Matrix Games Forums

I am looking for people with a background in Artificial Intelligence who would like to critique and otherwise particiapate in the design of the AI Assistant and AI Opponent for MWIF.

Hello,
I've no background in AI, but very interested in taking part in this part of MWIF.
Let me know what I can do for you and I 'd be happy to help.

Alain

I coded "expert systems" in Lisp a long time ago [:(]and I still need some "heuristics" at work time to time. Is this the kind of AI that you envision?

sorry... can't help there.

I'm having enough trouble with my own inteligence at times so I fear I won't be able to help here either. My only experience in this area is to find out that no ai has ever been satisfactory to me...

Marc aka Caran...

One of the (few) things I know about artificial intelligence is that they need some table of values against which to do their stuff eg

• "Gibralter is more important than Cyprus"
• "One day the USA will declare war on Germany"
• "The US needs to conserve carriers in the first year or so of the Pacific War"

While us non-PhD types might not be able to speak with a Lisp, maybe we can start getting together some of those rules for you?

ORIGINAL: Greyshaft
One of the (few) things I know about artificial intelligence is that they need some table of values against which to do their stuff eg

• "Gibralter is more important than Cyprus"
• "One day the USA will declare war on Germany"
• "The US needs to conserve carriers in the first year or so of the Pacific War"

While us non-PhD types might not be able to speak with a Lisp, maybe we can start getting together some of those rules for you?

Yes. Let me give you a context.

I currently have 8 decision makers for the AI Opponent, and as a group they cover all the various sections of the rules. I do not want to go into the details here because there are a lot of details and how they are addressed requires explaining data structures and the language for decision making.

Be that as it may, the decision makers will be familiar to all WIF players: Grand Strategist, Commander in Chief, Manufacturing Council, Foreign Liaison (coordination with allies), Joint Chiefs of Staff, Admiralty, Air Marshall, and Field Marshalls. Note that there may be several Field Marshalls with different ones assigned to different theaters of operation. Each country will have its own set of 8 decision makers; there will not be any grand coordinator of all the major powers for the Axis or Allied sides.

If you want to, you can write up your thoughts on:
(1) one problem one of these decision makers faces in WIF,
(2) what choices they have to choose from,
(3) what information they need to make a decision, and
(4) what criteria they should use to make their decision.

To use your first example (and to only skim the surface of the problem):
(1) The Joint Chiefs of Staff of the Commonwealth should have a stronger defense of Gibraltar than Cyprus.

(2) They need to choose what land and air units to base at these isolated locations to defend against invasion.

(3) Necessary information is: the availability of Axis invading forces, the balance of power in the air (i.e., fighters) in the sea areas neighboring Gibraltar and Cyprus, the balance of power at sea - both naval air and naval surface fleets, and the likelihood of the CW units being put out of supply.

(4) The criteria for making the decision is that the Axis should be denied the opportunity of attacking Gibraltar with 25% chance of success before Cyprus gets any resources. If the Gibraltar attack has less than a 10% chance of success, then Cyprus can recieve excess resources.

You can, and should, be much more detailed in your description of all 4 of these items than I have in the quick and dirty write up I gave in this example. The more details the better.

In selecting a problem to write about I recommend you pick something small to start with. It would be best if it only involved a single unit type (air, land, or naval) and it should be something where you are confident in your analysis. I strongly recommend not starting with something controversial. The idea is to get your feet wet before trying to swim the Atlantic Ocean.

You should choose one of the 8 decision makers and a specific country. That will help you frame the analysis and stay focused on a single problem. There is a very strong temptation to expand a problem and bring in a whole host of related problems. Bad idea. Keep the focus as tight as possible and use specifics (an 8-4 white print infantry) to clarify what you mean.

The Gibraltar example is a very poor choice for your first attempt because it involves multiple unit types. Which is why it would be handled by the Joint Chiefs of Staff.

If you think in terms of Strategic, Operational, and Tactical decisions, you might include that in the heading for the problem. The Gibraltar example is a Operational decision because it helps decide which units are sent to which areas of operation. Rearranging units in a front line to optimize defense would be a Tactical decision. Deciding when to go to war would be a Strategic decision.

