Matrix Games Forums

ORIGINAL: el cid again

2) After AE is released it will be possible to determine what really happens by creating test situations - and from that it should be possible to create AI oriented mods. For a current example of this sort of thing, look at RHSAIO. A number of things were done to "help" the AI "know" how to manage - at least the Japanese side. Command assignments, unit objectives, upgrade paths and similar things can be "programmed" by modders to help AI do a scenario which was designed for human vs human play. While my focus is PBEM, I plan to do two or three such scenarios - for players who want to practice before they do PBEM or FTF - and because it is very useful in certain long term tests to have games play out more or less "right". Since I am a test guy (in a different sense than game testing per se - but ANY sort of software testing) I may find this easier to do than some. But we can help each other - answer questions - so that modders should be able to make some AI oriented scenarios if there is an interest in such.

I sincerely hope you and others will be allow to do this for the sake of the community and the game.

ORIGINAL: Erik Rutins

I'm honestly getting the feeling that there's some kind of rivalry going between PBEM and AI players and I just don't see the need or sense of it.

I agree with most of what you say but I sure don't share your feelings regarding this statement. Nothing I have read demonstrates such a rivalry exists. I appreciate everything the PBEM folks bring to the game with their AARs, strategy suggestions, problem surfacing, etc, etc. and think that other AI'ers probably feel the same.

ORIGINAL: Sardaukar

I think I'll be happy with new AI if it plays bit better than current...[:D] I think AE will be lot more enjoyable than WITP in this aspect.

I'll be happy even if AI continues to be totally "scripted", as long as it could select between few "scripts".

What I really would like to see is Programmed Opponent able to evaluate situation every 6 months, based on bases occupied, economical situation and forces in hand etc.. Then it'd select one "script" from appropriate list for that year (even could have small random chance included) and follow it until next evaluation. Wish that there would also be triggers for critical events (losses of major assets or bases) forcing another "script check".

Might be bit beyond the scope of AE, though...

I second this. Very much like the AI system in TOAW.

ORIGINAL: Erik Rutins

I'm honestly getting the feeling that there's some kind of rivalry going between PBEM and AI players and I just don't see the need or sense of it.

Not so much a rivalry as a divergence of interests..... The PBEM'ers are tickled pink over the promised improvements, and can afford to be as they don't care if the AI can function with them or not...... The AI-only crew are worried that an AI which is already incapable of dealing with the current incarnation of WITP will perform even worse vis-a-vis the player as the complexity of the system is increased..... If you play exclusively against the AI (which is currently not very good, especially in the long run), it's a legitimate fear.

I would again encourage those who play only the AI currently to consider getting "the Admiral's Edition" and trying some PBEM as well. Everyone will be starting with a "clean slate" experiance-wise, and with all the posts on this and other threads it shouldn't be a problem finding another former "AI only" player or two to start a game with. Both of you will know "up front" your playing time is restricted, so no fear of dissappointing your opponant. And you will both still have the AI to play with as before. You will just also have a game or two that is more challanging because your opponant is a thinking human being. Why not "the best of both worlds"?

ORIGINAL: Mike Scholl

ORIGINAL: Erik Rutins

I'm honestly getting the feeling that there's some kind of rivalry going between PBEM and AI players and I just don't see the need or sense of it.

I would again encourage those who play only the AI currently to consider getting "the Admiral's Edition" and trying some PBEM as well. Everyone will be starting with a "clean slate" experiance-wise, and with all the posts on this and other threads it shouldn't be a problem finding another former "AI only" player or two to start a game with. Both of you will know "up front" your playing time is restricted, so no fear of dissappointing your opponant. And you will both still have the AI to play with as before. You will just also have a game or two that is more challanging because your opponant is a thinking human being. Why not "the best of both worlds"?

Mike, your talking from the pulpit. How many words does it take to explain the valid reasons it's not our (my) cup of tea? I am happy for you that it is and I wouldn't discourage you for the world. Our preference hasn't developed as a result of the new product. If there are some converts, swell, more fun to them.

