AI for MWiF - USA

[Shannon V. OKeets]

ORIGINAL: Froonp
Coordination between allies is one of the keys to victory in WiF.
I talk about operational coordination, not only strategical.
A few examples of operational coordination are (speaking of WiF, not the true war) :
1) Transporting CW troops with US TRS / AMPH so that the US has a naval action and the CW a land action (or the reverse).
2) Transporting CW PARA with US ATR, thus savng the precious CW air missions of a CW land action.
3) Carpet bombing & Ground Striking with the CW bombers on the German European 44 line under an Air Offensive Chit while the US under a land offensive chit is positionned to assault & blitze the positions struck bu the CW. This is crucial. Without this, the Western allies are like a man with only one leg.
4) Sailing German ships and / or SUBs (and leaving them at sea eventually too) with a few Italian SUBS so that the Italians can initiate the searches while chooisng combined actions, and the German can focus on land actions.

Those 4 are only from the top of my head, but tactical cooperation is a must in playing some countries in WiF.

I think this can be achieved by the loaning feature of CWiF.
In 1) the US TRS / AMPH are loaned to the CW for this impulse, in 2) the US ATR are loaned to the CW for this impulse, in 4), the German ships are loaned to the Italian for this impulse & more.

Only for 3) I do not have a ready solution in WiF with a human and AI countries played as a team.

Cheers.

This is all excellent. Thanks.

[Froonp]

3) Carpet bombing & Ground Striking with the CW bombers on the German European 44 line under an Air Offensive Chit while the US under a land offensive chit is positionned to assault & blitze the positions struck bu the CW. This is crucial. Without this, the Western allies are like a man with only one leg.

(...)

Only for 3) I do not have a ready solution in WiF with a human and AI countries played as a team.

Maybe there could be some sort of "Demands to the AI" dialog for those special US/CW and Italian/German close tactical cooperation (there's also the mighty Italian Air / German Land in Russia).

Imagine the following :
USA : Human (could be German).
CW : AI (could be Italian).

(Could be the reverse too)

The USA could have a special dialog in which they would say : "Listen my old friend Winnie, I would like your air force to wreak havoc into this zone, killing units and disrupting units".
- The "zone" could be definited as an hex with a range around which the zone is. For instance it could be Paris & 3 hexes around.
- The "wreak havoc" could be scaled up or down from a full scale air attack (the least powerful) to a ground zero air attack (the most powerful), even if there are only 2 steps in this scale, so that the AI know is the player would like an Air Offensive Chit (OC) to be used or not.
Ideas for the scale :
Step 1 (weakest) : Air attack on the most likely targets (for the following land attack by the human ally) & assets (assets are HQ, ART adjacents to the hexes likely to be attacked). The AI could fulfill this even with a simple combined action depending on its own air mission needs.
Step 2 : Air attack most of the enemy units, targets, assets, and replacements (replacements are units that could move to block the way to an exploit after a successful breakthrough). This would need an air action, and could lead to 7-10 ground strikes and 2-3 25-strengh carpet bombings (more with OC).
Step 3 : Same as 1, but with an OC.
Step 4 : Same as 2 but with an OC.
- The "killing units and disrupting units" would ba a choice by the requesting player. If only "disrupting", the AI should understand that only ground strikes are needed. If "killing" is demanded, the AI should understand that it should mount some 25 strengh Carpet Bomb missions on especially juicy enemy targets.

The AI would then decide if it can use an OC for that, decide the most likely targets considering the enemy fighter force and its own, decide how many Carpets mission it runs considering its own carpet bombing power and the availability of juicy targets (HQ not in forest, Armor concentrations, powerful hex to be breached, etc...).

This AI request dialog should also have an estimation of the number of turns in the future when this offensive will be carried, so that it can prepare itself by advancing the necessary offensive chitters HQ and the bombers.

I renew what I said before, this is crucial to break the Germans in 44 & 45 to be able to use this cooperative tactic, wreaking total & devastative havoc with CW planes and blitzing through the remnants with an US land OC on Eisenhower (or Bradley or Hodges).