WIF has hundreds of these problems (if not thousands). Feel free to give away as many of your own secrets to success in WIF as you like.

I have a (slightly rusty) AI background and was going to help out Robert Crandall on the AI.
What hapened to Robert Crandall, btw ? Is he still part of the MWiF effort ?
Are you going to do the AI, Steve ?

Mille Grazie... in that case I will start with the rules for the Grand Strategist for Italy.
Thinking... thinking... ah... bene

Steve,
Is this the sort of thing you were after? If so then I will expand my points.



WHILE NEUTRAL:
I do need to be at war with the allies at some point during the game.
I do not need to go to war with the Allies immediately (Sep/Oct 1939)
It is politically better that the Allies declare war on me than I declare war on them, however the military benefits of an unexpected declaration of war on the Allies may outweigh the political benefits of waiting for the Allies to declare war on me.
When I am at war with the Allies I will suffer the economic consequences of their blockade.
In 39/40 I can declare war on Yugoslavia/Greece in preference to declaring war on the Allies.
I need to determine which VP cities I will seek to conquer.

WAR STRATEGY:
The French will be defeated and removed from the game so there is no long-term benefit in fighting them if it can be avoided. Let the Germans force the French to surrender.
After the French surrender it is better that my units be employed to assist the Germans than that they sit back performing no function.
(Some of) My units in the Med do serve the function of dissuading the CW from attacking me.
(Some of) My units in the Med can serve the function of conquering the Balkans
Taking Gibraltar is the extent of my primary ambitions in the west.
After Gibraltar I may have some secondary ambitions in the west (eg raider/sub patrols in Atlantic)
Taking the Suez Canal/Cairo/Palestine is the extent of my primary ambitions in the east.
After Suez Canal/Cairo/Palestine I have some secondary ambitions in the east (eg taking Iran/Iraq and raider/sub patrols in Red Sea.)

BATTLE CRITERIA:
I need to conserve my fleet.

PRODUCTION
I need to maintain my fleet.
First cycle Fleet production after 1941 is not recommended unless the Axis is winning (UK conquered?)
Building carriers is not essential. Land Based air can cover the Med.
Supplying units to assist the Germans should trigger a contra of help to conquer the Balkans or North Africa.


PROBLEM: When to declare war on the Allies?
<<to be continued if I am on the right track>>

ORIGINAL: Greyshaft

Steve,
Is this the sort of thing you were after? If so then I will expand my points.

PROBLEM: When to declare war on the Allies?
<<to be continued if I am on the right track>>

Yes, but this is a very large problem (or several problems) that you should examine in pieces.

---------------------------------
DECISION MAKER: Italian Grand Strategist

(1) PROBLEM: Assigning priorities to victory cities

(2) CHOICES: (a) Gibraltar > Suez, (b) Gibraltar = Suez, (c) Suez = 0 until Gibraltar taken, ...

(3) INFORMATION NEEDED: overall strategic plan [or this could be a fundamental part of the overall strategic plan]

(4) CRITERIA: probably none; maybe a randomly chosen strategic plan

---------------------------------
DECISION MAKER: Italian Commander in Chief (who makes decisions about going to war)

(1) PROBLEM: When to declare war on the Allies (i.e., France and Commonwealth)?

(2) CHOICES: (a) stay neutral and do nothing to provoke war, (b) stay neutral with Allies but attack Yugoslavia, (c) stay neutral with Allies but attack Greece, (d) do both b & c, (e) stay neutral with Allies but prepare for attack on France, (f) stay neytral with Allies but prepare for attack on Commonwealth, (g) attack France, (h) attack COmmonwealth, ...?

(3) INFORMATION NEEDED: Victory city priorities, France's land, naval, and air strength on Italian border, France's ability to send more units to Italian front, Commonwealth's land, naval, and air strength in Mediterranean, CW's ability to send more units to Mediterranean, the likely date that Germany will defeat France, ....