ORIGINAL: Mike Scholl

ORIGINAL: Erik Rutins

I'm honestly getting the feeling that there's some kind of rivalry going between PBEM and AI players and I just don't see the need or sense of it.

Not so much a rivalry as a divergence of interests..... The PBEM'ers are tickled pink over the promised improvements, and can afford to be as they don't care if the AI can function with them or not...... The AI-only crew are worried that an AI which is already incapable of dealing with the current incarnation of WITP will perform even worse vis-a-vis the player as the complexity of the system is increased..... If you play exclusively against the AI (which is currently not very good, especially in the long run), it's a legitimate fear.

I would again encourage those who play only the AI currently to consider getting "the Admiral's Edition" and trying some PBEM as well. Everyone will be starting with a "clean slate" experiance-wise, and with all the posts on this and other threads it shouldn't be a problem finding another former "AI only" player or two to start a game with. Both of you will know "up front" your playing time is restricted, so no fear of dissappointing your opponant. And you will both still have the AI to play with as before. You will just also have a game or two that is more challanging because your opponant is a thinking human being. Why not "the best of both worlds"?

A lot of people (myself included) prefer to play the AI because you can play at your own pace and the war itself moves along quite nicely. I could sit down to play a game and play for five or more hours and get weeks or a month of action in. You can play when you want, as long as you want. In PBEM, the time you have to play the game is really dependant on your opponents life not your own. A lot of people can only play 1 turn a day if that. That is several years, if the game is even finished.

A game of this scope and length is very difficult to play consistenly and completely in PBEM. The problem is, is that the AI is really not that intelligent. And for people who actually dont have time to dedicate years of their life to one campaign, there should be a challenge available so that they may see their hard work through to the end and be able to play at their own pace, and be challenged at the same time!

BEACH and SUNSHINE. Please read what I wrote again..., I'm on your side insofar as your concerns about if AE will offer any significant improvement for the AI-only player. Given the old code and it's restrictions, I think you have a legitimate worry.

My second paragraph was only a suggestion that AE "might" offer a good chance to try PBEM with some other folks who have the same problems with PBEM currently as you do. Unless it's outrageously over-priced, I'll bet many of you will buy it just to see what all is "under the hood". If I'm right about that, then I was just offering some encouragement to try PBEM with someone else in the same boat. Nobody says you have to finish the game, or meet any time and performance restrictions. Just a suggestion to get some additional value out of something you have.

With all the talk about the AI being "scripted", is there no way to allow users to contribute to the AI behavior? I understand that a lot of AI behavior might depend on parameters that the developers want to keep secret, but any sort of user-modification ability would be fantastic. Would this be possible at all?

You need to ask someone who knows the programming. I'd suggest the AE General Thread...

ORIGINAL: Mike Scholl

BEACH and SUNSHINE. Please read what I wrote again..., I'm on your side insofar as your concerns about if AE will offer any significant improvement for the AI-only player. Given the old code and it's restrictions, I think you have a legitimate worry.

My second paragraph was only a suggestion that AE "might" offer a good chance to try PBEM with some other folks who have the same problems with PBEM currently as you do. Unless it's outrageously over-priced, I'll bet many of you will buy it just to see what all is "under the hood". If I'm right about that, then I was just offering some encouragement to try PBEM with someone else in the same boat. Nobody says you have to finish the game, or meet any time and performance restrictions. Just a suggestion to get some additional value out of something you have.

Oh, Im absolutely buying it (I own nearly every Matrix WW2 title), but I just really, really want a challenging game vs the AI. Hell Im excited about the changes, it will feel like a totally new game. I just wish AI would be a priority as well. We've waited this long, why not push release back until fall of '08 and spend a few months on working on the AI too.

And, yes, Ive tried PBEM, but I lose interest after awhile, either waiting on a turn to be returned, or wanting to play longer than just one turn, or even forgetting what the heck I was doing because of a break forced by an opponent.

To me it's about winning, which means coming close to finishing the game.

