The AI

[bo]

First an apology for my negativity about this coming game, but thats my middle name
Mr. negativity. I think I was totally frustratred with release dates that I did not use the noodle to well. I want to ask all programmers and other players who understand how an AI works. I know or think everything is scripted [you do this the AI does that] in every war game, and for the most part the AI cannot beat a good gamer [dont have to be great] Why is it that a decent computer chess game AI can crush very good chess players
with ease, I am not talking about Big Blue, I am talking about a game like Chessmaster which on level 6 beats my brains in and I am not a bad chess player. Not knocking Steve here or any other programmer in the industry of computer war games but what kind of breakthrough will it take to get an AI to compete with a human, or is this far down the road of wishful thinking can anyone tell me what it will take to do this, or are the programmers just going to keep on scripting with the same results.

Bo

[pzgndr]

Check out Using Strategic Command 2 AI to Play Advanced Third Reich

I wrote that about 18 months ago based on the Strategic Command 2 - Patton Drives East expansion. Since then there's been a Pacific Theater expansion and now an upcoming Global Conflict expansion. With each iteration, the generic AI algorithms have improved and the editable scripts have improved to provide better scripting. I've been working on my little Advanced Third Reich mod for like 5 years now and I'm pretty happy right now with the Global Conflict version results. The PDE version results were OK and considerably better than the earlier versions, but the point is it takes lots of time to fine tune everything. And often new game features and/or new scripting capability was needed so what I can do today was impossible to do several years ago.

I'm optimistic WiF can and will eventually provide a challenging computer opponent. I'm realistic enough to not expect it upon release. It may take time. In the case of Empires in Arms now 2 years old since release, the AI still has issues. Hopefully WiF won't experience the same growing pains EiA has, but players should be patient with the process nonetheless. The goal is to eventually have a great PC wargame with challenging computer opponent. Sooner would be better than later, but it may take time regardless. [:)]

[bo]

ORIGINAL: pzgndr

Check out Using Strategic Command 2 AI to Play Advanced Third Reich

I wrote that about 18 months ago based on the Strategic Command 2 - Patton Drives East expansion. Since then there's been a Pacific Theater expansion and now an upcoming Global Conflict expansion. With each iteration, the generic AI algorithms have improved and the editable scripts have improved to provide better scripting. I've been working on my little Advanced Third Reich mod for like 5 years now and I'm pretty happy right now with the Global Conflict version results. The PDE version results were OK and considerably better than the earlier versions, but the point is it takes lots of time to fine tune everything. And often new game features and/or new scripting capability was needed so what I can do today was impossible to do several years ago.

I'm optimistic WiF can and will eventually provide a challenging computer opponent. I'm realistic enough to not expect it upon release. It may take time. In the case of Empires in Arms now 2 years old since release, the AI still has issues. Hopefully WiF won't experience the same growing pains EiA has, but players should be patient with the process nonetheless. The goal is to eventually have a great PC wargame with challenging computer opponent. Sooner would be better than later, but it may take time regardless. [:)]

thank you pzgndr great article, I removed SCWA from my computer several months ago what do I have to do to get advanced 3rd Reich to play om my computer I loved the board game bought the computer game and removed the computer game after 2 months of frustration. I know I have to download it from a mod place to my computer, di i have to hace Patton goes east or will SC weapons and warfare do?

Bo

[Shannon V. OKeets]

ORIGINAL: bo

First an apology for my negativity about this coming game, but thats my middle name
Mr. negativity. I think I was totally frustratred with release dates that I did not use the noodle to well. I want to ask all programmers and other players who understand how an AI works. I know or think everything is scripted [you do this the AI does that] in every war game, and for the most part the AI cannot beat a good gamer [dont have to be great] Why is it that a decent computer chess game AI can crush very good chess players
with ease, I am not talking about Big Blue, I am talking about a game like Chessmaster which on level 6 beats my brains in and I am not a bad chess player. Not knocking Steve here or any other programmer in the industry of computer war games but what kind of breakthrough will it take to get an AI to compete with a human, or is this far down the road of wishful thinking can anyone tell me what it will take to do this, or are the programmers just going to keep on scripting with the same results.

Bo

Most AI programs for chess do not evaluate positions very well. What they do is use a look-ahead algorithm and minimax the end points. That is, they simply evaluate all possible moves, and moves in response to each of those moves, and moves in response to those moves ... Each side's move is labeled a "half move". When both White and Black have made a move that is considered a "full move". Mostly this labeling is historical and reflects how players record their games.