(4) CRITERIA: ?

---------------------------------
DECISION MAKER: Italian Manufacturing Council

(1) PROBLEM: Should naval units requiring 2 cycles be started?

(2) CHOICES: (a) yes, (b) no

(3) INFORMATION NEEDED: current date, is Axis winning, is CW conquered?, ...

(4) CRITERIA:

----------------------------------

I hope this quasi-example helps you see how I think about these things from the AI's point of view. The problem should be broken into pieces so the AI can look at it from the point of view of one decision maker (DM). The DM might have to rely on the other DMs providing information and/or actually carryng out the decisions. For instance, the Commader in Chief might decide that the Italian navy is inadequate for going to war so he assigns a higher priority to producing naval units. The Manufacturing Council then uses that higher priority in their decision making.

When humans make decisions, they range all over the place, as is clearly demonstrated in your post. We know there are a lot of different aspects that have to be considered and jump around pulling bits and pieces of information together, trying hard to not miss something that is important. After we get all the relevant facts 'gathered' (in the forefront of out thoughts), we them latch onto those that are major factors and use them to narrow down the choices. Once that is done, we try to use all the facts to make the final choice between (a) and (b). Of course I am speaking in generalities here and there could easily be as many exceptions as there are cases where this sequence of events unfolds as I have described.

For the AI Opponent, I need to neatly compartmentalized all decisions and then build the linkages between the pieces. It is not a common thought process for people and strikes most of us as a downright weird way of thinking.

ORIGINAL: Incy
I have a (slightly rusty) AI background and was going to help out Robert Crandall on the AI.
What hapened to Robert Crandall, btw ? Is he still part of the MWiF effort ?
Are you going to do the AI, Steve ?

I have been trying to reach you via email but it appears that Rob Crandall's old email address for you no longer functions. Please send me a personal message (PM) with your current email address so we can discuss things off-line.

I have some level of experience with Ai-programming, once coded an AI for (Reversi, Othello) using minimaxi algorithm with added heuristics.

When at university I attended a couple of PHD level AI courses, thats many years ago though.

ORIGINAL: Shannon V. OKeets

I am looking for people with a background in Artificial Intelligence who would like to critique and otherwise particiapate in the design of the AI Assistant and AI Opponent for MWIF.

DECISION MAKER: Chineese Fieldmarchal
(1) PROBLEM: Command frontline in southern China
(2) CHOICES: (a) withdraw line to defensive line at mountainhexes and Kwei-Yang, (b) maintain current line (c) attack japaneese hex with 6-3 units (d) Attack Canton
(3) INFORMATION NEEDED:
- Weather
- estimated number of impulses left this turn
- prioties on conserve own troops inflict enemy losses.
- Priority on preserve troops/preserve territory/gain territory
- Predicted chance that japan makes an attack on his pulse
- Predicted troop/land losses on japaneese attack on most vulnerable hex/hexes
- Predicted japaneese/Chineese losses on attack on Canton/hex NE Canton

(4) CRITERIA:
If goal is preserve own troops & predicted chineese losses on japaneese attack >1 corps then withdraw line

If goal is preserve own troops & predicted chineese losses on japaneese attack <=1 corps & predicted Japaneese losses > 1 corp then maintainline

If goal is to gain land & predicted chineese losses on attack on Canton <=1 corps & chance of taking hex >45% Attack Canton.

etc.

ORIGINAL: c92nichj
DECISION MAKER: Chineese Fieldmarchal
(1) PROBLEM: Command frontline in southern China
(2) CHOICES: (a) withdraw line to defensive line at mountainhexes and Kwei-Yang, (b) maintain current line (c) attack japaneese hex with 6-3 units (d) Attack Canton
(3) INFORMATION NEEDED:
- Weather
- estimated number of impulses left this turn
- prioties on conserve own troops inflict enemy losses.
- Priority on preserve troops/preserve territory/gain territory
- Predicted chance that japan makes an attack on his pulse
- Predicted troop/land losses on japaneese attack on most vulnerable hex/hexes
- Predicted japaneese/Chineese losses on attack on Canton/hex NE Canton

(4) CRITERIA:
If goal is preserve own troops & predicted chineese losses on japaneese attack >1 corps then withdraw line

If goal is preserve own troops & predicted chineese losses on japaneese attack <=1 corps & predicted Japaneese losses > 1 corp then maintainline

If goal is to gain land & predicted chineese losses on attack on Canton <=1 corps & chance of taking hex >45% Attack Canton.

etc.

