How would you rate the AI??

[Adamo]

Just wondering how you guys all feel about the AI in this game (meaning your AI opponent). I'm kinda new so I'm pretty happy with it now, but I'm wondering if many of you 'experienced' players still play against the AI and enjoy it.

thanks

adam

[Redwolf]

It is one of better ones in existing wargames. And those which are better are either exploiting a huge advantage in number crunching (TOAW) or are partly scripted (TacOps). CM's AI is self-sufficient and actually needs to do the same thing the player is doing, so it is certainly the most impressive I know.

[Moon]

Thanks Redwolf.

Adamo, this of course still does not mean that it's anywhere close to a human player, especially not an experienced player. The AI has many handicaps, one of them is not being able to learn, while the human player gains experience, both with game-specific features, as well as the tactical environment as a whole. After a while, you'll start beating the AI on *most* occasions, especially when you play a lot of similar scenarios (e.g. QuickBattles).

While the AI is "fuzzy" (i.e. it doesn't always choose the exact same solution when faced with the exact same problem more than once, but can actually choose from a number of random options), it still is not endlessly so, meaning that you will learn some rough patterns on how it approaches tactical problems. Bit this is the same thing, really, as when you'd be playing against the same human opponent all the time.

Martin

[junk2drive]

a lot of players claim the ai is too easy to beat. maybe qbs, which i dont play, but try a scenario made by someone with experience and playtested. some of those are tough. if the designer says to give the ai some extra, do it. if it is too hard too beat, restart with less.

weather in operations is random. i have had to restart some to get the weather i wanted for the tactic i wanted to employ.

to answer your question, yes i find playing the ai satisfying.

[SKELLEN]

Originally posted by Adamo:

Just wondering how you guys all feel about the AI in this game (meaning your AI opponent). I'm kinda new so I'm pretty happy with it now, but I'm wondering if many of you 'experienced' players still play against the AI and enjoy it.

The AI is definitely better/best in defence than offence as you may becoming aware, but it's also no push-over in offence either as I've discovered a number of times.

On occasions I've beaten the AI and on others I've had my butt kicked too, and I'd consider myself an average player also, so the AI is pretty astonishing in my book.

There're problems you will notice with the AI for sure like:-

1) Battalion and Company HQs being sent ahead of plattoons or squad in offence and usually resulting in the AI losing them unnecessarily.

2) Traffic jams or to put it another way, the gathering of its armour in the same area or small area - easy targeting.

3) When in defence the AI sometimes sends its units on walkabout from it's foxhole.

4) MG and Bazooka/Panzershrek teams have a tendency to move within full view of my units without support.

You will obviously discover other niggles too, but in one QB I played a while ago with the AI in defence and which I lost btw, at the end of the game I was gobsmacked seeing how competently the AI had placed its units and especially MG units. I'm not sure I could of done much better and that's no word of lie (praise to BFC), so yes in conclusion the AI is not perfect and never will be, but you have to admire the way BFC have created probably the best AI they can at present and the mind boggles thinking what the AI will be like in CMX2 or any future version of CM.

[ September 01, 2004, 08:41 AM: Message edited by: SKELLEN ]

[Panzer76]

On defence, pretty decent.

On offence, out right bad.

Play versus humans, the AI is only good at learning you the game. But dont learn too much or you will be picking up some bad habits

[Pzman]

The AI can be very good on the attack, given the right conditions. As noted before QB battles tend to show all the AI's weakness. In a scenario you can setup VL's to guide the AI where you want it to attack. Its one of the best AI's around, and its getting close to four or more years old. There have been some changes, but not that many from what I can see.

[mrpwase]

While the AI certainly is very good (relative to other wargames), the 'niggles' stand out a lot more. For example: in CMAK I was defending against the Amis, and was bewildered when 2 Stuarts appeared on my right flank, moving along a hedge - making them easy, slow prey for my AT guns.

Yes, the hedge probably provides a tiny bit more cover than open ground - but it also slows the Stuarts down to the speed of a HMG team, so they were both dead within seconds. Whereas if they'd have been moving along the nearby road, they could have reached cover before I even got them in my sights.

