STILL no AI?

[Neilster]

ORIGINAL: Klydon

AI also usually has different difficulty settings. This is often achieved by certain "cheats" That may be tough to implement with a board game. One I can think of is upping the resources available to the AI. Doing other things like adding 1 to die rolls to help might be too much and a bit problematic.

I remember a discussion long ago where I became convinced that giving the AI "cheats" was a bad idea. I can't remember all the arguments but it involved people playing around them and the whole game getting skewed. As has been said, a tremendous amount of work has gone into the AI and I remain to be convinced it won't be any good. It's very easy to miss things as a player and the AI won't, and will punish you.

Cheers, Neilster

[pzgndr]

ORIGINAL: Neilster
... I remain to be convinced it won't be any good. It's very easy to miss things as a player and the AI won't, and will punish you.

Certainly there are many veterans of the boardgame who may not be challenged by a computer opponent and won't bother with it anyway. But for every one of them, there's probably a hundred other players who are newbies or casual gamers just looking for some decent gameplay at their convenience. I don't expect any AI to be brilliant, but if it's competent and challenging (even with optional difficulty settings) then that's good enough for most.

[Hansstory]

How is an AI going to take into account all the possibilities that a human does in an instant by looking at the map? Should I move more fighters over here, or more tac air, or how shall I stack my armor and AT? Can it look at the other sides builds and disposition of forces and be able to decide where he is going in the future? No, the AI can't do that or a hundred other things a human can do and does in a real game. Think about how you decide to make decisions and why. Then ask yourself how will the AI do that? With hundreds of units on all kinds of different fronts. This isn't chess with a finite amount of moves possible. This isn't even chess on steroids. It's something chess can't even comprehend, or fathom that it exists.

And the assertion that I, or anyone, is insulting the designer is just misguided. I respect what he has done and the massive amount of work put into this. It is a computer program form of a Da Vinci. But reality is unbiased. It has taken how long to get the game to the point it is today? 10 years? More? And this is the easy part! If the AI can't do a naval convoy line, then how can it hold the eastern front versus the Germans in 1941?

How does the AI decide if it should defend behind a river but one hex back from another friendly unit or should it defend one hex in front of the river to close the gap? How does the AI think about that? How does it look at the other sides units to see what is appropriate? It can't and won't. And WIF isn't an area movement game. It has hexes that matter. Too many digressions and permutations and value judgments need to be made for an AI to work with WIF.

And my ego is fine Jose. Thanks.

[Cataphract88]

An A.I. would greatly help new players to learn the game in the first place, particularly all those who have never played the board game, and consequently vastly increase sales.

[juntoalmar]

ORIGINAL: Hansstory

Think about how you decide to make decisions and why. Then ask yourself how will the AI do that? With hundreds of units on all kinds of different fronts. This isn't chess with a finite amount of moves possible. This isn't even chess on steroids. It's something chess can't even comprehend, or fathom that it exists.

Land, air, naval, resource & production planning, balancing between different fronts... It is like an AI that should be playing chess, poker, go... combined. I agree that a "dumb" AI would help people to learn how to play on a subset of the game (Barbarossa & Guadalcanal, even the western front), but for a full global game... it is just too complex, and there are too few resources to try to implement it.

Regarding ego issue. I think I'm a bad player (although I did well on my only multiplayer game) and as a Computer Engineer I would LOVE to be defeated by an AI. But I think it would need a bigger team working only in the AI here. Big enough to be not economically feasible for a company to develop it.

Guys, we need to introduce MWIF to some Google big fish or Elon Musk so they get in love with the game and start some crazy project with zillions of $ and hundreds of engineers... [:D]

[Kull]

Steve, the guy who's actually coded the entire game, says it can be done and has laid out the process. But then we have some random complainers on the internet who claim it's impossible. Who to believe? [8|]

[juntoalmar]

ORIGINAL: Kull

But then we have some random complainers on the internet who claim it's impossible. Who to believe? [8|]

It's easier to attack the people who have a different opinion to yours than to engage in an intelligent conversation and debate. If you want to discuss about what is feasible regarding AI at the present moment and this particular project resources, I'm all ears.

[Kull]

ORIGINAL: juntoalmar

ORIGINAL: Kull

But then we have some random complainers on the internet who claim it's impossible. Who to believe? [8|]

It's easier to attack the people who have a different opinion to yours than to engage in an intelligent conversation and debate. If you want to discuss about what is feasible regarding AI at the present moment and this particular project resources, I'm all ears.

I did. My initial post contains all the information provided by Steve on how the AI is going to work and progress to date. So I'll turn this question back on you. What do you find faulty in his approach? Why won't it work?