Nice.

This leads to the followng tasks:
(1) estimate the number of impulses left in the turn;
(2) define a metric (scale) for "conserve own troops versus inflict losses on the enemy";
(3) define a metric (scale) for "conserve own troops versus gain territory";
(4) estimate the probability that Japan will attack China in the next impulse; and
(5) estimate our (Chinese) expected losses for a Japanese attack on each of our hexes.

If we can come up with a way to answer to #5, then we can use the same system for any land attack anywhere in the world. We might have to do a dozen or so well defined situations (German-USSR frontline in Barbarossa), (Maginot line in 1940) before we have a methodology that works everywhere. This is the way I like to develop 'rules': start with specific cases, refine them until they are as close to perfect as possible, do a bunch more specific cases, and finally, develop a general methodology based on all the cases.

INFORMATION NEEDED :
How many land actions can I afford in this turn?
Can I execute the choosen plan with the land movements of combined actions?
Even in 1938, there are corps to ship oversea, fleets to redeploy to new bases and you must carefully weight the importance of a front against others when choosing actions. Just my 2 cents, but I'd add "relative importance of the South China front" and "number of land movements and attacks required by each posture" to the informations required.
In fact, in my opinion, at the beginning of the turn, each "front" should report the type, importance for them (I can wait for the rebase but I must move that corps NOW) and number of activities they would need. Then the GHQ would prepare a "queue" of action types and "allocate" one each impulse depending on the relative importance of the fronts: in 1942, typically for Japan, the first impulses would be naval or combined and "China" would get a land action only if the turn lasted long enough. BUT, China should get 1 or 2 land movements early if it prevents a disaster so a naval action could have changed to a combined in the first impulse. Here, I'm not only talking of choosing the action type but also of allocating the activities: a rebase in North Africa can be more important than a ground strike in Ukraine but if you just blindly use the air actions as long as they are available, it may never be done.
Now you can't all base on your own plans, you must time to time react to the enemy's [:)] so there should be a provision for unexpected change of plans. And as I'm writing this, I wonder if it wouldn't be better to chose the action types "on the fly" at each impulse. Less efficiency but better ability to adapt to the circumstances. An ideal AI would have both : a country with the initiative, such as Germany 40, would use the "pre-planned method" while the same Germany in 1944 is often only able to react to the Allies last move.

ORIGINAL: Rotor
INFORMATION NEEDED :
How many land actions can I afford in this turn?
Can I execute the choosen plan with the land movements of combined actions?
Even in 1938, there are corps to ship oversea, fleets to redeploy to new bases and you must carefully weight the importance of a front against others when choosing actions. Just my 2 cents, but I'd add "relative importance of the South China front" and "number of land movements and attacks required by each posture" to the informations required.
In fact, in my opinion, at the beginning of the turn, each "front" should report the type, importance for them (I can wait for the rebase but I must move that corps NOW) and number of activities they would need. Then the GHQ would prepare a "queue" of action types and "allocate" one each impulse depending on the relative importance of the fronts: in 1942, typically for Japan, the first impulses would be naval or combined and "China" would get a land action only if the turn lasted long enough. BUT, China should get 1 or 2 land movements early if it prevents a disaster so a naval action could have changed to a combined in the first impulse. Here, I'm not only talking of choosing the action type but also of allocating the activities: a rebase in North Africa can be more important than a ground strike in Ukraine but if you just blindly use the air actions as long as they are available, it may never be done.
Now you can't all base on your own plans, you must time to time react to the enemy's [:)] so there should be a provision for unexpected change of plans. And as I'm writing this, I wonder if it wouldn't be better to chose the action types "on the fly" at each impulse. Less efficiency but better ability to adapt to the circumstances. An ideal AI would have both : a country with the initiative, such as Germany 40, would use the "pre-planned method" while the same Germany in 1944 is often only able to react to the Allies last move.

