Are Wargame AIs Fated to Suck?

[Kuokkanen]

ORIGINAL: Ostwindflak

@Matti: I have never actually considered travelling to Finland, but wouldn't rule it out.

If you are ever coming over and need some info, just ask me. And around here are also other guys from over here.

[edit]
And if it is of any interest to you, here are some game events where also war games can be played. Warhammers are popular, my preference is BattleTech, but I don't think historical ones are ruled out.

[aaatoysandmore]

I think it's lack of programming time put into AI's. I feel the game should be built around the AI not the other way around.

Take for instant Battlefront's AI in the Combat Mission series. It's more or less just a charge in your face AI that doesn't take into consideration flanks and feints. Two simple tactics that they couldn't take the time to program a counter or defense for.

Most games I've played though this is a common factor in them. The AI doesn't consider flanks or feints. Just charge like the Total War series AI does. Now, with some "patching" in Rome II they finally did add some flanking maneuvers for the AI but not protecting the flanks or feints.

That's why now I prefer more games of skill like "Conquest Medieval" as the AI doesn't have to worry about flanks and feints, just you breaking it's supply line. It plays a rather good game on a 4 or 5 difficulty setting.

I'd love the Field of Glory AI to be good as that's the kind of ancients game I like. Would prefer a campaign attached to it though.

[sullafelix]

What I am talking about is using the rules against a human or an AI.

When you are playing someone a war boardgame and you notice a flagrant or stupid rule, you just ignore it. That is if you are playing with someone who is a friend or not out just to get a win and game the situation.

There are so few rule in chess etc. that there is no wiggle room. This is not the case with a computer or boardgame that has twenty+ pages of rules, or more.

Many players will try to make a grab at ' victory hexes' at the end of a lot of games. When they can do that without having a line of supply or other troops in the area it is the fault of the designer or programmer.

Seriously, because you are able to capture a capital city of two million or more with a regiment of AA guns etc.. With no other troops around and no supply and are sure to be slaughtered if there is a next turn. That is gaming the system.

You are gaming a historical event. It does not take much to read up on the event and realize what the capabilities of both sides were. In some games the fact that Jubal Early is in a hex with a bridge to Washington, might be a game winner. In reality it meant nothing. I am talking about 1864 here.

[wodin]

Not sure how long ago you played CMx2 game..but since the patch that brought in triggers and aslong as I play a campaign or scenario that's been handmade and not a quick battle the AI is fine by me.

I'd love for you to play a scenario or two a CMx2 game and then post a screenshot of the end of game AAR o prove how easy you find the AI in that game.

ORIGINAL: aaatoysandmore

I think it's lack of programming time put into AI's. I feel the game should be built around the AI not the other way around.

Take for instant Battlefront's AI in the Combat Mission series. It's more or less just a charge in your face AI that doesn't take into consideration flanks and feints. Two simple tactics that they couldn't take the time to program a counter or defense for.

Most games I've played though this is a common factor in them. The AI doesn't consider flanks or feints. Just charge like the Total War series AI does. Now, with some "patching" in Rome II they finally did add some flanking maneuvers for the AI but not protecting the flanks or feints.

That's why now I prefer more games of skill like "Conquest Medieval" as the AI doesn't have to worry about flanks and feints, just you breaking it's supply line. It plays a rather good game on a 4 or 5 difficulty setting.

I'd love the Field of Glory AI to be good as that's the kind of ancients game I like. Would prefer a campaign attached to it though.

[aaatoysandmore]

ORIGINAL: sulla05

What I am talking about is using the rules against a human or an AI.

When you are playing someone a war boardgame and you notice a flagrant or stupid rule, you just ignore it. That is if you are playing with someone who is a friend or not out just to get a win and game the situation.

There are so few rule in chess etc. that there is no wiggle room. This is not the case with a computer or boardgame that has twenty+ pages of rules, or more.

Many players will try to make a grab at ' victory hexes' at the end of a lot of games. When they can do that without having a line of supply or other troops in the area it is the fault of the designer or programmer.

Seriously, because you are able to capture a capital city of two million or more with a regiment of AA guns etc.. With no other troops around and no supply and are sure to be slaughtered if there is a next turn. That is gaming the system.

You are gaming a historical event. It does not take much to read up on the event and realize what the capabilities of both sides were. In some games the fact that Jubal Early is in a hex with a bridge to Washington, might be a game winner. In reality it meant nothing. I am talking about 1864 here.

Kinda like Brazil marching into Berlin in Paradox's HOI? Then on to Moscow. Is that gaming the game or just stupid AI?

[IainMcNeil]

Currently neural nets are nowhere near being able to play even a simple wargame so there is no way to use them or make reusable code. Maybe at some point in the future but there are simply too many possibilities for it to work it out currently. There are just vast - billions of irrelevant moves that can be made that a human dismisses instantaneously that make it very hard for a neural net to evaluate. There are also many options that seems good in the short term but have serious long term consequences or put you in trouble against someone with any experience which a neural net would find almost impossible to see coming, at least now.