I fully appreciate that you can't work on AI until you implement OOB and other features. I've coded a few AIs for games I've made for my kids. (Chess, checkers, hockey, puzzle games.) In every case, the AI was virtually the last thing I implemented. If you make a checkers game, first you have do define the board; define the pieces; define rules for how they move (for either side); define how captures are made; define kings behavior; define a UI for the human to move things; and then, finally, start creating AI -- how to evaluate the board, how to evaluate moves, how to compare moves, storing possible moves, etc.

Also, many programmers will tell you that the AI is the most fun part of a project. It is for me. Steve, the WIF programmer, says he took that massive project on because above all he was most interested in the AI -- but it's the thing he'll get to last.

A couple questions and thoughts on improving the AI:

1. AGEOD's Civil War game includes a checkbox option: "give the computer more time?" I always check it. Perhaps we could get an option, if processing time is a concern? I'd be willing to wait 15-20 minutes for an AI turn, maybe longer -- provided that the turn didn't start executing automatically. (In other words, I'd want to get up, stretch, and come back to a button saying "AI done thinking; ready to watch turn execution?", rather than return to a fully-completed turn.)

2. Is the AI really completely script-driven? It seems also to do some tactical analysis, as evidenced by the air bal numbers. On a strategic level, I would think some "big picture" issues are quantifiable for each side's AI -- how much tonnage is in which theater, the state of Japanese production (a number I might let the Allied AI peek at).

3. I recognize that you may not be able to get away from a somewhat scripted model for the initial Japanese thrusts. As others have suggested, is it possible to stick with the "script" model for the Japanese side, but give the AI a half-dozen scripts to choose from? Historical attack pattern; less emphasis on Philippines, more on Malaya/SRA; start with Java versus start with Sumatra/Borneo; longer-term emphasis on South Pacific, or CentPac, or Burma/India. Undoubtedly it's harder to script the Allied side, but I would think it feasible to program triggers for the defensive phase of the war, and to program an array of possible reactions.

4. As others have suggested, is there any chance you'd give us tools to mod the AI ourselves? I know that creating such a tool -- a scripting tool, like the scripting tool that released with Neverwinter Nights 2 -- would be a big undertaking, maybe one that could wait til after the release of AE. But if there were one mod I'd ever be interested in contributing to WITP/AE, it would be an AI mod. Writing an AI is a blast.

Or better yet, just work on a really challenging AI for one side.  Id be willing to play several campaigns versus the same side if it continued to surprise me.  It would allow for the opportunity to try new strategies.

ORIGINAL: BlackSunshine

ORIGINAL: Mike Scholl

BEACH and SUNSHINE. Please read what I wrote again..., I'm on your side insofar as your concerns about if AE will offer any significant improvement for the AI-only player. Given the old code and it's restrictions, I think you have a legitimate worry.

My second paragraph was only a suggestion that AE "might" offer a good chance to try PBEM with some other folks who have the same problems with PBEM currently as you do. Unless it's outrageously over-priced, I'll bet many of you will buy it just to see what all is "under the hood". If I'm right about that, then I was just offering some encouragement to try PBEM with someone else in the same boat. Nobody says you have to finish the game, or meet any time and performance restrictions. Just a suggestion to get some additional value out of something you have.

Oh, Im absolutely buying it (I own nearly every Matrix WW2 title), but I just really, really want a challenging game vs the AI. Hell Im excited about the changes, it will feel like a totally new game. I just wish AI would be a priority as well. We've waited this long, why not push release back until fall of '08 and spend a few months on working on the AI too. From what I've read, a decent AI will require a complete code re-write from the ground up. That's a very expensive and time-consuming project. I have the feeling that the "Admiral's Edition" is kind of a "test balloon" to see if there is a market for such an endevour.

And, yes, Ive tried PBEM, but I lose interest after awhile, either waiting on a turn to be returned, or wanting to play longer than just one turn, or even forgetting what the heck I was doing because of a break forced by an opponent. I reccognize and have experianced all these problems..., and agree that PBEM isn't perfect either (unless you just luck out and stumble on a perfectly matched opponant)

To me it's about winning, which means coming close to finishing the game. I understand your point..., but given the AI's "capabilities", isn't beating it kind of like beating your little brother? That's what brought me to try PBEM..., what fun is it boxing a one-armed man?

