AI vs. Real Player

[Titi]

ORIGINAL: Shannon V. OKeets

A couple of comments here:

1 - MWIF should make naval moves much easier (less time consuming), assuming the player knows what he wants to do. The units are easier to access (compared to large stacks in over-the-board games) and it is easier to see all the other units at sea (even when those are large stacks). The naval review details and summary forms were designed specifically for that purpose. Not to say that all the complexities of performing the Commonwealth naval moves go away, but they should be mitigated significantly.

2 - The non-phasing player could very well need to make a decision about naval interception. So, the Axis player needs to be available during those "30 minutes", just in case.

3 - I expect PBEM games to be able to transition to NetPlay and back again later. But games that start as NetPlay will not have that option. One consideration is that PBEM is only for two players while NetaPlay can have up to 6 players (3 per side).

First about 2) Only one player per side need sometime to be involved in interception, could always buzz or take the decision for an ally in predetermined portion of the sequence of play and on predetermined theater.



The only part missing for a complete integration of netplay and PBEM in a whole multiplayer unique option is to have PBEM for more than 2 players.

It would be great to have this option as a 2 players PBEM could become a 3 players game by intoducing a new player. Same for a three players becoming a four players game. Same for a 4 becoming a 3 ...

It would be even greater if PBEM & Netplay allow the option of spectator watching the game. Best way to interest and recrut future new players [:D]

In fact i think that multiplayer (Netplay or PBEM) must been as near as possible to what can happen to a real WiF game.
To be honest i had a distant experience with WiF (Version 4 of the rules and only two players) but i had much more with Empire in Arms and Europa Universalis (what a rule nightmare [X(]).
In my experience the game must be very flexible to go to the end :
- allow a player to be added during a game, to resign, to change.
- allow player to be on leave for one day or two : leaving his country to another player or being frozen and playing alone at the start of next session with others players having a veto for any move that could be a problem in the timewrap.
- allow player to take a short break to drink a beer, listen to music, chat, smoke, make a phone, watch a movie ... in a time when not involved. If presence needed, that player is quickly called back.

That will cover computer problems, connection problems, not at home problems, firewall problems, etc that can hinder a multiplayer game.

Last chrome for a multiplayer game would be the option of doing a draft before the actual game. Planning the D-Day or Barbarossa before the time to play it to speed the game when the real time come. Of course the option to stop or edit it must remain.

You spend a lot of time to translate WiF to CWiF. Just take the time to adapt the interface to humans that will play it over a very long time and have a real and social life to interfere with CWiF [:-]

[Shannon V. OKeets]

ORIGINAL: Titi

ORIGINAL: Shannon V. OKeets

A couple of comments here:

1 - MWIF should make naval moves much easier (less time consuming), assuming the player knows what he wants to do. The units are easier to access (compared to large stacks in over-the-board games) and it is easier to see all the other units at sea (even when those are large stacks). The naval review details and summary forms were designed specifically for that purpose. Not to say that all the complexities of performing the Commonwealth naval moves go away, but they should be mitigated significantly.

2 - The non-phasing player could very well need to make a decision about naval interception. So, the Axis player needs to be available during those "30 minutes", just in case.

3 - I expect PBEM games to be able to transition to NetPlay and back again later. But games that start as NetPlay will not have that option. One consideration is that PBEM is only for two players while NetaPlay can have up to 6 players (3 per side).

First about 2) Only one player per side need sometime to be involved in interception, could always buzz or take the decision for an ally in predetermined portion of the sequence of play and on predetermined theater.



The only part missing for a complete integration of netplay and PBEM in a whole multiplayer unique option is to have PBEM for more than 2 players.

It would be great to have this option as a 2 players PBEM could become a 3 players game by intoducing a new player. Same for a three players becoming a four players game. Same for a 4 becoming a 3 ...

It would be even greater if PBEM & Netplay allow the option of spectator watching the game. Best way to interest and recrut future new players [:D]

In fact i think that multiplayer (Netplay or PBEM) must been as near as possible to what can happen to a real WiF game.
To be honest i had a distant experience with WiF (Version 4 of the rules and only two players) but i had much more with Empire in Arms and Europa Universalis (what a rule nightmare [X(]).
In my experience the game must be very flexible to go to the end :
- allow a player to be added during a game, to resign, to change.
- allow player to be on leave for one day or two : leaving his country to another player or being frozen and playing alone at the start of next session with others players having a veto for any move that could be a problem in the timewrap.
- allow player to take a short break to drink a beer, listen to music, chat, smoke, make a phone, watch a movie ... in a time when not involved. If presence needed, that player is quickly called back.

That will cover computer problems, connection problems, not at home problems, firewall problems, etc that can hinder a multiplayer game.

Last chrome for a multiplayer game would be the option of doing a draft before the actual game. Planning the D-Day or Barbarossa before the time to play it to speed the game when the real time come. Of course the option to stop or edit it must remain.

