[vonB]

Over the years, the subject of the AI comes up regularly.  I was minded to post some comments in another thread, but on second thoughts, it probably is better in its own thread?

I will admit that the issues regarding AI fascinate me.   I am minded to consider some of the posts by new players who basically say "how can I stop being whupped by the AI".  Does that mean the AI is strong or the player is stupid?  It could of course be neither.  The 2 most likely explanantions are that the player is not familiar enough with the weaponry and behaviours, and has not become familiar with the way the AI behaves.  Once these two things are sorted, the Single Player game becomes a very different animal. I remember my very first games playing CC2.  I did not always do so well!...

As any H2H player knows, H2H is a completely different game.  Now imagine if the AI was as good as the very best players, how many would find the game so compelling then?  I am sure we would get shouts for how to dumb down the AI.  It's human nature.  You have to be able to win sometimes!

The kicker is how to get software to perform like this.  To get it to 'think'.  It is easy enough to create logic which will give an advantage to the AI (to try and make the players' game tougher), but that's not necessarily good tactical logic being used, just 'stacking the odds', i.e. rigging the game.

For anyone who thinks this is easy, is probably not a programmer, and has little concept of the challanges this presents to a machine, and a very stupid machine at that.  So it has to be 'told' what to do.

The AI breaks down into many layers.  The first of which would probably be 'awareness', or what does it know.  Then you have to work out 'how' it knows what it knows.  Then you have to decide how it applies what it knows, and if you are very ambitious, how it 'learns' from the results.

The second is probably behaviour, which can be broken down into 2 areas; Tactical and Psychological.  The Tactical is the 'objective' logic, and the psychological is anything but.

It can help to gain a better perception of the challanges if you try to convert the actions you take into simple logical steps that can be used as a logical base for a computer to process.  Think that's easy?  In some cases it might be, but on the whole it is far more complex than you may think.

This can be broken down into other areas, such as deployment, manouevering, batttle posture, etc.  Take a look at deployment.  Take a look at how you deploy on a Map.  Now try and convert that into something that a computer would be able to use.  After you have tried to nail the 'why' you then have to translate that into what.  You will likely find that the logic you come up with to explain your own actions are neither clear, nor something if applied logically will come up with the same solution.

Why do you put a Team in that building?  How do you logically arrive at that decision?  Because, if you cannot come up with a logical reason for it, then a computer is not going to be able to, and what you might think is logical may be far from it.

For myself, I tend to be a 'gut feeling' kind of player.  I like the shape, symmetry, and 'sense' of where the strengths and weaknesses are.  While much of this will be based on a very good undertanding of the capabilities of the weaponry, it is not a 'logical' process, and therefore will introduce flaws into my play.  So be it.  I still get a lot of fun from it, so it doesn't bother me, and I am not a professional soldier with real lives on the line, so I do not need to be so particular.

Perhaps we should throw out the challange?  Write a logical explanation on how a computer can perform as an AI in Close Combat?  Don't attempt to map out everything in one hit.  You will not be able to do it.  All you will get is high level macro type logic which is useless to a computer, until it knows how to process EVERY detail, so at some point you have to nail the details.  What are those details?

If you were presented with a challange, say to take a building, you might be able to explain your own 'logic' as to how you would set about doing it, and I think you will probably come up with a number of exceptions, i.e. if the building is defended, then... if not, then.... but if this, then.... and so on.

I am not saying that every particular about the AI is difficult, nor is it impossible.  Some things are easy, but beware, what seems easy to you a computer does not understand.  If someone fires at you, you will seek cover.  Seems simple?  On your level it is.  For the computer you have to start from the position of how does it know it is being fired at?  Maybe not so hard.  Then it must do something.  This can be carrying on with whatever it was doing, or changing something.  Changing what?  How? Where? When?  Can that logic be applied globally to get a standard behaviour, and does standard behaviour apply ruthlessly, or can it be moderated by other factor.  What other factors?  How does it decide to override?