[chemkid]

.

[wings7]

ORIGINAL: aaatoysandmore
Kinda like Brazil marching into Berlin in Paradox's HOI? Then on to Moscow. Is that gaming the game or just stupid AI?

It's called freedom of choice on how you want to play HOI...one size does not fit all. [8|]

[aaatoysandmore]

ORIGINAL: Iain McNeil

Currently neural nets are nowhere near being able to play even a simple wargame so there is no way to use them or make reusable code. Maybe at some point in the future but there are simply too many possibilities for it to work it out currently. There are just vast - billions of irrelevant moves that can be made that a human dismisses instantaneously that make it very hard for a neural net to evaluate. There are also many options that seems good in the short term but have serious long term consequences or put you in trouble against someone with any experience which a neural net would find almost impossible to see coming, at least now.

There's no way to make an AI that stores information at least of how the player moves? For instance you couldn't make a program that reminds the AI that the player likes to move to the flanks and do feints pretty often? No way to program it to search for flank movement or patrol for it?

I've always thought that player information should be stored in a data file (we have large enough hard drives for that to happen now). Based on what it has stored when it sees certain movement it triggers the data file to read what the player did before when it made this certain movement. Thus, causing it to at least prepare for a flank attack or rear attack or feint. I see way too many games that ignore this. Even one of my favorite Norbsoft's Civil War series the AI doesn't really look and counter flank attacks very well. It at least does a pretty good job flanking and surprising the player and I find that enjoyable.

Just placing start up units on the maps edges would be an improvement in some games. Not talking about operational games but more on the tactical games.

Teaching the AI to value what it has vs what it might obtain and playing for the victory not just a delaying action for the player to always win. I've seen AI that would have won games had it stayed in place in its bunkers vs going out of defense mode just to "try" to capture another victory hex that the player just liberated that it really didn't need for the victory.

It's the little things I think that would make an AI better. I'm not looking for Big Blue. Just something that plays to "win" not just another delaying action AI in a game.

Give players more options to tamper with the AI. Like you did with Spartan. With a few changes to the build templates I was able to make a very challenging AI opponent on the upper tier difficulty levels. We need more games where each individual player can adjust the AI performance based on his type of play and difficulty he can handle.

[Karri]

ORIGINAL: aaatoysandmore

There's no way to make an AI that stores information at least of how the player moves? For instance you couldn't make a program that reminds the AI that the player likes to move to the flanks and do feints pretty often? No way to program it to search for flank movement or patrol for it?

Sure information can be stored, but it's using the information that's important and currently impossible for computers. Not to mention that you first have to figure out how to even teach the computer what a flank is etc.

Also, if you give the AI all the information about player movements then the AI is cheating...and no one wants that kind of AI. Currently most games, be it wargames or fps or whatever, consists of the AI throwing things at the player until it has nothing left to throw. Making it any better than that requires a lot of resources, new kind of thinking, and of course a game that is made with AI in the mind.

I'd be fine with people adding the auto-resolve option to games that would take into account how I've beaten the AI earlier. I hate to face a bad AI a hundred times in similar battles where it does the same mistakes. For example in the Total War series I would rather have the computer automatically calculate the battles by comparing them with previous battles.

[sIg3b]

ORIGINAL: chemkid


what happens when your opponent's going to change its mind?
read about 'Safely Interruptible Agents'. introducing, the AI Kill Switch Paper.

enjoy! [:)]

Brilliant! [&o] [:)]

I normally have a good BS detector, but with this paper I was unable to decide if it´s satire or not.

[sIg3b]

ORIGINAL: sulla05

What I am talking about is using the rules against a human or an AI.

When you are playing someone a war boardgame and you notice a flagrant or stupid rule, you just ignore it. That is if you are playing with someone who is a friend or not out just to get a win and game the situation.

There are so few rule in chess etc. that there is no wiggle room. This is not the case with a computer or boardgame that has twenty+ pages of rules, or more.

I DO understand what you are saying. I just happen to think it´s an artificial problem.

Take the 3rd Reich boardgame. Some 40 pages rules. You are *expected* to adhere to the rules precisely as written, and to use all the tools you are given. I am sure there are some maneuvers that you would consider counter-intuitive, but if you are *NOT* using them, you are not playing the game as intended.

[sIg3b]

ORIGINAL: wings7

ORIGINAL: aaatoysandmore
Kinda like Brazil marching into Berlin in Paradox's HOI? Then on to Moscow. Is that gaming the game or just stupid AI?

It's called freedom of choice on how you want to play HOI...

Nah, if something like this is possible, that is generally called silly.

[sullafelix]

HOI is a sand box game that doesn't have a lot to do with history. All of Paradox's games are pretty much that way.

The start out with historical people and limits to a point and then you change history.