[AlbertN]

An example why the AI would not work - if I remember the priorities I've read in the past - is that for instance the airplanes are checked in value accounting some details.

The AI will never manage to understand "I need a long range fighter to cover the 2-3 box of the sea" whereas maybe around there is a 7 Air-to-Air but 3 Range FTR; and the other is a 4 Air-to-Air but 6 Range FTR. Or can understand which FTR is required instead for better cover on land operations.
The AI only see a "sum" value of statistics (which some can be weight x1.5 or 0.5 or so).

[juntoalmar]

ORIGINAL: Kull

ORIGINAL: juntoalmar

ORIGINAL: Kull

But then we have some random complainers on the internet who claim it's impossible. Who to believe? [8|]

It's easier to attack the people who have a different opinion to yours than to engage in an intelligent conversation and debate. If you want to discuss about what is feasible regarding AI at the present moment and this particular project resources, I'm all ears.

I did. My initial post contains all the information provided by Steve on how the AI is going to work and progress to date. So I'll turn this question back on you. What do you find faulty in his approach? Why won't it work?

I asked about your personal opinion, not about Steve's reports. So far it only seems "because Steve says so, and I believe him more than some random complainers on internet". Not very polite to label as "random complainers" to people who you are talking to [:-]

Regarding my opinion, I already stated previously:

ORIGINAL: juntoalmar
If you have studied AI you know how complex it can be (let's call it n). If you have played MWiF you know how complex it is (let's call it m).

A problem of complexity [n x m] is really, really complex. So, being moderately skeptic seems a very sensible option.

ORIGINAL: juntoalmar
Land, air, naval, resource & production planning, balancing between different fronts... It is like an AI that should be playing chess, poker, go... combined.

If you think that the AI for MWiF is not so complex as I think, please let us know, explain why and maybe you convince me. Let's debate. I think there are too many levels interweaved together for taking decisions.

ORIGINAL: juntoalmar
And I hope with all my heart that we'll see the AI working for this game. But, I mean, we haven't fixed yet how to find the optimal route between resources and factories which is a problem a few orders of magnitude smaller than a simple AI for MWiF

If you think this is not true or that its complexity is not a few orders of magnitude smaller than the AI, let us know and we can discuss about it.


[Edit]: BTW, apparently Steve (which I deeply respect) hasn't "coded the entire game". Still, this makes no difference regarding his capacity of implementing or not implementing the AI for the game.

[Kull]

My "personal opinion" on the AI is completely irrelevant. The reason I posted all the the AI reports was so people would have actual data to draw upon, and could critique (or applaud) based on the plan and the actions taken to implement it. Once again you've declined to make a case based on the facts, but that's the only thing I'm interested in discussing.

Steve has stated that he has a plan for the AI, and has presented a sizable amount of information on that subject. Accordingly, the burden is on YOU and all those who those who naysay the possibility of a functional AI. What is wrong with Steve's plan? Why won't it work?

[juntoalmar]

[fact] Go is a game with two players, with games of 20-90 minutes, one set of rules, with 361 different positions and about 180 stones for player. All stones are the same and have no different behaviour or values

[fact] MWiF can have different amount of players, has more than 6000 unique units with different behaviours and values (not as Go), 70200 hexes with different characteristics (they are not all the same as Go), and a combination of 80 optional rules (not like Go) and a long and complex rule set. It has some other factors of complexity as production planning, resources, variable weather, entry options and random effects (use of dice) that need handling probabilities.

[opinion] MWiF complexity of playing or making a competent AI is several orders of magnitude greater than Go

[fact] AlphaGo beat Go world champion in 2016

[fact] Google created some own proprietary hardware to run AlphaGo and spent $500 million in buying the company that developed the algorithm. That company employees 400 people and was created by a former child prodigy in chess.

[opinion] There must be lot of resources, great engineers & developers working full time in that company

[fact] There is one developer for MWiF

[opinion] Steve has made a lot of work and I believe he’s very capable of programming and developing AI

[fact] AlphaGo runs on up to 1920 CPUs and circuits developed specifically for machine learning by Google

[fact] MWiF will run in one single computer PC

[opinion] Given the complexity of the problem (as stated before) and the amount of resources committed to it (as stated before), I believe that “being moderately skeptic seems a very sensible option" regarding having a competent AI.


[opinion] You say your opinion on the AI is irrelevant. From your comments I don’t think you have any background in AI, software development or engineering. You seem more interested in futile discussion of “you vs me” or “I’m right and you are not” than talking about AI or the possibility of developing it in this game.

[opinion] I would rather spend my time in an intelligent debate with someone more respectful and that has a relevant opinion than in such futile discussions

[opinion] There as a proverb/saying that says: Ignorance is bold. Science doubts.