At the end of the day, it is all down to decision making.  The quality of decisions will be based on what is 'known' (awareness), and what actions can be taken in response to a situation or objective.  To make a competent AI, you have to provide it with all these capabilities.  And then you have to apply it and make a game that is enjoyable.

So go easy on the programmers.  They will have neither the brief nor the funds to take the AI where we would like it to go.  The prospect is far more complex than many would appreciate.  Some great minds have been working on AI using the worlds most powerful computers, and have not yet come up with a 'machine' that is anywhere near to the overall capabilities of a human being.

So, if you are minded, why not try and describe your decision making in simple logic that can be translated into something that a computer could use.  For most, this will not be as easy as it may seem at first, but if you want an improved AI, you can either wait for someone else to produce it, or you can help make it happen yourself...?

[platoon_michael]

I can't see how I can participate in this conversation unless I start with the errors I see in WAR which I have already done.

While I do not like a game I can never win at I seem to be one of the few(If not the only one)who accepts the fact that the only way an A.I. is to be any better is if the odds are stacked in it's favor.


I am minded to consider some of the posts by new players who basically say "how can I stop being whupped by the AI".  Does that mean the AI is strong or the player is stupid?  It could of course be neither.  The 2 most likely explanations are that the player is not familiar enough with the weaponry and behaviors, and has not become familiar with the way the AI behaves.  Once these two things are sorted, the Single Player game becomes a very different animal.

This to me is the biggest hurdle.
Many of us here have 100's if not 1000's or hours playing CC.
The only A.I. that could overcome that amount of experience would be one that has the ability to learn it's players attributes after Battle and use that against them.

[lamurt]

i am involved with falcon4 development and that's where i am basing this opinion of mine on. falcon4 is psuedo-open source since 99 due to it's source code leakage, and in this decade of time frame literally hundreds of coders, novice and pro alike, have poked it's code base, and it's a total mess at the moment. however the end result with close combat is pretty similar from my pov.

problem with software development is the people who can code simply not having the slightest idea what they are trying to simulate emulate mimic etc etc....

to them, usually all it matters is IF a piece of code is bug free and works as intended without causing any additional headaches. past this, any comment or any kind of criticism to their work will be met with the ultimate ignorance on their part because what they created works properly hence can't be wrong.

from the beta tester, management, player pov, going easy on the coders is never an ideal option because if coders are left to their own devices then the end result would be properly working nothing, simple as that.

[platoon_michael]

The whole open source code argument I don't get.
Who's going to work/fix it?

Mafi?
The Blood?

Aren't those two already on the team?


Lets say someone could produce a software that has the ability to "Learn As"
Now every time I play the map Bastogne the A.I. learns my strengths and weakness's Tactics after every Battle.

As time goes by it then adjusts it's FP and tactics based on my previous Battles.
Just for argument lets set the "Learn As" to a Max of 40 Battles.
It always keeps 40 Battles in it's memory but drops your first games played as it learns from each battle played.
This way it isn't remember Michael's horrible tactis he had when he first played the game too.

Eventually I am going to be confronted with an A.I. that's going to be hard to beat,it shouldn't become impossible to beat because it too has a set FP and Model Psyche for soldiers deployment to contend with etc etc.
So lets look into the future shall we?

4 years later Michael is playing an A.I. that's giving him a run for his money but unfortunately Michael isn't learning much because he only plays the A.I.

Now add the ability to swap My A.I. file with say someone like Schrecken who's skills far surpass Michael's
Now Michael is learning from an A.I. that has learned from someone elses skills who as we already mentioned is better than Michael.


We have a software at work that has a "Learn As" feature for every job.
It's not perfect but it does get better.

