Question about the AI

[Mo74]

Hi all,

I liked the demo on I'm on the verge of picking up CMBN/Commonwealth - but a couple of things I read has put me off a bit. It concerns the AI.

In the Rock, Paper, Shotgun review of Commonwealth it mentions the static/scripted AI as a negative point.

In a thread the other day I read about the pros/cons of timed battles, there was a response again highlighting AI being scripted/static for each scenario.

The franchise I really enjoy is the Total War series: in the real-time battles the AI attacks/defends/reacts so every battle plays differently, even the same one if you reload it. No AI/game is perfect but it puts up a fight.

From what I have read CMBN scenarios are canned so once you have played it, you're done.

So, I'm wondering, what does the AI actually do in that case?

I know Total War and CM are very different games and both do their thing very well. I'm just interested in what I can expect from the CM AI.

I hope my question makes sense and sorry for the long post.

Thanks,

Mo.

[Paper Tiger]

Mo74

Yup, your question makes sense.

A good stand-alone mission that has been created for playing against an AI opponent will have several AI plans and so the missions can play differently each time you play them. There can be as many as five AI plans for the AI-controlled side so it might take you a while to play against each of them. There's no guarantee that you'll get a different AI plan each time you play the mission though.

Different defensive AI plans will place the defenders in different positions from plan to plan and so will prevent you from having prior knowledge of their set-ups when you replay a mission. They may also behave very differently in each plan. When attacking, the AI imay take different approaches to the objectives so again, you won't know where the real attack is coming from until the attack begins in earnest.

I think you'll find that that review also stated that the scenario designers who created the content did a very good job of scripting the AI. Clever scripting can present even a seasoned player with a real surprise when playing against the AI. So, if you like the demo, I doubt you'll be disappointed. And there will be a ton of stuff coming from the community to play as well.

Hope that helps.

[~Viajero]

Well, playing solo against the AI will definitely not give you the same challenge than playing against a fellow player that is for sure.

This game shines in human vs human play obviosuly and that is what I would first and foremost recommend to you.

Having said that even though the AI is scripted (be it to not move a certain unit, or to move it in a deternined way) it can still present enourmous challenges to a human opponent and can be loads of fun. You still need to figure out how to properly get rid of those keyholed units here or how to flank that pesky jgdpz in hull down over there etc. The challenge level is in no way disappointing, at least to me. Regarding replayability the number of designed battles and campaigns availble out there already is so big that I assure you, you ll have more than enough to contend with eventhough you may consider certain of them can not be "replayed".

[Humbug]

It does put up a fight if the scenario builder isn't incompetent. I'm not sure if bringing TW into this as a positive example is very accurate however. I own most of the TW games, have played them extensively, and if it's one thing the TW developer is known for it is certainly NOT well made AI. Many of the TW titles like Medieval 2 and Empire have completely atrocious AI and is a very common complaint. The newest TW game Shogun 2 has much improved AI but it still does a LOT of very stupid things like charging its general alone into your spearmen. Allied AI's ust running straight into a vastly superior force instead of forming up together with the human player, standing around with units instead of helping his own forces that are fighting, and much much more. But I digress.

The CM AI is by no means perfect, but it's not true that it will always do the exact same thing in one playthrough of a scenario to the next. Scenario builders can add multiple plans for the AI with different chances of execution to keep the player guessing, and the combat AI may not choose to execute a plan 100% to the letter every time. Unit positions will vary. It's not a very dynamic AI however and will not react to your decisions.

All in all though I'd say the AI may not be perfect, but it doesn't go off the wall and do utterly outrageous things very often either. A well made plan and it can be very deadly indeed as I'm sure you'll find out. If you want a truly dynamic opponent however I'd recommend you play against other humans like yourself. That's where most of the fun lies.

[YankeeDog]

Hi all,

From what I have read CMBN scenarios are canned so once you have played it, you're done.

So, I'm wondering, what does the AI actually do in that case?

Well, yes and no -- Yes, the computer player AI in CMx2 relies on scripted behavior. But it's a bit more complex than this.