[Froonp]

I think this can be achieved by the loaning feature of CWiF.
In 1) the US TRS / AMPH are loaned to the CW for this impulse, in 2) the US ATR are loaned to the CW for this impulse, in 4), the German ships are loaned to the Italian for this impulse & more.

It will be important to document the "Loan" feature of MWiF (which is completely stranger to WiF) in the MWiF documentation, and to illustrate it with those kinds of examples.

[scout1]

This raises another thought. I'm assuming there is strategic bombing (of one sort or another). In RL, there was much debate within the services as to the use of strategic air power. In fact, there was a big argument about Ike "borrowing" the 8th Air Force to prep for D-Day. A not so pretty discussion with the flyboys.

[Shannon V. OKeets]

Patrice,

The mechanism I will use for communicating importance between the AIO's is the value in terms of Combat-Strength-Value (CVs) of a land unit. This is what each AIO will use internally to evaluate between land, air, and naval units/activities/actions. So the CW AIO could pass along the inofrmation to the USA AIO that a ground strike is worth so many CVs. Actually what it passes on is a distribution that provides what the 1st , 2nd, 3rd, ... ground strikes are worth. Likewise for ground support and strategic bombing missions. What the allied AIO has to decide is whether to treat doing things for his ally just as if he were doing it for himself (CW CV = USA CV) or to impose some sort of penalty (say, 0.5 CV or 2.4. CV - whatever).

How to apply this to an AIA and human player is unclear, but a simple translation of English words (helpful, important, crucial) into CVs is one possibilitiy.

Your point about longer range planning is something I hadn't thought about, but in retrospect is obvious. I'll have to think about that.

[Shannon V. OKeets]

ORIGINAL: Froonp

I think this can be achieved by the loaning feature of CWiF.
In 1) the US TRS / AMPH are loaned to the CW for this impulse, in 2) the US ATR are loaned to the CW for this impulse, in 4), the German ships are loaned to the Italian for this impulse & more.

It will be important to document the "Loan" feature of MWiF (which is completely stranger to WiF) in the MWiF documentation, and to illustrate it with those kinds of examples.

Ah, something for the Help system and Tutorials. Thanks. When I asked for advice and suggestions for these two items back in September, there was only a splattering of responses. I have left off finishing their final design document because I believe it needs more than what I have gathered so far.

[Froonp]

How to apply this to an AIA and human player is unclear, but a simple translation of English words (helpful, important, crucial) into CVs is one possibilitiy.

I think that the player should be able to request things from the AIO mainly.
The reverse is somehow less true, because I'm talking about those situations where a close teamwork is needed (US & CW in western Europe & Italy, Germany & Italy in Italy and to a lesser extend Germany & Italy in Russia). In those situations, there is usually (in the real game) a sort of team leader that decides of the long term thrusts & offensives who are carried, and the other team member provides the assets to help the strategy being carried. He also carries his own needs & wishes, but it is best if the team work is as close as possible. Both can't try to do opposite things without destroying their hopes of achieveing something. Both must be close allies, and one of them has to decide, and I think that between the AIO and the player, the Player should decide.
Hence, that's why I think the player should ba able to request things from the AIO, and the reverse is less true because the AIO should be somehow "following".

[Froonp]

Another thing comes to my mind related to what I wrote in the previous post.
It is about close cooperation between allies, when one of these allies is the AIO. When it is a human it is irrelevant completely because players communicate using human means.

There is a need, at least in my conception of WiF play, for the close cooperating allies (close cooperating allies are countries which not only cooperate in the sense of the WiF rules, but also need to work as a close team -- nearly as one country -- on a given theater) to be able to have a plan (a deal) for sharing the possible objectives cities that could be conquered by the team between them.

For example, in a normal WiF real game, it is common to have the CW army mainly positionned on one wing of the Allies frontline, and the US army mainly positionned on the other wing. I write "mainly", because their armies are often deeply intermixed, with assets of one in the middle of the army of the other (i.e. for using the Air OC / Land OC ploy this is crucial to have an HQ amongst your ally army).

When this situation arise the players will often have deals to "share" the objective cities between them, and if one unit of your ally could conquer one city that would belong to your deal, it normaly avoid to conquer it and it leave it to you (provided the said objective city is not at risk of being retaken, and that you can take it immediately too).