You spend a lot of time to translate WiF to CWiF. Just take the time to adapt the interface to humans that will play it over a very long time and have a real and social life to interfere with CWiF [:-]

You concerns on accommodating the vagaries of real life when playing over the interent are all handled by the design. it might seem like a lot to do, but it isn't really. The program just gives the team leader the ability to dynamically control who plays which major power on his side. Then throw in the ability to change team leaders. Hopefully none of these things would be necessary in a game, but if the situation arises, then the team leaders can keep the game going.

Multi-player PBEM is not part of MWIF product 1.

I have a full design for creating, storing, and using "planned ahead" moves, but I had to remove it as a feature since it would require more of my time to implement.

[Shannon V. OKeets]

ORIGINAL: HansHafen

I think Steve needs to focus almost exclusively on the AI from now on. If this product won't work at all without a workable AI, as many have said in the "When" post responses to my suggestion, then that is the priority. I may be wrong in my understanding, but I think Steve has said that the AI has yet to make any decision in game. (Correct me if I am wrong). If this is so, massive work is still needed in this area. Four months doesn't seem to be long enough to get that done. And he is still needing to spend time on other things listed in his task list. It seems to me that the AI is going to be the most challenging aspect of this entire project, correct? Won't this take the longest to complete and test?

Well, if you ask the guys who want to play over the internet, NetPlay is the most important thing to do first.[;)] And if you ask the guys who want to play by email, the PBEM system needs to be worked on first.[:D]

I need them all to work.[8D]

The AIO is not a blank sheet of paper. I have put well over 1000 hours into it so far.

1 - The laguage specification (LAIO = Language for an Artifical Intelligence Opponent) is done
2 - Sample scripts written in LAIO are done
3 - Defined points within the sequence of play where the AIO needs to make a decision are all identified
4 - AIO modules written in Pascal to branch to the script for making the necessary decision at each decision point are done
5 - English text on how each decision will be made by the AIO (94 typed pages) are 80% done
6 - Strategic plans for each of the 8 major powers (over 200 pages) are 90% done

If I could turn over all the other tasks required to complete MWIF to someone else right now, I would happily do so. The only task I would not give up (though I willingly share the work) is the AIO. Coding the AIO has been my primary motivation since day one on this project, and that has never changed.

[Froonp]

ORIGINAL: Shannon V. OKeets
5 - English text on how each decision will be made by the AIO (94 typed pages) are 80% done

If you need a typing hand, I can do that for you.
Just tell me what I should type, and I can type it very fast.

[brian brian]

I tried to kick-start a discussion about an idea I had in the Italy thread a little bit, but at what point does the AI begin to question it's strategy? (Human beings do this endlessly, sometimes shifting disastrously, other times shifting strategy correctly just in time). For example the Germans aligning Turkey in the summer of 1942 changes things considerably for Italy and the CW and US forces just starting to enter the war. Should the Wallies shift gears from developing air bases on the Italian Coast sea zone and commit more forces to attempt to re-open a land connection to a Russia in dire straits? Should the Italians or the Germans send units across the Dardanelles to complete the subjugation of the Middle East, or would that be a waste of time as a major US fleet looks poised to dock in Bombay? These questions are occupying my solitaire-WiF time right now, and weren't on my mind when I contemplated the strategy for each side at the beginning of the war.

It looks like the AI programming will be the project of the summer, and I hope we can all help hash out those LAIO documents if they aren't complete.

[HansHafen]

Cool. Looks like you have alot of the work done. I await Spetember!

[HansHafen]

How does the AI decide to stack units? Will it stack two big infantry units in one hex with a division also and then leave one division alone in the next hex? Will it divide it up to one infantry and one division each? Can you program this type of decision making?
 
How do you program the AI to consider and prepare for double moves? Offensive chits? Para drop possibilities? SRing?
 
Can the AI discern where the enemy Armor is and move AT divs etc to that sector? For example, if the German puts all his armor in AG South, can the AI see that and move AT assets down there? Or will it leave them where they are and not be able to respond to that type of tactic?
 
Will the AI garrison non-threatened areas deep in the home country when those troops are needed at the front?
 
How will the AI react to building strategies of its opponents? Can it?
 
We the AI be able to see and respond to a massive air buildup in England?

[Shannon V. OKeets]

ORIGINAL: HansHafen

How does the AI decide to stack units? Will it stack two big infantry units in one hex with a division also and then leave one division alone in the next hex? Will it divide it up to one infantry and one division each? Can you program this type of decision making?

How do you program the AI to consider and prepare for double moves? Offensive chits? Para drop possibilities? SRing?

Can the AI discern where the enemy Armor is and move AT divs etc to that sector? For example, if the German puts all his armor in AG South, can the AI see that and move AT assets down there? Or will it leave them where they are and not be able to respond to that type of tactic?