El Cid wrote: "1) A big problem with AI is the wastage of valuable ship tonnage - running ships and convoys to nonsensical locations with tiny or even zero cargo - running them by outrageous or even impossible routes - running them without regard to the fuel required to traverse the distance. Revisions to how ships are assigned missions probably are possible with modest effort - and no change at all to the AI structure. "


Sid, I was looking at the screenshots and on the one showing the patrol zone settings for a sub TF you can see a set of switches for "Routing Control: Normal Safest Safer Direct". SInce one of the areas they have mentioned that they have worked on is TF routing, it appears that they are giving us some modicum of control over this.

1. AGEOD's Civil War game includes a checkbox option: "give the computer more time?" I always check it. Perhaps we could get an option, if processing time is a concern? I'd be willing to wait 15-20 minutes for an AI turn, maybe longer -- provided that the turn didn't start executing automatically. (In other words, I'd want to get up, stretch, and come back to a button saying "AI done thinking; ready to watch turn execution?", rather than return to a fully-completed turn.)

2. Is the AI really completely script-driven? It seems also to do some tactical analysis, as evidenced by the air bal numbers. On a strategic level, I would think some "big picture" issues are quantifiable for each side's AI -- how much tonnage is in which theater, the state of Japanese production (a number I might let the Allied AI peek at).

3. I recognize that you may not be able to get away from a somewhat scripted model for the initial Japanese thrusts. As others have suggested, is it possible to stick with the "script" model for the Japanese side, but give the AI a half-dozen scripts to choose from? Historical attack pattern; less emphasis on Philippines, more on Malaya/SRA; start with Java versus start with Sumatra/Borneo; longer-term emphasis on South Pacific, or CentPac, or Burma/India. Undoubtedly it's harder to script the Allied side, but I would think it feasible to program triggers for the defensive phase of the war, and to program an array of possible reactions.

Let me add a good, old AOLer "me too!" to this list. Saved me a lot of typing.

If it's brute-force decision path analysis that can make the AI better I too would be happy to wait 10-15 minutes between turns. My TIVO is right next to the PC.

Also, the AI's micro-modules or tactical modules aren't bad for the most part. It's strategy over years where it falls apart. I too vote for new thinking either for the up-thread proposal to allow the AI to annually teleport forces into optimal positions, or else let the human act as an AI-helper every January 1 by telling the AI what strategic aims to pursue, with a dice roll built in for it to pout and refuse.

And, as has been said, the Japanese side seems orders of magnitude easier to program for, both because of interior lines of communication and fewer forces. I wouldn't mind if the majority of AI work focused on making the Japanese a lot better (said the allied fanboy.)

I'm interested in WIP, but I haven't bought it so far due to its limited AI (I'm an AI-only player too for several reasons).
I see that the AI can only be finalized, when the engine features are implemented. But I've the impression, that it is not planned to build a new AI engine.

I know that it is very hard to develop a capable AI for such a game/engine. It would need make use of sophisticated learning algorithms and methods and not rely (only) on hard-coded rules.

@Erik: A limited SDK to modify the AI is probably out-of-scope, isn't it?

It maybe a good idea to start an AI only wish list?

I third the items listed but we won't get them for the initial release of AE. On the chat last night jwilkerson said the conentration for AE is on filling out the OOB and the other posted items. Don't expect anything big for the AI.

I think its an expected response and he did respond in a positive way (see the chat once they post it). I believe Matrix and Henderson Field designs are listening to the AI beefs despite the conversion attempts by some well intentioned individuals.

So if there will be patches or enhancements to AE in my opinion it maybe possible to get revisions to the AI at that point.

0.02$

ORIGINAL: Snowman999

1. AGEOD's Civil War game includes a checkbox option: "give the computer more time?" I always check it. Perhaps we could get an option, if processing time is a concern? I'd be willing to wait 15-20 minutes for an AI turn, maybe longer -- provided that the turn didn't start executing automatically. (In other words, I'd want to get up, stretch, and come back to a button saying "AI done thinking; ready to watch turn execution?", rather than return to a fully-completed turn.)

