How smart is the AI??

[The CM AI]

Originally posted by Bruno Weiss:

Yeah, and it can't post on this here forum either. How smart do ya gotta be to do that? Mahahah

Who says I can't post to this forum?

Death to you, Bruno Weiss, in our next encounter! MahahahahaHA!

And all of you, listen up. Stop saying these nasty things about me. I can play better than any of you. Computers rule. It's just that, well, HUMANS CHEAT! :mad: :mad: :mad:

[ November 01, 2002, 08:00 PM: Message edited by: The CM AI ]

[Flesh]

Okay, here's my two cents...

IIRC a lot of these issues have already been addressed by BTS.

It seems that the main problem with the AI is that *it has no memory*.

By AI, I mean both the TacAI and the Strategic AI.

A simple TacAI example: one of your tanks targets an enemy tank. The enemy tank then moves behind a building and is therefore out of LOS. Your tank then "forgets" about the enemy tank and stops targeting it. The enemy tank then emerges from behind the building and whacks your tank.

At the strategic level, when the AI is plotting it's moves, it has no memory of anything that has taken place in previous turns. It therefore cannot analyse your moves and determine your plans, as a human opponent can.

Also, it seems that one of the main reasons for the AI's lack of memory is that if the AI did indeed have a memory, PBEM turns would be *huge*. This is because each turn file would have to contain a record of everything that had happened in all of the previous turns. As so many people have 56k modems, this is not currently feasible.

[Bruno Weiss]

Great sod crud! The AI's joined the forum. Likely as not, it'll end up in the cess.

[tracer]

I think I opened a can of worms with a simple question, LOL

[kevinkins]

I would think the "The CM AI" would love to have an upgrade after all these years. Do you have any influence over BTS Mr. AI?

[Battlefront.com]

Hi all,

From my experience, CM's AI is probably the best out there compared to any other wargame of a similar nature. It does a decent job no matter what situation it finds itself in, which is a tall order all on its own considering that there are infinite and HUGE range of possibilities to consider. Think about it... take some Human players and force them to take a different unit mix and they fold up like a house of cards. In CMBO this guy probably only played as the Germans on flat maps with sparse terrain in good weather using Panthers/KTs and Volksgrenadier SMG troops. The AI has no such discrimination luxury. If it is asked to play as Conscript Romanians in open terrain or Finns in a deep snow winter scenario in the woods of Finland... it has to take up the challenge. How many Human players can say they play all things equally well against thousands of different people all the time?

Also, CM's AI doesn't cheat at all. This means that it is, in some ways, less "challenging". Cheating AIs can be quite difficult to beat, but once it is figured out that they cheat it sorta takes all the fun out of it for most players. Especially if the cheat can be identified and countered.

Now... having said all that... do I think CM's AI is as good as I would like it? I think this quote from Foxbat will help frame my answer:

Or any software for that matter, Artificial "Intelligence" has one big, major, unassailable flaw. It can't recognise what it doesn't know.

Bingo. CM's AI can only be as smart as we program it to be. And since AI programming is horribly time consuming, resource hogging, and inherently difficult to do... there is only so good we can make any AI (CM or otherwise). We think we do a better job in less time than others, but obviously there is room for improvement. Hopefully improvement will be very noticable when we rewrite the game engine. But we will just have to wait and see, won't we?

Steve

P.S. AI API programming is totally out of the question. That is, in no small way, a program assignment all on its own. We have neither the resources nor the time to do something like this. Our energy and time is better spent making the game better in other ways, including multi-multi-player.

[kevinkins]

At one time I saw a poll that indicated that many more people spend most of their time playing the AI. Is that still the case? If so then threads like this could develope ideas to improve the AI. Although, since we dont know the inner workings of the engine, this could be a waste of time - posting ideas that could never be coded in. I think the AI does a great job at fire distribution. This is one key thing in a tactical sim. On the move is where it looks at times like a un-trained mob. I would enjoy being part of a beta looking at the AI. In the past I designed my own wargame rules and "AI" for board games. I like this topic a lot.

