completely competent military simulated opponent.

I have done my best not just to create an opponent that tries to win, but an opponent that approaches tactical problems the way a real, professional military commander would. I have strived to use "decision point tactics", just as modern military commanders do, not just to model a thinking enemy, but an enemy that thinks LIKE the military enemy you would face on the modern battlefield.

If you beat the game, it doesn't mean the AI sucks, it means you appropriately synchronized combat power at the decisive point and defeated the enemy. Any force that successful does this will beat any other force on the modern battlefield, no matter how smart they are.

"Sometimes you get the bear, sometimes the bear gets you!"

[CPangracs]

...

"Sometimes you get the bear, sometimes the bear gets you!"
Why does this sound so familiar? :D

[Chelco]

Thank you gentlemen for your considerations.
It transpires then that the tactical quality of an OPFOR in a particular scenario will reflect the tactical knowledge of the designer of it.
I never meant the AI sucks or things of the sort. In fact, it beats me to pieces every time. I find some relief reading that even Cpt. Proctor has troubles sometimes beating it. I was just pointing out some thoughts about the AIs in general and looking for some feedback.

[Deltapooh]

Thank you gentlemen for your considerations.
It transpires then that the tactical quality of an OPFOR in a particular scenario will reflect the tactical knowledge of the designer of it.
I never meant the AI sucks or things of the sort. In fact, it beats me to pieces every time. I find some relief reading that even Cpt. Proctor has troubles sometimes beating it. I was just pointing out some thoughts about the AIs in general and looking for some feedback.

This is a fair assessment. I spent days setting up Sand Castle 2.0 defenses, and KBluck still blew through it with ease. :mad: It also depends on one's under- standing of the scenario editor. I have plenty of ideals that would simulate a more intelligent force. However, getting all these features to work in an AI plan can be a challenge.

I would like to see an event trigger that allows the game to switch between battle plans. This would permit a scenario builder to develop several battle plans and set specific triggers to initiate or cancel activation. Right now it is an "either/or" situation. You set your AI forces to do one thing and develop an AI plan to respond to another (more than likely) player action. Multiple Battle Plans would allow you to create behavior closer to combat.

Of course, we are fighting at company level. There isn't alot you can do when the enemy is 5km and advancing at top speed except fight, run, or surrender.

Another advantage would be more control of missions. This is the best feature of the game. I'd like to be able to use it more in scenarios (not to mention the game in general), but it is difficult to control what units will do and where.

[CPangracs]

This is a fair assessment. I spent days setting up Sand Castle 2.0 defenses, and KBluck still blew through it with ease. :mad: It also depends on one's under- standing of the scenario editor. I have plenty of ideals that would simulate a more intelligent force. However, getting all these features to work in an AI plan can be a challenge.

I would like to see an event trigger that allows the game to switch between battle plans. This would permit a scenario builder to develop several battle plans and set specific triggers to initiate or cancel activation. Right now it is an "either/or" situation. You set your AI forces to do one thing and develop an AI plan to respond to another (more than likely) player action. Multiple Battle Plans would allow you to create behavior closer to combat.

Of course, we are fighting at company level. There isn't alot you can do when the enemy is 5km and advancing at top speed except fight, run, or surrender.

Another advantage would be more control of missions. This is the best feature of the game. I'd like to be able to use it more in scenarios (not to mention the game in general), but it is difficult to control what units will do and where.Missions are, currently, the games biggest limitation, IMO. Although, the easiest thing to do is create an "Enemy in Defense" scenario. Though this is the main theme in the game I'm working on, I am trying to spice things up a bit without creating intricate triggers, which slows down the overall simulation. I'm waiting for Pat to play the first two scenarios I have completed to get a general idea of how this pans out. Again, sometimes the simplest things prove to be the hardest for the player to overcome!:devil:

[kbluck]

This is a fair assessment. I spent days setting up Sand Castle 2.0 defenses, and KBluck still blew through it with ease. It also depends on one's under- standing of the scenario editor. I have plenty of ideals that would simulate a more intelligent force. However, getting all these features to work in an AI plan can be a challenge.

One thing to remember, however: we don't want to be creating scenarios that are "unbeatable". Challenging, certainly. But somebody out there is going to figure out the key to unraveling the enemy's plan, no matter how well considered it might be. My variant on Sand Castle, for example. I've made some changes that I think will complicate matters for the Blue force, but make no mistake: it can still be decisively beaten, and in much the same way. As Pat pointed out, if one beats the scenario, it probably reflects the quality of the player's plan moreso than the quality of the AI.

The standard of AI success, in my mind, is this: did both sides fight in a "fair" way? In other words, did the player do things that would be reasonable in real life, and did the AI respond in ways that a real enemy commander could reasonably be expected to do? If so, and the player beat the AI, congrats are in order. If ridiculous effects occur, such as enemies zapping entire companies of dismounts with Spandrels, then something needs to be fixed. Of course, there is often a fine line between the truly "ridiculous" and things simply not working the way a player might like.

In general, speaking from a scenario design standpoint, I would like to see it easier to organize things and to set alternatives in motion based on triggers. The "Prerequisite Data" is a good idea, making it possible to store useful things in an easily referenced manner. Three things that are notably missing from that paradigm are event regions, fire missions, and heirarchy missions. It would be *great* to have a cache of alternate missions available to assign to units as events develop. It would be really nice to be able to organize and easily reuse event regions for different things. It would be nice to be able to assign your own fire missions to artillery units on a trigger. And DP had a good idea about being able to "select" battle plans. Perhaps I would qualify that by changing the term "Battle Plan" to "Order of Battle" to vary the force mix; that would be the "random" part. Then allow multiple AI Heirarchy Orders per OOB that can be selected on triggers.

In general, the major difficulty in creating scenarios is the rather primitive interface for the order editor. It is very difficult to keep track of everything that is supposed to happen without a single overarching "view" of the event path. It would also be very helpful to at least know what everything does and what data it expects, not least the things that don't actually do anything (yet).

--- Kevin