Your points are all valid but they would be easier to understand if they were separated and given more structure. So, I have reformatted them below. I do this to help communication. If I made any mistakes and changed your meaning, I apologize and you should let me know.

SJH paraphrase:
---------------------
DECISION MAKER: Joint Chiefs of Staff (Japan)
PROBLEM: Choosing Action Type
CHOICES: Land, Air, Naval, Combined, Pass
INFORMATION NEEDED: year/month/impulse, weather, probability of end of turn in 1, 2, 3, ... impulses, risk of suffering defeat in an opponent's land attack during opponent's next impulse - if Naval Air, or Pass Action taken (no land moves/attacks allocated) / if Combined Action is taken (limited #) / if Land Action is taken (unlimited number), importance of South China Front

---
DECISION MAKER: Joint Chiefs of Staff (all countries)
PROBLEM: Choosing Action Type
CHOICES: Land, Air, Naval, Combined, Pass
INFORMATION NEEDED: from each Field Marshall: importance of getting 1, 2, 3, .. rebases / land moves / land attacks / air missions

---
At the beginning of each turn the Joint Chiefs of Staff (JCS) would gather information from the Field Marshalls and use it to plan the Action Types to be chosen on the 1st, 2nd, ... impulses. This plan would be reexamined each impulse and revised if necessary.
-----------------------

SJH comments:
I would measure the importance of a front in absolute terms and then let the JCS compare them to get a relative importance. The absolute importance of each front would be set by the Grand Strategist.

The field Marshalls should also be accessing and reporting to the JCS on attack opportunities (ripe fruit to be picked). Indeed, the Admiralty and Air Marshall should also be reporting risks from enemy actions and opportunities for attacks / moves.

DECISION MAKER: Admiralty (any country)
PROBLEM: Choosing Task Force composition
CHOICES: Which ships to affect to a given mission
INFORMATION NEEDED: nature of the mission (port attack, raiding,..), target seazone or port, list of assets that the enemy can use against the task force : land based aviation/carriers/scs/submarines, weather, importance of the mission, list of ships not already selected for another task, optionally starting harbour, in other cases, the JCS could leave the admiralty the choice between, say, Gibraltar and Alexandria to select the best suited ships for the mission.

ORIGINAL: Rotor
DECISION MAKER: Admiralty (any country)
PROBLEM: Choosing Task Force composition
CHOICES: Which ships to affect to a given mission
INFORMATION NEEDED: nature of the mission (port attack, raiding,..), target seazone or port, list of assets that the enemy can use against the task force : land based aviation/carriers/scs/submarines, weather, importance of the mission, list of ships not already selected for another task, optionally starting harbour, in other cases, the JCS could leave the admiralty the choice between, say, Gibraltar and Alexandria to select the best suited ships for the mission.

Ok.

Now, you could take one of the missions (say, port attack) and figure out the criteria for setting the #of planes, air-to-sea factors, and # of fighters to send. Because the problem was task force composition, I assume we are not automatically sending everyone. How do we use the information to make the decision?

DECISION MAKER: Field Marshalls (any country)
a case can be made for letting the JCS deciding globally of the necessary reorganizations
PROBLEM: Reorganize
CHOICES: Which units to reorganize with wich hq
INFORMATION NEEDED: expected remaining length of the turn, expected weather (reorg planes of not), which units could effectively be reorganized, what are we going to do in the coming impulses : advancing, retreating, heavy fighting.., do we expect a situation where we would need the hq special abilities (emergency supply & offensive/defensive support irc), would inverting the hq put it in danger (case of retreat)..
imho it will be the toughest part to do right. Grand strategy looks easy in comparison. But maybe it's just me...