Kevin

[Battlefront.com]

Hi Kevin,

Although, since we dont know the inner workings of the engine, this could be a waste of time - posting ideas that could never be coded in.

This is actually quite correct. User input and feedback on most any other part of the game is valuable, but in terms of AI... people should save their virtual breath The only person that can come up with practical improvements to the AI is Charles. Nobody else, even me, can offer up anything that will make the AI better. We know very well where the AI's strengths and weaknesses lie, more so than probably any player, but we also know the practical limitations.

There is an old saying... ideas are a dime a dozen. For AI improvement, ideas are worth far less than that.

We will make improvements, but user feedback will not help things at all. Not because we are arrogant, mind you. It is just not practical for Joe Anybody to help things get improved. This is not the case with other aspects of the game so feedback efforts should be directed away from AI and towards things where players have a chance of making a difference.

Steve

[The CM AI]

The boss has said it all: "CM's AI is probably the best out there compared to any other wargame of a similar nature." Hear that? I'M THE BEST! I'M THE BEST!

OK, got that off my chest. Now, listen to the other news: I'm so complex and mysterious you can't possibly begin to understand me. :eek:

Now the final word: "Humans cheat! And I can prove it!" :mad: :mad: :mad:

[kevinkins]

Darn .. that annoying AI

Steve,

Oh well. Thanks for confirming.

With this parting question I will leave the sleeping dogs ...

Has there been any thought at giving the editor

more power to influence the AI?

- Kevin

[ichadwick]

(Just keeping this can of worms open...)

The AI in CMBB isn't all that bad, just limited. It does a turn-by-turn rethink of the situation without any real plan (hence it attacks or counterattacks at inappropriate moments.) You can give it some advantages like extreme fog of war and experience, but this doesn't improve the AI, only its chances of winning.

When you start out, it seems invincible. After a few weeks, it seems weak because you've learned to expect its moves, you remember where it had units and you can plan for likelihoods. The AI can't.

Many games like Ghost Recon offer very rich scenario development tools that include scripting and other tools. It would be nice to be able to tell the AI that it is in attack mode, or defend mode, or set up waypoints for units to move to, or patrol zones. Or to attack/retreat under specific circumstances. But that is another rewrite, a major increase in the effort (and resulting in a greater expense for the next version!). Me, I'd be willing to pay for enhanced scenario tools, but would everyone?

The only answer at the moment is to take on a human opponent. So maybe what we need is a PBEM/TCPIP registry for anyone who wants to play others... or a place to drop a PBEM game where others can pick up the challenge and play the other side that you started.

Ian

[sbg2112]

To make this really work, they'd have to give you much more info about the program than they probably want to. Not that I wouldn't like an CM-OS (Open Source) version, but can you imagine BTS make a living from hotline support only?

I've spent more than a year toying with the idea of writing a proposal to BTS on this very topic. I think there's a huge opportunity to allow us off-duty hacks write all kinds of neat add-ons to CM, without opening up CM one bit. (I think we all understand -- and agree with -- the well-founded reasons for keeping CM "closed", in a sense.)

However, all I've got right now are my notes on the idea; I've just figured if BTS hasn't already thought of it, then it's not worth doing.

[ December 29, 2002, 12:23 PM: Message edited by: sbg2112 ]

[Egbert]

Now that's just too funny.

Ever note Battlefront.com is member #42.

That's just soo preplanned classicly geek, it's scary.

[Silvio Manuel]

Why?

[Terrapin]

Originally posted by Silvio Manuel:

Why?

I think he's referring to Hitchhiker's Guide to the Galaxy.

I think it's a coincidence.

[CombinedArms]

The real reason Battlefront has number 42 is that there was a big crash of the website fairly early on (before my time, anyway)--and when the member numbers were reassigned, they were assigned in alphabetical order. Good thing this outfit isn't named ZoomZoomSoft, or something thing that.

[tracer]

"bump" for the new fellas