First, the scenario designer only sets general parameters of the AI plan. The designer sets the initial deployment of units, organizes them into groups, and then scripts instructions generally for what these groups will attempt to do: If, when, and how they will try to move, etc. But the exact moves of individual units within the AI groups is still determined by the computer. In other words, the scenario designer script determines the general plan, but the computer AI fills in the details. And within the restrictions of the AI script, the computer has leeway to adjust and react to what it sees of the human player's actions.

So within the AI script there is some flexibility; even when executing exactly the same script, the computer player's actions probably won't look exactly the same in two different play-throughs.

Secondly, scenario designers can and usually do create multiple scripts for a scenario. So, just because you've played a scenario once, doesn't mean you're going to see the same AI plan the next time you play the same scenario. There might be, e.g., one plan where the AI attacks up the left flank, one where it attacks up the center, and one where it attacks up the right. Just because you've played a scenario once, doesn't mean you'll see the same plan of attack on your second attempt.

Overall, CMx2's AI system does have its advantages and drawbacks. In a well-designed scenario where the designer knows what he's doing and has scripted good AI plans, it can be very good indeed -- probably far better than any AI system that simply left everything up to the computer. However, the AI does lack flexibility. It can adjust the details, but it can't make major shifts in tactics once the battle has started. For example, it can't send recon units up several possible routes of advance, and then make a decision based on resistance encountered as to which route to take with its main force -- once the battle starts, it's locked into a general plan.

Practically speaking, I don't think there was really an alternative to scripted AI behavior for CMx2. The game is so complex that I don't think a computer AI could really play competently unless BFC had the resources to hire a full team of programmers to work on the AI routines only, and with these kind of resources I'd guess the AI would still be a pretty mediocre opponent. In any event, it's moot because BFC has exactly two coders and so obviously doesn't have the resources to put even one coder on the AI full-time, let alone a full team of specialists.

If you want more info, I would suggest reading the sections of the manual on scenario design, which will give you a good idea of what is and isn't possible with the AI scripting in CMBN. The demo scenarios also have multiple AI plans, so you can see this feature in action by re-playing these scenarios.

Cheers,

YD

[hoolaman]

A scripted AI can provide an aggressive and surprising attack or a rush down a flank that a dynamic AI would never be competent enough to pull off.

but

The AI will not react to your movement on a tactical or strategic level in any way. It is pretty easy to set an ambush for the AI where they will just run squad after squad into your trap until there is literally a pile of corpses in front of your MGs.

I dream of some sort of hybrid dynamic AI but since CMSF there has been virtually no improvement to this part of the game. TacAI yes, there has been massive improvement, but the "StratAI" that used to be touted in CMx1 is just not there.

[phil stanbridge]

If you enjoyed the demo, forget the reviews. Just go and buy it and help support BFC. I play Normandy on a daily basis against the AI and enjoy it immensely. It really can put up a good fight!

[Mo74]

Hey all,

Thanks so much for the quick and comprehensive replies. I feel encouraged by what you say. I like the point made that, while the AI might not be able to adapt massively, it follows a general plan and makes up the smaller details - and how it might react could be a bit different each time.

And being able to make your own maps/scenarios means you can make up whatever you want.

I'm not sure if bringing TW into this as a positive example is very accurate however. I own most of the TW games, have played them extensively, and if it's one thing the TW developer is known for it is certainly NOT well made AI. Many of the TW titles like Medieval 2 and Empire have completely atrocious AI and is a very common complaint. The newest TW game Shogun 2 has much improved AI but it still does a LOT of very stupid things like charging its general alone into your spearmen. Allied AI's ust running straight into a vastly superior force instead of forming up together with the human player, standing around with units instead of helping his own forces that are fighting, and much much more.

Yeah, I agree with you about this. Absolutely. I didn't mean Total War was a shining example of AI, more just it reacted to you somewhat on the fly, and that I enjoyed that kind of sandbox play, even if what gets served up is a bit of chaotic jumble.

Thanks again,

Mo.

[RockinHarry]

Maybe BFC threw out the strat AI, because it would freeze RT play whenever used by the game engine?

[Mo74]

Sorry, just read this bit.

TacAI yes, there has been massive improvement, but the "StratAI" that used to be touted in CMx1 is just not there.

Did CMx1 AI have the scope to react on a different level?