Will the AI garrison non-threatened areas deep in the home country when those troops are needed at the front?

How will the AI react to building strategies of its opponents? Can it?

We the AI be able to see and respond to a massive air buildup in England?

Combat value (CV) of a unit is the fundamental building block. When on defense, that depends on terrain etc., and likewise when on the attack. Speicialized unit capabilties are included in these calculations.

Choices are every other hex defense if the enemy is at least 1 hex away, or a continuous line defense [a few specialized alternatives also exist]. Averaging the defense strength along the front is usually best, though enemy attacks can be 'directed' by purposely weakening one section of the front line.

I don't really have time to go into all the details. And your questions didn't mention such things as unit mobility and supply, both of which are crucial factors for deploying units.

CV determines what is built and when, though in conjunction with the overall strategic plan and projected operational considerations.

==
Sorry, but I need to focus on programming.

[HansHafen]

Thanks, understand.

[Anendrue]

ORIGINAL: Shannon V. OKeets
The AIO is not a blank sheet of paper. I have put well over 1000 hours into it so far.

1 - The laguage specification (LAIO = Language for an Artifical Intelligence Opponent) is done
2 - Sample scripts written in LAIO are done
3 - Defined points within the sequence of play where the AIO needs to make a decision are all identified
4 - AIO modules written in Pascal to branch to the script for making the necessary decision at each decision point are done
5 - English text on how each decision will be made by the AIO (94 typed pages) are 80% done
6 - Strategic plans for each of the 8 major powers (over 200 pages) are 90% done

If I could turn over all the other tasks required to complete MWIF to someone else right now, I would happily do so. The only task I would not give up (though I willingly share the work) is the AIO. Coding the AIO has been my primary motivation since day one on this project, and that has never changed.

Thanks for the detailed breakdown of the progress on the AI. Especially when your energies are highly directed toward the finishing line. Hopefully this will silence negative comments that the AI was tossed in at the last moment.

[macgregor]

I need to take responsibility for my decision not to be a beta-tester. This kind of leaves me on the outside looking in. Any commentary I offer is bound to be controversial, and perhaps out of the mainstream as I am not in the mainstream. For my own sake as well as any of those who see my POV as less than constructive, that I should be a little less talk and more read. I'll check back but I think I probably should refrain from posting.

[Anendrue]

macgregor, I was not impying your comments. There qre a lot of new people with very few posts that have only read one or two status reports and have made decisions regarding that. I am just happy Steve clarified this issue as one which has been in work for a lengthy time as you have also stated. I believe his post puts to rest the concept thet he is going to build an AI in one or two months which even to an untrained eye would be irrational at best. However I do believe there will have to be a few updates on the AI as players figure out how the game plays, find exploits and gambits.

[Bibs]

Have any of you seen the performance of the AI for other Matrix games like Empires in Arms and War in the Pacific? I'm being kind when I say they are not very good. They can OK tactically sometimes, but strategically they will follow a script no matter how the situation changes. As long as you play to that script it will be so-so, but if you go another direction that's it. I use WitP as an example, playing the Allies I deliberately waited for the Japanese to take guadalcanal before I did anything - by mid 1943 they still hadn't taken the most important spot in the Solomons which was undefended.

Look at Grigsby's World at War - a much simpler game and the AI is still awful.

This isn't a knock on Shannon, there are simply to many variables to make a decent AI for these kind of strategic games without spending the ten years working on it. Just think about the number of factors that can go into deciding whether to intercept a ground strike missions and imagine having to code that logic. The game needs to have an AI (especially for those who never played the board game), but it will sink or swim on Netplay/PBEM. After 1-4 games playing the AI won't be worth the effort.

[IKerensky_alt]

The AI from the others matrixgames (since SPWAW wich have a worst AI than SP-WWII) and notably the strategical ones you quoted was one of the reason why I was so reserved about MWiF.
 
Then I read the AI Thread and it was an eyes-opener. I really, really, like the way Steve is approching this thing and I really think it could work. I dont think it will play as good as a human but , heck, even the biggest computer cannot play chess strategically as good as human. They can compute extremely large amount of data but they need a human when the situation contain no logical or statistical advantage so they cannot make a choice. Look at one of the game of Kasparov against Deep Blue (I think), the situation was clearly at the human advantage building on with a long term strategy who was obvious to even a beginer but the computer wasn't able to judge the situation strategically and ressort to useless moves as it doesnt see the danger building until too late.
 
I think Steve job is a first step and very important one and the calculation power and multi-cpu of modern PC could work really well with what he planned. Funny how this remember me of Amiga computer of my youngs that works with dedicated CPU, one for graphic, one for music...
 
I would really love to be able to have an insighter on how is this thing going to work.
 