[Edited]: added bold

[pzgndr]

ORIGINAL: Kull
Steve has stated that he has a plan for the AI...

Bingo, end of discussion. Unless and until Steve himself states something different, the plan remains to eventually implement the AI. After the bug fixes, after the NetPlay implementation, after the other optional rules, and after the half-map scenarios. After all that, Steve will get back to what he has already started and intends to finish.

ORIGINAL: juntoalmar
If you think that the AI for MWiF is not so complex as I think, please let us know, explain why and maybe you convince me. Let's debate.

No, it is not for you or others to be convinced, or for us to debate. Steve has his plan. The plan appears reasonable. It is for us to be patient while the plan unfolds, slowly but surely. I will wait.

[Centuur]

Steve has got his plan and it's not for us to be convinced if that plan will result in a good AI or not. That's something which can be concluded if the AI is there. However, to say that there should not be a debate about this, is not something which is productive. Good ideas are sometimes the result of a good debate...

Now, I'm not a programmer and I don't know anything of how an AI is coded, but one thing I'm convinced about. The more input somebody gets who has to do something, the better he can decide how to proceed. There's been a lot of discussions (of which a lot can be find in this forum if one looks at old topics) done for the AI in the past. So, if one really wants to see what's been done and also how Steve thinks about this AI, I suggest some good reading. Comments and questions are always appreciated. Discussions are always welcome.

But to simply say that it will be a good or bad AI? That's something only time can tell. To build an AI for this game is the real challenge Steve wants to take on. That's why he started coding this monster in the first place...

[Kull]

ORIGINAL: Centuur

Steve has got his plan and it's not for us to be convinced if that plan will result in a good AI or not. That's something which can be concluded if the AI is there. However, to say that there should not be a debate about this, is not something which is productive. Good ideas are sometimes the result of a good debate...

Now, I'm not a programmer and I don't know anything of how an AI is coded, but one thing I'm convinced about. The more input somebody gets who has to do something, the better he can decide how to proceed. There's been a lot of discussions (of which a lot can be find in this forum if one looks at old topics) done for the AI in the past. So, if one really wants to see what's been done and also how Steve thinks about this AI, I suggest some good reading. Comments and questions are always appreciated. Discussions are always welcome.

I completely agree. The main reason I posted the full list of reports was to try and generate a more useful discussion. Based on Steve's plan and what has ACTUALLY BEEN DONE, what do people think? Are there specific areas where folks have ideas on how the plan could be improved? Are there areas where perhaps others could assist?

And thanks for noting that other AI suggestions have been posted. Anyone willing to do some research would do us all a service by posting some links (and perhaps a synopsis).

But to simply say that it will be a good or bad AI? That's something only time can tell. To build an AI for this game is the real challenge Steve wants to take on. That's why he started coding this monster in the first place...

In complete agreement again. My problem with many of the posts in this thread is they amount to little more than personal opinions - "it's too complicated" or "it will never work". Gee thanks, that is so helpful.

Edit: WIF is an extremely large, hex based war game, but it's not the only one. WitE has 25K hexes and over 4000 units yet still features a competent AI, so comparisons with chess and GO are meaningless. The "standard" isn't an AI capable of beating grand masters, but rather something that is reasonably competent by commercial war game standards.

[Kull]

To hopefully move the discussion in a more positive direction, here are two things which I find to be particularly encouraging:

1) Decision Points:

There are 143 decision points in the sequence of play where the AIO needs to decide what to do. 122 of these use universal logic, in that they do not depend on which major power is making the decision (e.g., rail moves, strategic bombing, naval interceptions, placing partisans). For the other 21, the logic does depend on which major power is making the decision. Examples of those are: strategic plans, declarations of war, production plans, and trade agreements. I have 200 pages of text describing strategic plans. Each major power has their own section and the plans are well organized. For the French, I’ve started encoding their strategic plan as data.

For the 122 ‘universal’ decision points, I start by writing plain text that describes how the decision is to be made. I’ve done 73 of them as plain text so far. The second step is to translate the plain text into a LAIO script. All of that work remains to be done.

Comment: Since over 85% of the decision points are "universal", that consolidates a lot of the AI code. When the same subroutine utilized by the US is also used by Germany (and everyone else), that should greatly reduce the coding effort. It also means a fix for one is thereby a fix for all. Furthermore, 60% of the universal subroutines have already been created, just not coded.