[Therion]

[vonB]

Open Source is a mixed blessing.  One the one hand, it can be an 'enabler' by allowing many people to contribute (many hands make light work?).  On the other hand, unless the direction and tracking is ruthlessly moderated, the results can be a total mess.  However, from the resulting mess, seeds can be gleaned which can help promote the objectives, but it is not the most efficient way.  Consequently, turning over the CC Code to the Public will not necessarily guarantee anything unless a small dedicated and very competent group were to take it on.  But you will never see the CC Code Open Source unless there is a breach of security.  Destineer own it, and there is no way that they will ever turn it over in my opinion.  It doesn't cost them a penny to keep it, and they can enjoy free (if modest) revenue from it.  LSA has just been released.  I rest my case.

The other issue is that there is no 'standard'.  Your opinion on what AI is or should do may be different from mine and anyone else's.  However, I think there would be much consensus between intelligent people.

I am absolutely on platoon_michael's wavelength, that is an AI that 'learns'.  We do, so why shouldn't it?  The way you play now may have little resemblance to the way you played your first games.  Why?  Because you learned.  If you follow this logic to it's ultimate conclusion, you will end up with an AI that you can never beat (at worst) and merely exact a stalemate (at best).  But I would see that merely as a need to 'dumb down' the AI's thinking in order to make it more 'beatable' if that is what you want.  Much easier to do that than to try and program an impossibly good AI in the first place.

To pick up on a comment in another post (someone said they didn't want to play 'Rommel').  Why not?  AI could have its own 'flavours'.  platoon_michael touches on this.  From his comments, eventually the AI would become particulalry good at beating HIM, but how good would it be at beating someone else who plays differently?  Now it gets interesting.

Now to the programmers.  I said don't be too hard on them becasue it is not in their brief to develop a competent AI.  I am sure the Publisher would love it, but the Publisher/Owner is only really interested in profit.  As long as it can do that, then they are happy.  Most 'game' programmers probably (I accept the possible offense this may give to some) could not program a competent AI to save their lives anyway.  It's not just about Code, but about 'intelligence'.  The ability to construct an application to do a job using Code does not in itself imply adequate intelligence.

Most game programmers are focused on 2 things, doing it as quickly as possible (due to commercial pressures) and with as few bugs as possible (as already mentioned by lamurt).  Indeed, they may not have the slightest interest or knowledge in the issue of AI, and if they are contract programmers, even with the product itself.  If it works then pay me, and so long.  Next job.

It seems to me the focus of effort in the game world is on 'prettiness'.  How flash the graphics are.  This is easy compared with AI logic.  Draw an object on the screen and get it to move from A to B.  Not 'why' move from A to B, or how to decide whether to move from A to B, and if the decision is taken to move from A to B, then how to deploy and manouevre?  That's tricky stuff if you want to try and model an 'intelligent' process, rather than a proscriptive scipted type of approach.  Not that scripting doesn't have a part to play, it can and should.  There is a mechanical level even in AI, but it is not the 'intelligent' component.

In fact, CC has some pretty clever stuff, but suffers from not being able to manage the processes sufficiently in order to remove the 'stupid' behaviour.  The spinning tanks is a result of fundementally good logic which is then turned into a mess, because it cannot follow through or complete the logic in sufficient depth to moderate the behaviour.  There was no intention to program spinning tanks, so the logic is incomplete.  The AI is inadequate.  It needs to be able to do more.  It may not be as simple as just adding Code that says 'if the tank is spinning, then stop it spinning'.  Doing that may break the whole logic that allows the tank to behave as a tank, and other performance characteristics may suffer.

The problem with the CC Code base is the domino effect.  The logical permutations are probably scarily high, and the knock on effects of changing even the smallest piece of logic can manifest somewhere else, and not necessarily for the good!  So you change one thing to 'fix' something, and a whole heap of crap then follows somewhere else.  You then fix that, and start another chain of crap somewhere else.  In the end you can end up with a truly broken system.

Unfortunately, the commercial environment is not conducive to such things as AI development, with perhaps some notable exceptions, such as the Sims, but then look at how much money that made!  The AI development was a crucial part of making a mass market product, and it worked.  You will never see this for a minority interest tactical war game.  period.