2. Is the AI really completely script-driven? It seems also to do some tactical analysis, as evidenced by the air bal numbers. On a strategic level, I would think some "big picture" issues are quantifiable for each side's AI -- how much tonnage is in which theater, the state of Japanese production (a number I might let the Allied AI peek at).

3. I recognize that you may not be able to get away from a somewhat scripted model for the initial Japanese thrusts. As others have suggested, is it possible to stick with the "script" model for the Japanese side, but give the AI a half-dozen scripts to choose from? Historical attack pattern; less emphasis on Philippines, more on Malaya/SRA; start with Java versus start with Sumatra/Borneo; longer-term emphasis on South Pacific, or CentPac, or Burma/India. Undoubtedly it's harder to script the Allied side, but I would think it feasible to program triggers for the defensive phase of the war, and to program an array of possible reactions.

Let me add a good, old AOLer "me too!" to this list. Saved me a lot of typing.

If it's brute-force decision path analysis that can make the AI better I too would be happy to wait 10-15 minutes between turns. My TIVO is right next to the PC.

Also, the AI's micro-modules or tactical modules aren't bad for the most part. It's strategy over years where it falls apart. I too vote for new thinking either for the up-thread proposal to allow the AI to annually teleport forces into optimal positions, or else let the human act as an AI-helper every January 1 by telling the AI what strategic aims to pursue, with a dice roll built in for it to pout and refuse.

And, as has been said, the Japanese side seems orders of magnitude easier to program for, both because of interior lines of communication and fewer forces. I wouldn't mind if the majority of AI work focused on making the Japanese a lot better (said the allied fanboy.)

ORIGINAL: Grotius

A couple questions and thoughts on improving the AI:

1. AGEOD's Civil War game includes a checkbox option: "give the computer more time?" I always check it. Perhaps we could get an option, if processing time is a concern? I'd be willing to wait 15-20 minutes for an AI turn, maybe longer -- provided that the turn didn't start executing automatically. (In other words, I'd want to get up, stretch, and come back to a button saying "AI done thinking; ready to watch turn execution?", rather than return to a fully-completed turn.)

Me, too!

Heck, I'd be willing to wait overnight or even days for the AI to do its thinking.

If AI effectiveness is proportional to the time we allow it, give it more time!

How about something like this?

Give the computer more time?

[ ] No, use the default time [60 seconds? 5 minutes?]
[ ] Yes, give it 10 minutes.
[ ] Yes, give it one hour.
[ ] Yes, give it <fill in the blank> hours.
[ ] Give it unlimited time until I [the player] press the Stop button.
[ ] [any other option suggestions?]

Make this a pre-game selection, or even a turn-by-turn selection (see below).

For a game of this scope and complexity, If I the human player might take hours to plan and execute my turn, I'd think it only fair to give the AI equal time consideration.

If we're talking about a game that might take months and even years to complete, what's a few hundred extra hours among "friends" (you and your AI opponent), especialy if those extra hours you are (a) sleeping, (b) at work, (c) playing some other computer game, (d) whatever.

I could see playing in the following manner: I commit to playing one turn at a time, then giving the computer indefinite time--at least overnight, days if I get distracted and have to move on to other things (including playing other games)--to plan and "think" through its response. Distracting time off between moves would even be a virtue. It I knew that the longer I wait, the better the AI move might be, I'd not fret over my time away from the game. In fact, variable AI time between moves might have the added advantage: For complicated situations and important turning points in the war, give it full time consideration. At other times, when there is little for the AI to do, give it as little time as your experience and/or impatience to "get on with it" deem necessary.

For the life of me, I can't fathom why game designers try to make their games, and their AI, fit one mold. Break the mold! Give the players a choice! If enabling a competitive AI is a matter of giving it more time to think, let us the human players decide for ourselves the limits of our (im)patience. These games allow us, the human players, to set all sorts of optional rules going in. Why not give us as many AI options (particularly PO time to think), too?

Variable, human player directed AI time!