[YankeeDog]

Maybe BFC threw out the strat AI, because it would freeze RT play whenever used by the game engine?

Couldn't say for sure, but I doubt it's about computing resources. Good AI routines aren't necessarily resource-hungry, they're just notoriously difficult to program and debug. Seriously, if BFC could actually program a computer player AI that could credibly play CM, they'd have the Pentagon knocking down their door, looking to buy the code.

Sorry, just read this bit.

Quote:

TacAI yes, there has been massive improvement, but the "StratAI" that used to be touted in CMx1 is just not there.

Did CMx1 AI have the scope to react on a different level?

CMx1 was a much simpler game, and there were no AI scripts in this game -- the computer AI determined everything, from the higher-level plan, down to the moves and actions of individual units. The highest level of planning was usually referred to as the "StratAI".

The StratAI in CMx1 was also very predictable and easy to fool. It tended to overreact to the first sighted enemy units near a victory location, making it very easy to draw it into ambushes. It also had a very hard time with more complex tactics. For example, it had a very hard time coordinating artillery usage with tactical movement.

In my opinion, the CMx2 computer player system, while it has its limitations, is vastly superior to what was in CMx1.

[Kat Johnston]

Playing thoroughly random (random forces, objectives, and map) quick battles (not scenarios, which are better and harder) against the AI, I'd say the AI provides a proper (though beatable) challenge about one time in three, and most battles are enjoyable in any case. I rarely lose random quick battles against it, but it can happen.

The main drawback to playing against the AI is that once you've a feel for how it operates, you can get away with doing things that you know a human opponent would spot and act against - you can concentrate forces where a human would realise and mortar them, you can flank knowing that the enemy likely won't change position if they see what you're doing, that sort of thing.

Oh, and it doesn't do recon-by-fire, or other forced fire, which we humans love.

If you fancy more challenge from the AI you can always stack the odds by increasing the forces available to it using the force adjustment settings, or by deliberately taking on fights using "unsuitable" forces.

[Macisle]

Mo74,

I think that at the very least, you will feel that you got your money's worth with this game. And more likely, you'll want to pinch yourself that it was ever possible to get this much entertainment, and get so close to a realistic WWII battlefield experience (from the armchair side of course!), for so little money.

I had the first MTW game and I certainly don't think the CMBN AI suffers at all in comparison. On the defense, it can certainly put up a good fight. On the attack, well...not as often, but sometimes--yes.

As for scenarios being canned, it's mainly a question of how much work the designer wants to put in. The tools are there for them to put in more plans than you will probably want to play for that given scenario. And from what I've seen, many designers are good about letting you know how much replay value there is in a scenario they've uploaded to the depot.

Also, something to consider here is that BF has given you all the tools you need to make your own maps and/or take existing ones and alter them. So that means, terrain, units AND AI plans. For example, you could grab a map, create a few devilishly clever AI plans and then leave them for awhile while you play other battles. Then, when you've forgotten about what you did, go back and play against your custom, heightened AI plans.

So again, while the AI is not perfect, it can be very fun to play, and you can even try to make it better through your own scenario tweaking.

Really, the game is in a class by itself. Not perfect--just better than anything else for what it is trying to do.

Macisle

[Georgie]

The StratAI in CMx1 was also very predictable and easy to fool. It tended to overreact to the first sighted enemy units near a victory location, making it very easy to draw it into ambushes. It also had a very hard time with more complex tactics. For example, it had a very hard time coordinating artillery usage with tactical movement.

In my opinion, the CMx2 computer player system, while it has its limitations, is vastly superior to what was in CMx1.

A lot of people are calling for triggers,including myself. Won't triggers make the AI more like CMx1 and thus easier to beat? Maybe I should reconsider what I am asking for.

[mjkerner]

Georgie, triggers are easier to beat? If AI trigger is "when Human player moves into hedgerow at "x", AI squad/platoon/vehicle move to "y" and assaults "x" (and human doesn't know the AI plan), how does that make it easier? I am not being sarcastic...I really don't know because I never played any game with AI triggers. There are obviously some aspects of triggers I don't know about....

[MikeyD]

Won't triggers make the AI more like CMx1 and thus easier to beat? Maybe I should reconsider what I am asking for.