I think that there is a need for the AIO to be able to take such deals with the player, and to stick or betray it as they see fit. The need is indeed for the player to be able to have such deals.

i.e. CW player "talking" to the US AIO : "OK, Italy & France is already completely yours (Paris, Marseille, Milan, Rome), my army is on the left wing in France, close to the sea, I would like Amsterdam, Antwerp & Kiel to be mine, and I leave to you Munich. Berlin will be for the first comer."

Do you think such communication will be possible ?

[JanSorensen]

To be honest I dont think the AI should bother with the intricacies of individual victory. In all likelihood doing so will only weaken its ability to play a good team game by making it take actions that are tactically or even strategically unsound (just like it does for not so few humans).

For that reason I think it would be an error to let the AI go down that road. Forget about the diplomatic aspect and concentrate on the military one instead as those are at odds with the latter being several orders of magnitudes more important.

WiF is not Diplomacy afterall.

[Shannon V. OKeets]

ORIGINAL: Froonp

How to apply this to an AIA and human player is unclear, but a simple translation of English words (helpful, important, crucial) into CVs is one possibilitiy.

I think that the player should be able to request things from the AIO mainly.
The reverse is somehow less true, because I'm talking about those situations where a close teamwork is needed (US & CW in western Europe & Italy, Germany & Italy in Italy and to a lesser extend Germany & Italy in Russia). In those situations, there is usually (in the real game) a sort of team leader that decides of the long term thrusts & offensives who are carried, and the other team member provides the assets to help the strategy being carried. He also carries his own needs & wishes, but it is best if the team work is as close as possible. Both can't try to do opposite things without destroying their hopes of achieveing something. Both must be close allies, and one of them has to decide, and I think that between the AIO and the player, the Player should decide.
Hence, that's why I think the player should ba able to request things from the AIO, and the reverse is less true because the AIO should be somehow "following".

Yes, to all of the above. A good point.

Please try to keep the distinction between the AI Opponent (AIO) and the AI Assistant (AIA). In the preceeding post you are talking about the AIA, because it is playing on your side.

[SurrenderMonkey]

I tend to agree with Jan. One of the simplest ways to fail at software development is to attempt too much. Have the AI's focus on the military situation, and force a human player in an AI-human alliance to deal with the AI's focus as a kind of diplomatic problem in itself.

Otherwise, I can hear the endless complaints coming already ...

[Shannon V. OKeets]

ORIGINAL: Froonp

Another thing comes to my mind related to what I wrote in the previous post.
It is about close cooperation between allies, when one of these allies is the AIO. When it is a human it is irrelevant completely because players communicate using human means.

There is a need, at least in my conception of WiF play, for the close cooperating allies (close cooperating allies are countries which not only cooperate in the sense of the WiF rules, but also need to work as a close team -- nearly as one country -- on a given theater) to be able to have a plan (a deal) for sharing the possible objectives cities that could be conquered by the team between them.

For example, in a normal WiF real game, it is common to have the CW army mainly positionned on one wing of the Allies frontline, and the US army mainly positionned on the other wing. I write "mainly", because their armies are often deeply intermixed, with assets of one in the middle of the army of the other (i.e. for using the Air OC / Land OC ploy this is crucial to have an HQ amongst your ally army).

When this situation arise the players will often have deals to "share" the objective cities between them, and if one unit of your ally could conquer one city that would belong to your deal, it normaly avoid to conquer it and it leave it to you (provided the said objective city is not at risk of being retaken, and that you can take it immediately too).

I think that there is a need for the AIO to be able to take such deals with the player, and to stick or betray it as they see fit. The need is indeed for the player to be able to have such deals.

i.e. CW player "talking" to the US AIO : "OK, Italy & France is already completely yours (Paris, Marseille, Milan, Rome), my army is on the left wing in France, close to the sea, I would like Amsterdam, Antwerp & Kiel to be mine, and I leave to you Munich. Berlin will be for the first comer."

Do you think such communication will be possible ?

Here I think that you, as the human player, essentailly can take credit for all the victory cities held by both you abd the AIA at the end of the game.