But that's just whineing on my part - the AI is good in most places.

[Redwolf]

The AI can be turned into a halfway decent attacker by giving it its forces as reinforcements in an order that makes sense and ensures it doesn't run in the artillery spotters first.

Then, you should have lanes of flags guiding it. And you should turn flag display off when playing a scenario specially designed that way. Your goal should be to break the AI's force, not take flags, the flags are for steering the AI and you must not see them so that you don't spoil the enemy "plan".

Give it elite artillery spotters. But I am not a friend of generally raising the experience level, it is a bad substitude for good tactics. And it makes for bad training if you want to play humans later. Some HQ bonusses are probably a good idea, though.

I also don't like giving it more forces, that turns the whole scenario into something else and again is bad training in preparation to play humans.

Instead of beating the AI, your goal should be to break the enemy force with minimal losses of your own. Your victory conditions should be 1) mission accomblished or not and 2) friendly losses. 20% friendly losses should be inacceptable.

[Jack Carr]

I enjoy battles against the AI very much. Sometimes it does foolish things but I still get whooped by it on occasion.

This is the best damn game in history!

[JasonC]

Good enough to teach you how to play, to use units sensibly, etc. Reasonably dangerous if you give it high odds (2-3 to 1) or large amounts of superior armor (platoons of heavy tanks vs. ATGs, or companies of reasonably thick tanks vs. modest armor yourself). But overall, no match for any human who knows what he is doing in standard odds scenarios, and sometimes comically weak.

[danielh]

As Panzerman already wrote, quite good in defence. But really weak on the attack, tanks bunching up generating endless traffic jams is one example. It is almost impossible to create a massed tankattack as was so common on the russian side. Infantry also has the fatal habit of bunching up in ridicolous amounts. Only by creating "unnatural" exactly designed scenarios is it possible to reduce this a bit (every new "row" of tiles has to have several good cover tiles which have the same distance to the good cover tile of the previous row). A kind of scripting in addition to the actual system would be very, very helpful for the scenario designer, as would be a "Debug" mode where the designer could play the scenario in "god" mode seeing every unit, while the engine would play with the given FOW setting (Extreme fog of war normally) -> checking AI behaviour. Now reducing FOW settings changes the behaviour of the units.

Greets

Daniel

[JasonC]

Actually there is another way scenario designers can get the AI to perform reasonably well on the attack, without using unrealistic terrain. You send the attackers in several distinct groups with sensible internal tasking, e.g. by using reinforcement groups. Also, put small flags at reasonable, accessible locations along multiple avenues of advance.

If instead the AI starts with everyone on the board, and the flags all deep in the enemy rear, they tend to pick a supposedly "best" route and send far too much stuff along that route. With predictably awful consequences. The defenders often are waiting, obstacles are in the way, arty gets them bunched up, they pile up at chokepoints along the route like a difficult to cross open field etc.

Instead put a small flag on route A and a starting force in a reasonable position to get close to A. Put a reinforcement marker somewhere else, along that edge or an adjacent one, closer to a different small flag reachable by route B. Arrival something like minute number 5. Then put another reinforcement group behind the initial force, arrival something like minute number 10 - perhaps with armor this time. You get the idea.

The AI will deliver multiple blows and the strength and composition of those blows will change over time. The designer has wired those strength changes into the set up rather than leaving them to the AIs discretion, so they aren't based on reactions to human positioning etc. But they are unknown to the human player, different for different scenarios, make some sense as to force composition and terrain, etc.

You can't really rely on the AI doing this high level tasking and overall map plan, sensibly. But you can approximate a high level script with reinforcement groups.

[CombinedArms]

As many have noted, the AI is better on defense than offense. But it does have a few foibles on defense as well. Perhaps the worst of these is a tendency to mount ill-advised counterattacks on a forward flag that you as attacker have taken. Units in foxholes and trenches around a large flag will sometimes jump out of that cover and attack across open terrain to try to retake a small flag that has already fallen and that is firmly in the attacker's grip.

They generally get mowed down in the process and often leave the large flag undefended as a result.