I tell people (jokingingly) that I'm the only anti-trigger nay-sayer out there. Everybody, including the developers, wants triggers. My own fear is that designers will become over-reliant on triggers and won't bother with the finer points of AI construction. As it is, getting people to include carefully coordinated 'Exit After/Exit Before' AI timeing is like pulling teeth.

[mjkerner]

Good point, MikeyD.

[Rocky Balboa]

I tell people (jokingingly) that I'm the only anti-trigger nay-sayer out there. Everybody, including the developers, wants triggers. My own fear is that designers will become over-reliant on triggers and won't bother with the finer points of AI construction. As it is, getting people to include carefully coordinated 'Exit After/Exit Before' AI timeing is like pulling teeth.

I'm confused by

carefully coordinated 'Exit After/Exit Before' AI timeing

Please elaborate

[Erwin]

CMBN and the CW module are superb and there is no close competitor out there regarding the quality and detail of the CM2 series of games.

We can discuss AI scripting and other details, but that is like debating the color scheme of a moon rocket as a determinant as to whether to launch or not.

BF has provided excellent free demos. If you like em you will love the full games. If you don't like the demos you probably won't.

[Dave85]

I too used to be a complete Total war junkie, on Med2 empire and napoleon I clocked up over 3000 hours doing campaigns and online battles.

I havn't touched it for 6 months.

If like me you loved total war but were always frustrated by CA's complete disregard for realism and historical accuracy, and you have a keen interest in ww2 then this game is for you.

With regards to AI I always felt CA tried far to hard to make its AI too clever without resorting to prescripted AI plans. They failed miserably.

I loved campaigns, multiplayer battles and especially multiplayer campaign on napoleon, but never had any time for single player battles. They were far too easy. The AI would charge its cavalry straight at your musket line from the off, leave its artillery undefended on the flanks, often with cannons on reverse slopes bouncing shot 100 meters over yr line, and feed its infantry peicemeal into yr waiting wall of lead.

No matter how outnumbered you were you always seemed to be able to get fire superiority.

Since getting CMBN I have only plyed single player and have been having a whale of a time . There are dozens of scenarios and campaigns designed by people who are far better tactically than I am. So in a sense I am not really fighting AI, I'm fighting the designer, its just that he had to make all his decisions before the battle and then cant change them.

So its kinda like when the huns invaded france in 1940 and the british and french had to wait 3 weeks to get permission from high command to fire a rifle or move a section 10 yds.

Having said that, on the defensive a well planned AI strategy will punish you severly if you take your eye off the ball. The designer will always put AT guns in excellent positions, but not the ones you expect. The first time you get some tanks to command, you will no doubt lose them all before you see a single german.

On the offensive the designer will just throw overwhelming force at you often with a feint attack on one flank and a hard thrust down the other, all the while stonking yr forward positions with mortars and artillery.

In quick battle the AI is not so much of a challenge and I generally dont play them as the attacker but if you do a defensive one and give the AI a 4 to 1 advantage in men and tanks, it can be a hell of a lot of fun.

Sorry for going on, I'll get to the point:

If you found TW AI challenging, and are worried that CMBN AI wont be.....

well prepare to get yr arse handed to you

[aka_tom_w]

You might be interested in reading this thread where the game is considered/suggested to be too hard because the time deadlines are too tight.

Its a good discussion :

http://www.battlefront.com/community/showthread.php?t=103215&highlight=too+hard

[Rocky Balboa]

You might be interested in reading this thread where the game is considered/suggested to be too hard because the time deadlines are too tight.

Its a good discussion :

http://www.battlefront.com/community/showthread.php?t=103215&highlight=too+hard

Good point this ... defeating the AI in CMx2 is much more involved than just which side has the most men on the field at the end of the battle. The designer in CMx2 can do things with time limits and victory points for casualties that can make up for the lack of AI creativity.

[hoolaman]

We can discuss AI scripting and other details,

thanks

but that is like debating the color scheme of a moon rocket as a determinant as to whether to launch or not.

It's more like debating whether it's better to be a solo test pilot that dies in a flaming pure oxygen atmosphere or part of the team that gets to walk on the moon.