2) AI Code Runs "in the background":

Another change I made for NetPlay, that has direct bearing on the AI Opponent code, was that I created a separate thread to run some code “in the background”. It is a minor technicality but I am happy to now have a working example of how to code background threads using Delphi. Virtually all the AIO code will execute in separate (i.e., background) threads. The intent is to have the AIO figure out what to do, while its human opponent is moving and clicking the mouse and using the keyboard. When the time comes for the AIO to decide something, it should be able to do so quickly.

Comment: To me, this is HUGE. In most strategy wargames, you have an IGO/UGO system where the human player completes their turn, and THEN the AI starts processing. In WIF, most of that processing will go on DURING the human players turn. So not only will the computer turns appear to proceed more rapidly, but it also means the code can be written to do a LOT of different things without having to worry about imposing huge delays during the computer-player's turn.

[Klydon]

As a long time veteran computer war gamer (my first computer was a Tandy color computer I got in 1981 that had a whopping 16k of memory) and in chatting with a couple of my friends that happen to have backgrounds in computer programing along with other things I have read, there is one big factor in favor of a "good" AI for MWIF.

The programmer has a clue how to play the game and I assume is at least decent at it. In addition, the programmer has gotten a lot of input from the MWIF community for strategy, etc. Other computer games really don't get that type of input and feedback. This is possible for MWIF because it is closely based on an existing board game.

Many times "bad AI's" are not a result of poor programing, but rather a result of the programmer not having a clue on how to really play the game they are prepping the AI for or not being very good at it. This all too often results in a "dumb" AI.

[pzgndr]

FWIW, I think Steve's approach is sound. Based on what I've read over the years. There's a high-level grand strategy routine to make the big decisions. Then there are operational-level routines to consider the strategies and the force levels in the various areas and assign forces to objectives, either offensive or defensive. And then there are the tactical-level routines to execute the nuts & bolts of game play. Sounds fine in theory. That's how people play, yes? Somewhere are numerous posts where players have submitted their setup and strategy ideas for each country, and AFAIK these have been considered (maybe not coded yet but filed away). Frankly, I'm not sure how much more debate and discussion Steve needs or wants right now? I'm not saying there should not be a debate, but I am suggesting this has already been debated over the years and there may not be much more to discuss until Steve gets into more AI work. There are other priorities and when the time is right he will circle back to working on the AI and probably ask for more comments and suggestions. I'm OK with all that.

[Hansstory]

I greatly admire Steve for his work and am glad he has done it! So I don't mean to personally attack him, but I need to use an example. I think in the beginning he estimated that the game would be finished and ready to release in like 24 months. (It could have been 18 or 30) That was his estimate. He was off by an order of 4 to 5 if I count correctly. He underestimated the work necessary and how long it would take to do the easy part; make the game enforce the rules. I believe, with all due respect to Steve, that he is too optimistic about the ability of anyone to code the AI to be good enough to beat a reasonably good human player on a global scale and probably a small front of the war. It has to do with the fact that the computer is "blind" and unable to make the value judgments we can make where we take variables into consideration and make a judgment call. For example, with air units, we judge their type, speed, attack etc in deciding where to place them, and why to place them. The computer can't be programmed to do that for each chit and to relate that evaluation to other chits as well. It isn't lack of drive, or skill on the part of the programmer, it is because a computer still can't think like we do. There is no HAL out there. And all the AI programmers that have come before Steve have failed to, for the same reason. Can AI be developed that does a relatively good job for some games, yes. But not for a game like this.

[joshuamnave]

ORIGINAL: Kull

My "personal opinion" on the AI is completely irrelevant. The reason I posted all the the AI reports was so people would have actual data to draw upon, and could critique (or applaud) based on the plan and the actions taken to implement it. Once again you've declined to make a case based on the facts, but that's the only thing I'm interested in discussing.

Steve has stated that he has a plan for the AI, and has presented a sizable amount of information on that subject. Accordingly, the burden is on YOU and all those who those who naysay the possibility of a functional AI. What is wrong with Steve's plan? Why won't it work?

2 years ago we had the last "state of the game" put together by Eric and Steve. In it was a list of priorities. Although no set time frame was given, the post said they expected to have them finished over the next six months. Two years later, not ONE of those items has been checked off the list.

Coming up on 2 months ago we were promised a new state of the game. A month went by with not a word until I asked about it. Turns out that there was a plan to do one, but Eric had family issues and Steve had health issues, but we would get it soon. Apparently nobody else in the company was in on the plan or they might have stopped by the forum and given us a heads up.

What's my point? I don't doubt there is a plan. There is always a plan. Matrix has given me ample reason to doubt the execution of that plan.

When you consider that five years (we're up to five now, right? I've lost track) the AI for convoy routing still does not work, all the plans in the world are meaningless to me.

I plan to be the first man on Mars. The burden is on you to explain why that won't happen.