A WW2 Stratagic Game

[Dan Cook]

Originally posted by laurent Favre:
Dan,

Even if I guess chances to see this game completed are around 0%, I will post on the open source wargame forum of the Wargamer board. A bottle in the sea... and a motive for me to learn C++. So maybe around 2015...

2015? By that time, we'll be able to recreat WWII with VR goggles!

-Dan

------------------
Spam.....Is there anything it can't do?

[bpolarsk]

Originally posted by Mogami:
quote:You can get computers rated as high as 2100 but that is a blitz rating most pc programs perform at about 1600-1800 rating. Most under 10 year old humans are 1100 most humans 1400 so computers look good intill they meet real chess players.

Situation has changed since many years. Give a Try to Arasan 5.3. it is free and is rated ELO +2200 at 2h game length.
http://www.best.com/~jdart/arasan.htm

There are lot's of excellent free chess game now. Check them at :
http://www.enpassant.dk/chess/softeng.htm

They are all free.

Bernie

[sapperland]

Originally posted by bpolarsk:
Ouch ! Exact number is 2**64, Even for a computer it is unlimited. There is no difference between Unlimlited and 10 time unlimited. The difference between chess AI and wargame AI is 20 years of AI developping on the SAME game with SAME rule. I don't think than any company will ever produce alone a good AI. However, if you compare on what happened with chess, the solution is to have AI as an external open module, standartized within an expert system with thousands of rules (forget heuristic model). This model must be external to the game (rules of game are coded) and the model is open source so that it can be modified and reworked by many people in time.
There is an attempt to create such model at "http://openwarsim.org/". As far as I know it is the only one on the Web which is about wargeme and is active. We try to build rules and then attract people from Gameai.com to build various engine to compete each other. but you will need patience.

Bernard

Lets do some math:

Pawns 8x2 =16
Queen 1x27=27
King 1x8 = 8
Knight 2x8 =16
Bishop 2x13=26
Rook 2x14=28

SUM equals 121
This is (in theory) the most number of moves a side can choose from in one turn. In practice the number is more like 40 since not all pieces can move their max due to location and other pieces. Also as pieces get spread out some are eliminated.
Now to forecast future moves you will put 40 to the power of the number of moves you want to predict. i.e. to forcast the next 3 moves for yourself and the other side you get 40*6
which is 4 billion. This number for computers is not a big deal.

Now compare that to a war game inwhich you have say 100 units and you can move all none or some. The key here is you can move more than one unit a turn. And the average distance the units can move is 2. Now you get (100*100)x2. This is 2*200. Now forecast that for 3 turns for yourself and the other side.
you get 2*200*6 which is 2*64000000000000.

You now see that you are getting a very BIG number here.


QED
Steve

[GulFalco]

I hope not much time is wasted on developing better solo game AI, even though I have played considerably more by myself than with others,but with the high speed connections that are now available this is becoming less and less true. What I want is AI that can interpret general odrers issued by a player into the more specifics the game requires.

------------------
Men are cheap, but women can never be payed for. Napoleon

[bpolarsk]

Originally posted by sapperland:

Now to forecast future moves you will put 40 to the power of the number of moves you want to predict. i.e. to forcast the next 3 moves for yourself and the other side you get 40*6
which is 4 billion. This number for computers is not a big deal.

Now compare that to a war game inwhich you have say 100 units and you can move all none or some. The key here is you can move more than one unit a turn. And the average distance the units can move is 2. Now you get (100*100)x2. This is 2*200. Now forecast that for 3 turns for yourself and the other side.
you get 2*200*6 which is 2*64000000000000.

This is quite optimistic view. In chess you must be absolutely precise in a plan, while wargame are more statisticals than absolutely precise. I am chess player with 2230 ELO and I do plan usually of 6 moves in advance and sometime more. Actuall computer goes frequently to 15 moves, so review your nice figures, as the reality is much painful. Also after each move you re-evaluate the while position, it is not just forecasting the physical move. In Wargame AI you play with targets and logical group and subgroup assigned and limited to the target. In this respect the notion of each individual move balancing the fate of the whole game is not so true as in Chess. If it were not true, then following your own figures, they will not any wargame AI on the market.

Bernard

[sapperland]

I was using the number 3 for ease. I have played chess from the age of 8. So lets just say almost 3 decades. I know that most good players forecast much more. I find 6 - 7 moves to be max for me.
AI's for chess can and do check each and every move that can be made. This is because the total number is somthing that can be handled. 3 turns, ie 4 billion, may take several seconds, going 6 turns out could take an hour depending on the set up. Unless you have a super computer that can do trillons per second, you will have a long game on your hands.
I just wanted to point out that making an AI perfect for war games can not work since the number of calculations would be so large. So AI's for war games must be done on "if then else" commands. Basically, the programer must write all the AI moves in advance. I for one would not even hope to play a winning game against another human with that handicap.

[bpolarsk]

Originally posted by sapperland:

I just wanted to point out that making an AI perfect for war games can not work since the number of calculations would be so large. So AI's for war games must be done on "if then else" commands. Basically, the programer must write all the AI moves in advance.

Other AI have faced and solved this problem, ie : the GO. They are quite good now, competing for world title. I agree than an heuristic AI on wargame is not fot the near future so that the only way to go is building an expert system. But as long as companies will build AI devoted to their own game, starting from scratch at each serie, then the level will be quite weak. The strengh of an AI is closely dependant of the number of years of work put in it. I still hold on my opinion, that companies could derive great benefit if they build they AI externally open source to enable third parties to improve. I definely suspect than most wargames obey to a certain amount of limited rules (and exception to rules) and that it is possible to translate the map and the specificity of the game in order to match the internal mechanism of a generic engine.