I was playing HOI3 once and gave directions to invade the SU.

The AI went up through Finland and down into the SU that way. So yes there are questionable AIs.

I am not sure how I can tell if a messed up rule in a wargame is an error or meant to be there?

[Kuokkanen]

ORIGINAL: sulla05

When you are playing someone a war boardgame and you notice a flagrant or stupid rule, you just ignore it. That is if you are playing with someone who is a friend or not out just to get a win and game the situation.

If flagrant or stupid rules get into the game manual, corrections can be written which can be downloaded from designer's or publisher's web site. Unfortunately there are those who don't care about those and insist on playing with flagrant or stupid or otherwise broken/misspelled rules.

Many players will try to make a grab at ' victory hexes' at the end of a lot of games. When they can do that without having a line of supply or other troops in the area it is the fault of the designer or programmer.

I'd like to blame scenario designer for not having enough victory hexes of sufficient value. Large value victory hex on a bridge leading to other victory hexes across the river would solve this problem, or part of it anyway.

Seriously, because you are able to capture a capital city of two million or more with a regiment of AA guns etc.. With no other troops around and no supply and are sure to be slaughtered if there is a next turn. That is gaming the system.

At least in Panzer Corps AA units can't capture cities (can occupy them though). In other games AA units can have sufficient escorts, and I have read that dakka from 40 mm automatic AAA has made enemy crews abandon tanks.

[sIg3b]

ORIGINAL: sulla05

HOI is a sand box game that doesn't have a lot to do with history. All of Paradox's games are pretty much that way.

The start out with historical people and limits to a point and then you change history.

I was playing HOI3 once and gave directions to invade the SU.

The AI went up through Finland and down into the SU that way. So yes there are questionable AIs.

I am not sure how I can tell if a messed up rule in a wargame is an error or meant to be there?

A rule that you *think* is messed up. [:-]

But yes, that´s what "designer´s notes" are for.

For 3rd Reich there was The Wargamer´s Guide to Third Reich. Pretty much explained most of the finer points. HoI, I don´t think there are designer´s notes. Explaining all the nonsense happening in this series would be a challenge, anyway.

[Mobeer]

Flanks:
For many a game though its essential to understand what a flank is:
- for tactical then how to gain attack bonus, or how to hide weak armour etc
- for strategic then how to weaken enemy (or avoid being weakened) by losing supply lines

Some games at least appear to have those abilities, others clearly lack them. Given that some games do have these concepts, this suggests that flanks can be identified as an idea within the AI if the designers can justify putting the effort in.



AI Learning:
One of my frustrations is when an AI repeatedly uses the same failed approach.

For example in Age of Kingdoms (RTS) the AI was limited by fog of war and would actively scout until all territory was observed, after which it would stop scouting. It therefore had no idea what defences exist until battle was joined. To attack it seems to choose the shortest open path the existed when the scouting was done, but then used this over and over with similar unit strength, never learning from failed attacks.

I can't imagine it would be that difficult to have some decision tree that could improve on this, for example:
1) have 3 paths via which attack can be launched
2) launch attack using a random path
3) record total value of damage done and cost of units lost
4) develop new army at cost minimum 10% stronger than previous total cost
5) if last attack value of damage done is more than 30% of cost of units lost then use same path to attack, else use different random path
6) go to 3
then randomise the percentages just a bit as you go along



Re-usability:
In terms of reusable code, I can understand that reusing actual scripts\behaviours is difficult. Still there must be a place for reusing script formats, script editors, behaviour tree implementations etc, but separating out the technology from the data.(?)

[sullafelix]

I think we are playing totally different kind of games.

I never really considered Third Reich a wargame or simulation it is a game that really abstracts war. The same goes for Panzer/Corps general.

I was talking about Drang Nach Osten, Wellington's victory and assorted heavier duty pc and board wargames, really most like this try to be almost simulations.

I try to stay away from beer and pretzels for my brain and my waist.

Yes, a lot of the rules that have been questionable on these games in the last 45+ years have ended up in errata on boardgamegeek or become house rules.

[aaatoysandmore]

I think we've reached the title of this thread: AI's are Fated to Suck.

What developers actually make are "what if's" what if the AI played this or this or this. Nothing is historical but the map, the OOBs and the era time frame. Everything else is just make believe with the AI playing it. Oh you get things like supply rules and radio commands to make it generally seem realistic for you as the player but after that when the AI takes over like someone said above it just becomes a sandbox game. All games basically become "Making History".

That generally wouldn't be so bad if the AI played to actually win. I feel I'm just playing delaying action games. I see the end before the end gets there in almost every game. I almost know the AI isn't going to try to win. Least in Conquest Medieval it doss try and does win from time to time. If they could only incorporate that type of AI into other games.

[sullafelix]

I am not and never said they don't need work, of course they do. I agree completely.

I hate how some game makers use the phrase " multiplayer " as an excuse to not work on the AI at all.