Steve, did you run test just with the LAIO ( without incorporating maps and counter data ) on the way the different AI sub-commander coordinate ? I would really like to see some debug log of the computer decision process.
 
The good thing is that, if MWiF AI is satisfying it will be a basis on wich build other AIO for other games, so every effort put into it is more than worthwile in the long term.

[Shannon V. OKeets]

ORIGINAL: Greywolf

The AI from the others matrixgames (since SPWAW wich have a worst AI than SP-WWII) and notably the strategical ones you quoted was one of the reason why I was so reserved about MWiF.

Then I read the AI Thread and it was an eyes-opener. I really, really, like the way Steve is approching this thing and I really think it could work. I dont think it will play as good as a human but , heck, even the biggest computer cannot play chess strategically as good as human. They can compute extremely large amount of data but they need a human when the situation contain no logical or statistical advantage so they cannot make a choice. Look at one of the game of Kasparov against Deep Blue (I think), the situation was clearly at the human advantage building on with a long term strategy who was obvious to even a beginer but the computer wasn't able to judge the situation strategically and ressort to useless moves as it doesnt see the danger building until too late.

I think Steve job is a first step and very important one and the calculation power and multi-cpu of modern PC could work really well with what he planned. Funny how this remember me of Amiga computer of my youngs that works with dedicated CPU, one for graphic, one for music...

I would really love to be able to have an insighter on how is this thing going to work.

Steve, did you run test just with the LAIO ( without incorporating maps and counter data ) on the way the different AI sub-commander coordinate ? I would really like to see some debug log of the computer decision process.

The good thing is that, if MWiF AI is satisfying it will be a basis on wich build other AIO for other games, so every effort put into it is more than worthwile in the long term.

Of the 146 decision points in the game, the vast majority (> 100) can be made autonomously. There are two primary places where joint decisions are needed: Foreign Liaison and Joint Chiefs of Staff. The former coordinates between major powers and the latter between the branches of service. The foriegn coordination requires an iterative process but the branches of service compete for resources based on ~ROI = how much benefit is to be gained by allocating the resource to the decision maker (e.g., who can make better use of a fighter, or an air mission, or a reorganization point).

[rmdesantis]

Given the need for a good AI for the success of the game, and given the liklihood that any AI will make mistakes and require correction - would it be possible to have the opportunity to 'make corrections' in mid-game? That is, if the AI does something stupid that makes it too easy to beat it, that there is a way to pause the game, manually change the locations of the AI's pieces, and then continue? I don't know that it would be feasible to undo attacks or allocation of resources, but a simple correction of unit locations might be enough to keep the game playable.
 
Is this even a possibility?

[Anendrue]

ORIGINAL: rmdesantis

Given the need for a good AI for the success of the game, and given the liklihood that any AI will make mistakes and require correction - would it be possible to have the opportunity to 'make corrections' in mid-game? That is, if the AI does something stupid that makes it too easy to beat it, that there is a way to pause the game, manually change the locations of the AI's pieces, and then continue? I don't know that it would be feasible to undo attacks or allocation of resources, but a simple correction of unit locations might be enough to keep the game playable.

Is this even a possibility?

I go not believe there is s save game editor being developed. However most gaming communities tend to develop editors for every game ever released from almost any company.

[Shannon V. OKeets]

ORIGINAL: rmdesantis

Given the need for a good AI for the success of the game, and given the liklihood that any AI will make mistakes and require correction - would it be possible to have the opportunity to 'make corrections' in mid-game? That is, if the AI does something stupid that makes it too easy to beat it, that there is a way to pause the game, manually change the locations of the AI's pieces, and then continue? I don't know that it would be feasible to undo attacks or allocation of resources, but a simple correction of unit locations might be enough to keep the game playable.

Is this even a possibility?

No.

This capability (PlaceUnits) is available to the beta testers for debugging purposes but it is extremely delicate. Screw it up and the program crashes. The problem is the high interaction between the rules. Some of the more annoying items are: foreign troop commitment, overstacking, placing units in neutral countries, minor country retrictions on units leaving their home country. Actually the list is quite long. At its heart, WIF has dozens of restrictions on where units can be placed, above an beyond the simple: "Don't put armor in the middle of the Atlantic" and "Don't put battleships in Kansas".

I spent quite a bit of time back in 2005 trying to remove bugs with the PlaceUnits routine but gave up after coding work-arounds/prevention/warnings for a couple of dozen situations.

[IKerensky_alt]

Also I think that if you move/change the AI piece, the AiO will have a hard time starting from the new position mid-game... or can it adjust to the moving of just a few piece to a new position ?

[Shannon V. OKeets]

ORIGINAL: Greywolf

Also I think that if you move/change the AI piece, the AiO will have a hard time starting from the new position mid-game... or can it adjust to the moving of just a few piece to a new position ?

Well, it might just move it back to the 'wrong' position.[:D]