This seems to be a part of the general coding of the AI to advance toward all enemy held flags. It would be nice if there were some kind of coding to tell the AI--"You're on defense, you idiot. Stay in your foxholes." It's possible to deliberately exploit this tendency of the AI's but generally I find myself exploiting it unintentionally. That is, I take a flag as a natural part of the progression of the attack, and then the ill-advised AI counterattack begins.

[Mad Russian]

Originally posted by JasonC:

Actually there is another way scenario designers can get the AI to perform reasonably well on the attack, without using unrealistic terrain. You send the attackers in several distinct groups with sensible internal tasking, e.g. by using reinforcement groups. Also, put small flags at reasonable, accessible locations along multiple avenues of advance.

If instead the AI starts with everyone on the board, and the flags all deep in the enemy rear, they tend to pick a supposedly "best" route and send far too much stuff along that route. With predictably awful consequences. The defenders often are waiting, obstacles are in the way, arty gets them bunched up, they pile up at chokepoints along the route like a difficult to cross open field etc.

Instead put a small flag on route A and a starting force in a reasonable position to get close to A. Put a reinforcement marker somewhere else, along that edge or an adjacent one, closer to a different small flag reachable by route B. Arrival something like minute number 5. Then put another reinforcement group behind the initial force, arrival something like minute number 10 - perhaps with armor this time. You get the idea.

The AI will deliver multiple blows and the strength and composition of those blows will change over time. The designer has wired those strength changes into the set up rather than leaving them to the AIs discretion, so they aren't based on reactions to human positioning etc. But they are unknown to the human player, different for different scenarios, make some sense as to force composition and terrain, etc.

You can't really rely on the AI doing this high level tasking and overall map plan, sensibly. But you can approximate a high level script with reinforcement groups.

You are absolutely right Jason. There are ways to help the AI attack and defend. The newer CMAK seems to want every scenario to be a meeting engagement so you simply phase in the forces as you suggested.

I particularly like the way the AI works on the Eastern Front. I believe that for the most part, except for the T-34 dance of death that it does a pretty credible job of being a Russian commander. The more regimented a combat system the better the AI will portray it. As Moon said the AI is the same each time out. So you learn it's strength's and weakness'.

Certainly in constructed scenarios the AI can give a good account of itself. Check out HSG KC Oberloskamp(shameless plug), at the SD, for a scenario that the AI does a fairly good job of attacking. Play it, then go into the editor and look at how the scenario is set up. You will see that the AI will attack you from different directions as Jason has pointed out.

Giving the AI some help is easily done but it doesn't always work out. That is why you have to do extensive playtesting for AI scenarios to make them work.

I think the biggest AI weakness is a scenario designed to do too many things at once. Like, play from either side vs the AI, and be a two player game as well. It is very difficult to have the AI play well, on both offense and defense, with the same objective set. The goal of the two games are different.

Unless you are trying for a meeting engagement where both sides have the exact same objectives.

Tring to match a H2H scenario to a vs the AI format is also very difficult. The human players will play better than the AI in most instances. So how can you balance it for both? The designer should decide going in which outcome he is looking for. AI on the attack...AI on the defense or two player.

That alone will increase the efficiency of the AI IMHO.

Panther Commander

[DavidFields]

The tendency of the AI to counterattack sometimes makes it easier to beat, but makes the game a bit more exciting. Close Combat III, as I recall, had more passive AI, for example, but was thus less exciting.

There has been more than one time where I thought the computer was launching a stupid counterattack, only to sweat it out as those "dummies" refused to die or retreat as expected, and made a flag neutral, or worse.

Also, this helps train the human player to advance in force, and not just throw a couple of units at a flag.

As to how good it is: I am actually taking a break from the game because I had my nose punched in by the AI at the end of a long scenario. So, when the AI does something stupid, we can laugh at it, but do we really want to have out head routinely handed to us when we are playing for "pleasure". I find it the best, most realistic, AI I have every seen. 30 years ago, the Department of Defense would have probably paid millions to build something which would have still been inferior. (I am willing to stand corrected if someone knows better)

[Sanok]

The AI makes two crucial errors, whether attacking or defending.