The best AI developments come from Educational Establishements and Research (such as NASA).  The psychological model used in CC comes from academia.  They do not have a commercial imperative in the same way.  In fact, they can even get away with it not working!  Put it down to research.  How much do you reckon Destineer, Matrix, or investors are prepared to front up for 'research' and do you reckon they will be pleased if the result was failure?  I reckon it is probably a bit less than $1, UNLESS they were convinced that it would make a big money spinner.  If you could do that, then you could get the money.

However, I do see the possibility for it to happen, but not if it is purely a CC thing.  Unfortunately that is also a problem in a commercial environment.  Who is going to pay for it?  But remarkable things have happened just because some people refused to give up and accept the status quo.  I am not going to pretend I have the slightest chance of making it happen, but that is not going to stop me from pursuing it in whatever way I can.

So I would say, forget about the CC Code and changing the AI significantly.  Focus on what is possible and moderate the game play using whatever resources are available.  If you want that 'better' AI, then it is something that is yet to come, and sadly, may not ever happen.  That's the way it is as I see it.  But do not underestimate the effects that the efforts of good men and true can have with patience and perseverence.  Small steps.  Keep making them.  It's amazing what can be accomplished...

[flick]

To be honest, playing against a Rommel would be great if possible!..... but as I said, I'd settle for an AI that has a basic grasp of direction. I've lost count of the times when I was playing on the defensive, bored out of my mind, while enemy tanks and men, crawled around in circles.

How do other games in the genre manage it? Is the tricky part because we're dealing with real time? Lack of budget?

[platoon_michael]

There is no way the A.I. could become unbeatable or always result in a stalemate.
If you factor in the following for both sides......

Soldier Psyche
Ammo count
Weapon Data
Vehicle Data
Deployment
Support Options


I also assumed there would be the option to turn off the "Learn As" feature just for those days you want to kill and blow things up. :)

Now granted I'm not a programmer but I don't see why it would be that difficult to create.
It could very easily be modeled to factor in the very things we use/do when playing............

Soldier Psyche
Ammo count
Weapon Data
Vehicle Data
Deployment
Support Options
Elements
Terrain (Understanding and use of)
Knowing the strengths and weakness's of troops and vehicles

I'm sure the list could include more but you get the idea.

I'm sure I am missing a lot more but you get the idea.

[vonB]

[Stwa]

Well vonB,

This all sounds like another really beeg project and I know you have already started on beeg project already.

I am hoping you will continue on with the project you have started.

That being said, all I am hoping for, is a CCMT patch, that fixes a few lame bugs that remain. If that can JUST happen, I am good and set for another decade, me thinks.

[vonB]

That's not a big Project Stwa, that's a Mega Project, and waaaaaaay beyond my personal scope to produce.

I am currently focused on CCODM (the Saved Game File Handler) for CC3/CoI and CCOPS (the external Campaign/Operation System), and while I may also be dabbling in some other bits and pieces, these will be keeping me occupied for some time to come!

The AI has always been a particular interest of mine.  I don't know how to get CC into the next generation (my pocket money will not cover the development costs even if I could get the approval from Destineer), but I will continue to keep pushing from whatever angle I can.

CCMT has GOT to have a 'Game' to make it attractive to the majority of CC addicts.  Playing one off Battles is fine in itself, abd perfectly enjoyable too, but it is still a game deficiency.  Possibly in the heyday of CC3 and the Zone, when online activity was massive, CCMT mioght have been very popular as a multi multi players battle game.  I would have relished the prospect.  When we were doing the 5 x 5 game testing, it was a VERY enjoyable experience!

However, for the majority of CC players, Single Player is a must.

[Therion]

From my experience, CCMT was a pretty good game in multi-player after roughly correcting the worst data errors. The main problem was the small amount of players with some of them having problems with connecting.

[Stwa]

Here is the definitive discussion on Multiplayer vs Singleplayer, as far as I am concerned.

Enjoy ...

Multiplayer vs Singleplayer