The stuff that happens over the board between allies in WIF as to who takes which city always stuck me as artificial. Sort of like whether I pay for a dinner out or my wife does. We have an common checking account, so what does it matter.

[Shannon V. OKeets]

ORIGINAL: JanSorensen

To be honest I dont think the AI should bother with the intricacies of individual victory. In all likelihood doing so will only weaken its ability to play a good team game by making it take actions that are tactically or even strategically unsound (just like it does for not so few humans).

For that reason I think it would be an error to let the AI go down that road. Forget about the diplomatic aspect and concentrate on the military one instead as those are at odds with the latter being several orders of magnitudes more important.

WiF is not Diplomacy afterall.

There 3 topics being mixed up here.

One topic is the question of who wins. My previous post answers that question with: "it doesn't matter when there is a human- AIA alliance".

Another topic is the communication between the AIO's for different major powers on the same side and whether one of them tries to score more victory points even if that means the overall score for their side suffers. This I intend to control as an internal parameter which is set to NO. As a game option it can be varied by the human opponent to make the AIO play weaker because of divisiveness within the AIO's side.

The third topic is the question of whether the AIO is designed as one monolithic opponent or as a set of individual opponents that work together. On this design question I decided (in July) for the latter instead of the former. There were several reasons for that decision, but the primary one is that as a human player controlling all the countries on one side, I think of them individually. I plan what to do each turn, each impulse, for each major power separately. The game design more or less forces this process upon the player because of the requirement to choose a Action type each impulse for each major power. Within my head, I make the trade off decisions between Germany wanting a land action and therefore Italy has to do a naval (or combined). It is easier to write the AIO mimicing the logic I use than to try and create a monolithic world view.

From my perspective these decisions have been made. I see no reason to revisit them, even in idle curiousity.

I have stated this back when I first started on MWIF, but not repeated it recently. So here is my position on design decisions:

For any design decision, I explore as many alternatives as possible given the time constraints. The alternatives do not need to be reasonable, and wild ideas are encouraged, to help us to think beyond any preconceived notions we might be unaware we have. I select those alternatives that seem best and develop the implications of using those designs in a little bit of detail. Then I let the forum members critique the reduced number of alternatives, arguing amongst themselves more than with me (hopefully). If it is something that has minor impact on coding time, I let the forum members decide. If it impacts the amount of effort required to implement it as code, then I'll decide after listening to all your voices. It circumstances where coding is very easy, I make all alternatives available and the player can choose which to use when he sits down to play (e.g., the flags). Once I have decided on a design and started to code it, further discussion is moot and of no interest to me. Only in exceptional circumstances will I trash code that has already been written simply to 'improve' an aspect of the game. The design flaw has to be fatal or a major annoyance, where I get to decide the level of annoyance.

The motivations for the above should be obvious, but just in case they aren't: Revisiting design decisions can make development of software (MWIF in particular) drag on for years. Design it right, code it once and development can get done before we all die of old age.

[fuzzy_bunnyy]

how, if at all, will a human player communicate long term strategic goals to the AI? I find that especially when playing the Axis all 3 countries should decide on what to do. If Japan decides to invade Siberia and Germany goes Sea Lion....yada yada, you get it. just wondering.

[Shannon V. OKeets]

ORIGINAL: fuzzy_bunnyy

how, if at all, will a human player communicate long term strategic goals to the AI? I find that especially when playing the Axis all 3 countries should decide on what to do. If Japan decides to invade Siberia and Germany goes Sea Lion....yada yada, you get it. just wondering.

A good question. I actually have the communication between human and AI Assistant as a lower priority than that between AI Opponents on the same side. Which means I will solve the latter problem first, and then modify that solution for the human - AIA interaction. Someone earlier stated what in hind sight is obvious: the human is in charge and the AIA will take strategic direction from the human. That still doesn't answer your question of how. But it does simplify matters somewhat from the programming side.

The solution to how to communicate is to define strategic plans in terms of variables and establish the various values each can take on. I have a good start on a list of variables but the second piece - the values they can take on - is less complete. I have a lot to work on right now getting things done for play test. The human-to-AIA interface is nowhere near the top. Sorry.