1. It doesn't keep its infantry in command range of their HQs.

2. It exposes its armor. On the attack, the AI will often lead the way with its main tanks, instead of scouting with infantry and keeping the armor back. On defense, it will see your attack coming and move its armor forward.

I think the AI is excellent at placing MGs and guns in places with very good fields of fire.

[DavidFields]

Yes, of course Sanok makes excellent points. Leading with the armor is something even an experienced human could do, either on purpose or by mistake. But the AI compounds the problem, I think, by often (usually?) not pulling back and trying another route when it is being clobbered. "Persistence" turns into "massacre"--though I have occasionally been guilty of that myself if I impatiently thought that "a slightly larger tank rush" was going to overwhelm a AT gun, for example.

Not keeping its units in command distance of the HQ is one of the most "inhuman" things which I see the AI do. Most of us humans, I think, are obsessive about keeping units in command.

[danielh]

JasonC,

Correct, but doesn't help much if one tries to simulate a massed attack, a real shockforce, as was often done by the Russians at their point of main-effort (and surely too by the germans if enough manpower was available). Where they tried to saturate the surviving enemy direct fire with numbers, and uncover them to the interspersed Tanks and the following Assaultguns and Tanks in the second wave. If done on a fairly broad front 4-10 km the enemy must have his artillery largely intact to have a chance to keep the first wave from entering the line. It's also bad the AI is not capable to execute a walking barrage in front of the first wave. Some planning tools for the designer would help much here (The AI tends to uload all Arty onto one spot, either near a flag or something spotted in the first turn).

Sanok,

Disagree, the AI when attacking, always tries to keep tanks level with the advancing infantry, if the infantry stalls the tanks most of the time wait forever. One can easily determine the frontmost line the infantry has reached within about 100m range even if nothing is spotted. I'm conviced that it is only possible to recreate the vast possible types of attacks only by giving the Scenario-Designer some additional tools to guide the AI to some extent.

True for your last remark about the placement of guns and MGs, they are almost always placed very well.

Greets

Daniel

[Chelco]

For experienced players, surely the AI is not gonna pass the Turing test.

For dummies like me, I am very happy with it and consider it among the best in my HD (along with Highway to the Reich and Korsun Pocket).

Besides the already present low-level tactical behaviours which are very convincing, we wouldn’t expect the AI to create elaborate plans. Check this out:

A research paper about how to create tactical plans for computer-generated forces

The paper is a bit old, but my point is that it takes whole research teams to make an AI which can make plans.

Cheers,

[Redwolf]

If that is the paper I have seen a few years ago (server down right now) it concentrates forces very well but totally ignores the threat of enemy artillery.

[Chelco]

Originally posted by Redwolf:

If that is the paper I have seen a few years ago (server down right now) it concentrates forces very well but totally ignores the threat of enemy artillery.

Indeed, there are not artillery units in this paper.

[Redwolf]

Must be what I have seen then.

This is a more fundamental flaw then it appears on first sight. You cannot "tag on" artillery behaviour later.

A systemd designed without fire support in mind is totally one-dimensional. A system designed with it in mind is a system developed around the tradeoff detween dispersion and concentration and, more importantly, around the change that this tradeoff goes through over time. In other words, the programmed opponent needs to have a sense when to concentrate and when not. The one-dimensional opponent on the other hand, while it sees changes over time, does not see its fundamental tradeoffs change over time.

[Chelco]

Originally posted by Redwolf:

Must be what I have seen then.

This is a more fundamental flaw then it appears on first sight. You cannot "tag on" artillery behaviour later.

A systemd designed without fire support in mind is totally one-dimensional. A system designed with it in mind is a system developed around the tradeoff detween dispersion and concentration and, more importantly, around the change that this tradeoff goes through over time. In other words, the programmed opponent needs to have a sense when to concentrate and when not. The one-dimensional opponent on the other hand, while it sees changes over time, does not see its fundamental tradeoffs change over time.

The absence or arty could be pointed as flaw, yes. But they are trying to model a single "toy problem", and it appears to be tough enough.

[Redwolf]

Fair enough. They approach an interesting and challenging programming problem. It's just that the solution wouldn't be a good opponent in most wargames (maybe in CM), not to speak of the real world.