Bernard

[JJU57]

I'll concentrate on the AI for 3R since most agree that it was pretty bad. My question was why? First off the opening moves are know much like chess. I found the placement of units in the game to be correct. However, the next setup was where the computer started to fail.

The game is actually pretty good at exploiting errors and using the computer pieces to surrond and cutoff extended units. The computer generally failed at a robust defense and in strategic analysis. And it really shouldn't have.

The key to the game is simply looking at the placement of units on the map. Let's say your playing England. The start is failry straight forward. Defend Africa and France. Prevent an invasion.

The program should have simply weighed the placement of the Axis fleets and airpower along with the paratroop unit. If they were on the east front then no invasion is planned. Therefore more British fleets and air should be in Africa to fight there. A simple check should be made to see if a quich attack in NW Germany can get to berlin through exploitaion. Otherwise it is farily straight forward.

As for defense in Russia the computer needs to give ground to form a two deep line. Don't fight where you stand. This also should have been failry easy to program.

The hardest part for any AI is co-ordinated attacks over multiple turns. However, if the AI for 3R did the simple things then it would have been a much better game.

One final point about the difference between chess and games. Chess has well defined runs on 'combat'. Games use probability and even a 5-1 attack just may lose. This makes it harder but not impossible.

[Grumble]

"I'm not sure about boardgames, unless you're talking about "monster games", say, AH's "Longest Day", the whole damn Normandy invasion at battalion level or some such crazyness. When speaking of boardgames it comes down to the scale and scope of the game. As for computer games the reason is simple. AIs are *stupid*. Even low level AIs meant to do the grunt work freeing the player to handle high level issues. I've never found a good one in wargames or strategy games, and boardgames of course can't rely on a low level AI, so the players inevitably end up doing more than is realistic for their position, like moving all units themselves..."
Well "Grand Europa" is another one, as well as WiF (an excellent game). WiF as case in point, the expansion counters go down to the battalion-level. That's got nothing to do with AI or directing a nation's war effort(except as a factor in game design), everything to do with demand driving the market. Without a gamer's desire to move around hundreds of counters, whether cardboard or pixels, the market wouldn't be producing them.

[Ed Cogburn]

"I'm not sure about boardgames, unless you're talking about "monster games", say, AH's "Longest Day", the whole damn Normandy invasion at battalion level or some such crazyness. When speaking of boardgames it comes down to the scale and scope of the game. As for computer games the reason is simple. AIs are *stupid*. Even low level AIs meant to do the grunt work freeing the player to handle high level issues. I've never found a good one in wargames or strategy games, and boardgames of course can't rely on a low level AI, so the players inevitably end up doing more than is realistic for their position, like moving all units themselves..."

I'm confused. I said this a long while ago. Are you responding to me?

Originally posted by Grumble:

Well "Grand Europa" is another one, as well as WiF (an excellent game). WiF as case in point, the expansion counters go down to the battalion-level. That's got nothing to do with AI or directing a nation's war effort(except as a factor in game design), everything to do with demand driving the market. Without a gamer's desire to move around hundreds of counters, whether cardboard or pixels, the market wouldn't be producing them.

I agree there is a large segment of the wargaming community that prefer games where they make *all* the decisions. Micromanagement run amok, true. However this does not change the fact that there is *no* alternative right now. AIs cannot be built currently that would satisfy the moderate wargamers, much less the I-want-to-do-it-all monster game wargamers. The former might except a level of basic competence, i.e. mediocrity, on the part of the AI, but AIs now are completely incompetent, and the latter kind of wargamer will only except an AI equivalent to their own competency.

[JJU57]

I don't understand why there aren't good AI's for games like Axis & Allies. There is a limited number of unit types and a limited number of important areas on the map. The logic tree would not have to be more then 100 levels deep. After all just analyze your thought process. You first look at the map and see where the enemy is and where you are. Then you develope a strategy for this. I think most AI programmers don't take into account the placement of enemy forces.

In chess each piece has a value attached to it and this value is modified slightly by it's location on the board. A knight trapped on the corner is worth less then one in the middle. Wargame AI's don't do this. I don't think they even take into consideration the value of units based on the overall strategy of offense or defense. Why? It may be some boring programming but it should would make a hugh difference.

After all optimization routines for large databases are more complex then some of these game AI's. Of course the database cost tens or hundreds of thousands of dollars.

[Grumble]

Hi,
I've been a gone a while, so yes I was responding.
"I agree there is a large segment of the wargaming community that prefer games where they make *all* the decisions. Micromanagement run amok, true. However this does not change the fact that there is *no* alternative right now. AIs cannot be built currently that would satisfy the moderate wargamers, much less the I-want-to-do-it-all monster game wargamers. The former might except a level of basic competence, i.e. mediocrity, on the part of the AI, but AIs now are completely incompetent, and the latter kind of wargamer will only except an AI equivalent to their own competency."

Boardgame, computer game it doesn't matter. Granted, AI's will never approach the ingeunity of a human opponent. But the AI has nothing to do with the scope of game design; one designs a workable game system then works on an AI to operate within that system. The alternative is to realize that the SCOPE of a strategic game should be on strategy, political and economic decision making, NOT whether this battalion is going to land here, or this squadron is performing CAS instead of Interdiction. Otherwise it's NOT a strategic game by definition; sort of a super Operational-level game and a bad simulation of strategic warfare.