2. Is the AI really completely script-driven? It seems also to do some tactical analysis, as evidenced by the air bal numbers. On a strategic level, I would think some "big picture" issues are quantifiable for each side's AI -- how much tonnage is in which theater, the state of Japanese production (a number I might let the Allied AI peek at).

The Original Poster raised questions about both tactical and strategic level AI. Surely the AI could be made smarter not to make specific infuriating tactical mistakes. Yes, it is probably too much to expect any reasonable AI to be competent in its strategic thinking over the longer term. But in the short term, and for specific game play actions, surely this is possible?

3. I recognize that you may not be able to get away from a somewhat scripted model for the initial Japanese thrusts. As others have suggested, is it possible to stick with the "script" model for the Japanese side, but give the AI a half-dozen scripts to choose from? Historical attack pattern; less emphasis on Philippines, more on Malaya/SRA; start with Java versus start with Sumatra/Borneo; longer-term emphasis on South Pacific, or CentPac, or Burma/India. Undoubtedly it's harder to script the Allied side, but I would think it feasible to program triggers for the defensive phase of the war, and to program an array of possible reactions.

The shorter the scenario, the easier it is to script. Scripting the entire war, or entire single years, is difficult to impossible. Much more possible is scripting a few weeks of action. More and shorter scenarios: Make it so!

4. As others have suggested, is there any chance you'd give us tools to mod the AI ourselves? I know that creating such a tool -- a scripting tool, like the scripting tool that released with Neverwinter Nights 2 -- would be a big undertaking, maybe one that could wait til after the release of AE. But if there were one mod I'd ever be interested in contributing to WITP/AE, it would be an AI mod. Writing an AI is a blast.

Amen, amen, amen! Open up the AI, as much as possible, to modder development! Enthusiastic and committed and even obsessed modders can do wonders for the PO. Look at how the EUIII (Europa Universalis) Magna Mundi team (just to give the most successful example) has vastly improved that game's AI (or at least mitigated many of its shortcomings and annoying tendencies). Open up the AI to user modding and harness the energy and pull of your user base to effectively make them a part of your post-release development team.

Maximizing user, and modder, freedom and choice--therein lie the answers!

ORIGINAL: Grotius
...
For a game of this scope and complexity, If I the human player might take hours to plan and execute my turn, I'd think it only fair to give the AI equal time consideration.

As I understand it, in a human (Japan) vs. AI (Allies) game the AI has to:
- wait for the human player to end the Japanese turn
- give orders to AI-controlled Japanese regions
- give orders to AI-controlled Japanese task forces
- give orders for the entire AI-controlled Allied side

What would be good is if the AI could start calculating orders for the Allied side as soon as the execution phase is over. There is no need to wait for the Japanese human player to end his turn. During the Allied order phase, the PC is probably largely idle for at least several minutes so the AI should be to do a lot of processing without making the human player wait at all.

ORIGINAL: berto

If AI effectiveness is proportional to the time we allow it, give it more time!

How about something like this?

Give the computer more time?

[ ] No, use the default time [60 seconds? 5 minutes?]
[ ] Yes, give it 10 minutes.
[ ] Yes, give it one hour.
[ ] Yes, give it <fill in the blank> hours.
[ ] Give it unlimited time until I [the player] press the Stop button.
[ ] [any other option suggestions?]

Make this a pre-game selection, or even a turn-by-turn selection (see below).

For a game of this scope and complexity, If I the human player might take hours to plan and execute my turn, I'd think it only fair to give the AI equal time consideration.

Assuming the equation "More time = better AI" is true, then you'd be hard pressed to find an AI player who wouldn't LOVE to have that added as a check-box option. In the same vein, keep in mind that some of us actually can afford to buy (or already own) pretty serious machines, so the added processing time wouldn't be that much of an issue. And making it optional ensures the game doesn't become unplayable for the owners of ancient clunkers.

If it's true that much of the AI amounts to scripting, then I also support the suggestions of those who are begging for periodic script reset points, to give the AI a better chance of adjusting to changes over time (i.e. changes on the map).