[Glen Felzien]

Could the human to AIA be as easy as selecting strategy from a list? Maybe select your top three picks for the AIA for each allied nation/power and then leave the AIA to do what it can with a percentage chance that it would do one of the other picks? For example a possible list for Japan could be Garrison Gains, Fortress Japan, China Total War, China Withdraw, China Limited War, Operation Vlad, Kill the Bear, The Jewel in the Crown (India), African Riches (Madagascar), Free the Colonies (Aus &NZ), Secure the Greater Prosperity Sphere (Pacific Island Grab) etc. The human player would select three and the AIA would best attempt the first if its internal die roll succeeded (80% chance it would pursue the strategy(?)) else it would pursure the second etc

Anyway, just an idea for a later time.

[scout1]

I like the idea of some variability. In WitP, there are players who pout about not being able to control EVERYTHING, ALL THE TIME. Nimitz to the supply sargent. They complain when the AIA performs a task that is not exactly what they want, where they want and when they want.

Variability is a good thing. Keeps things hopping.

[dhatchen]

I had some ideas on this and put them in post #83 in the Artificial Intelligence in World in Flames thread.

[Shannon V. OKeets]

ORIGINAL: Glen Felzien

Could the human to AIA be as easy as selecting strategy from a list? Maybe select your top three picks for the AIA for each allied nation/power and then leave the AIA to do what it can with a percentage chance that it would do one of the other picks? For example a possible list for Japan could be Garrison Gains, Fortress Japan, China Total War, China Withdraw, China Limited War, Operation Vlad, Kill the Bear, The Jewel in the Crown (India), African Riches (Madagascar), Free the Colonies (Aus &NZ), Secure the Greater Prosperity Sphere (Pacific Island Grab) etc. The human player would select three and the AIA would best attempt the first if its internal die roll succeeded (80% chance it would pursue the strategy(?)) else it would pursure the second etc

Anyway, just an idea for a later time.

My difficulty with agreeing to any of this is that is needs a lot of study.

I have gone into excruciating detail for creating the AIO and have an extremely long list of decisions that the AIO has to make to play this game at all competently. While you, as a human, are comfortable with a short title to convey a lot of information, I as a programmer, have to understand what that title means in terms of all those decisions the AIA would have to make.

I also would want any system to apply to any combination of countries that the human would want to assign to the the AIA and to himself. That means having the AIA play both the USA and China or Italy and Japan, or any of the other dozens of combinations. When you add the combinatorics of the different strategic plans for each major power with their different emphases to the north, south, east, or west, you get to a big number.

For a design to accommodate all that well is non-trivial. It isn't something that can be decided in 10 minutes while typing over the keyboard. I have been worrying at the problem for the last 7 months for the AIO. The AIA is standing in line behind his big brother.

[Glen Felzien]

As is obviously appearent, I am not a programing guru.

ORIGINAL: fuzzy_bunnyy

how, if at all, will a human player communicate long term strategic goals to the AI? I find that especially when playing the Axis all 3 countries should decide on what to do. If Japan decides to invade Siberia and Germany goes Sea Lion....yada yada, you get it. just wondering.

This was the original question I was trying to answer. Please allow me to better articulate my suggestion for a possible solution.

Regarless of what side Japan is on, Japan has to make a decision as to what overall strategy it wishes to pursue from the beginning of the game and during various points during the game. As an AIO, the player will not actively intereact in the decision making. There would be indirect interaction as the player and AIO use the game pieces to influence the others strategy. However, as an AIA (Japaen as an ally) there would need to be a direct method of interaction between player and AI thus those fancy tiltes representing possible strategic approaches. The titles simply represent the code behind that particular strategy.

I now referance my opening sentence when I ask the following: Why does the AIO and the AIA have to be different from a strategic point of view. Regardless of the side, a major power will have a host of possible strategic possiblities. These will not be any different if the power is an enemy or a friend. The only difference, as I imagine it, is the level of player AI interaction. I imagine a basic interection for strategic "diplomacy" as referred to in my previous post and a more detailed interaction when my units and the AIA's units share the same theatre.

So unless I have missed something brutally obvious, and goodness knows it has happened before, why cant the AIO and the AIA be fundamentally the same?