Now, at the end of each half move the program evaluates the position (using a fairly weak algorithm). It then tries to find the best move for itself, if it is the AI's turn to move; or the worst move for itself, if it the opponent's turn to move. This process has been called minimax, where you try to maximize your position when you move, and minimize your position when the opponent moves. Logically this is finding the 'best' move for each side. It explores all possible moves until it runs out of time (some fixed limit) or reaches 'quiescience' - a point where future moves do not change the overall evaluation.

So, the amount of 'knowledge' in the program about what constitutes a good chess move is completely contained within the evaluation scheme for each position. What is missing is any sense of purpose (e.g., attack on the king side, encourage exchanging pieces because we have a material advantage). Also sequences of moves that are 'forcing' can be underrated. When playing against a human opponent, your opponent only needs to get one of the 'forced' moves wrong and you will win. If the program looks ahead and sees that there is a good counter-move for each of 6 forcing moves in a sequence, it is likely to deem the sequence of moves of little interest. Basically the idea here is to give your opponent the opportunity to screw up.

The use of look-ahead and minimax is more or less a "brute force" approach to AI. If you throw in an excellent book or two on how to play openings and endings, the AI chess opponent can be quite powerful.

Any war game that involves dice or hundreds of counters can not be programmed using look-ahead + minimax. There are simply too many possible outcomes, millions/billions/trillions times more than there are in chess. Instead, the programmer has to do the hard work of having strategic, operational, and tactical plans.

But enough on this (one of my favorite topics).

[pzgndr]

Bo, wait for Global Conflict?  I'll be posting my mods at the Battlefront repository after release, and I use the plural because I now have working 1939, 1941, 1942 and 1944 campaigns.  OK, shameless plug mode OFF.  [;)]
 
Seriously, I support what Steve's doing with WiF and I'm anxiously waiting for it to get released.  I offer my perspective on AI as just that, a perspective.  AI programming and scripting and all the rest isn't trivial.  It's hard.  But it's also doable!  [8D]

[bo]

ORIGINAL: Shannon V. OKeets

ORIGINAL: bo

First an apology for my negativity about this coming game, but thats my middle name
Mr. negativity. I think I was totally frustratred with release dates that I did not use the noodle to well. I want to ask all programmers and other players who understand how an AI works. I know or think everything is scripted [you do this the AI does that] in every war game, and for the most part the AI cannot beat a good gamer [dont have to be great] Why is it that a decent computer chess game AI can crush very good chess players
with ease, I am not talking about Big Blue, I am talking about a game like Chessmaster which on level 6 beats my brains in and I am not a bad chess player. Not knocking Steve here or any other programmer in the industry of computer war games but what kind of breakthrough will it take to get an AI to compete with a human, or is this far down the road of wishful thinking can anyone tell me what it will take to do this, or are the programmers just going to keep on scripting with the same results.

Bo

Most AI programs for chess do not evaluate positions very well. What they do is use a look-ahead algorithm and minimax the end points. That is, they simply evaluate all possible moves, and moves in response to each of those moves, and moves in response to those moves ... Each side's move is labeled a "half move". When both White and Black have made a move that is considered a "full move". Mostly this labeling is historical and reflects how players record their games.

Now, at the end of each half move the program evaluates the position (using a fairly weak algorithm). It then tries to find the best move for itself, if it is the AI's turn to move; or the worst move for itself, if it the opponent's turn to move. This process has been called minimax, where you try to maximize your position when you move, and minimize your position when the opponent moves. Logically this is finding the 'best' move for each side. It explores all possible moves until it runs out of time (some fixed limit) or reaches 'quiescience' - a point where future moves do not change the overall evaluation.

So, the amount of 'knowledge' in the program about what constitutes a good chess move is completely contained within the evaluation scheme for each position. What is missing is any sense of purpose (e.g., attack on the king side, encourage exchanging pieces because we have a material advantage). Also sequences of moves that are 'forcing' can be underrated. When playing against a human opponent, your opponent only needs to get one of the 'forced' moves wrong and you will win. If the program looks ahead and sees that there is a good counter-move for each of 6 forcing moves in a sequence, it is likely to deem the sequence of moves of little interest. Basically the idea here is to give your opponent the opportunity to screw up.

The use of look-ahead and minimax is more or less a "brute force" approach to AI. If you throw in an excellent book or two on how to play openings and endings, the AI chess opponent can be quite powerful.

Any war game that involves dice or hundreds of counters can not be programmed using look-ahead + minimax. There are simply too many possible outcomes, millions/billions/trillions times more than there are in chess. Instead, the programmer has to do the hard work of having strategic, operational, and tactical plans.

But enough on this (one of my favorite topics).

Thanks steve I heard a good chess player thinks 6 to seven moves ahead, I have trouble with more than two [&:] and also a chess programmer only has to worry about 64 squares not thousands. My offer on Tony Lukes still stands[:D]

Bo

[bo]

ORIGINAL: pzgndr

Bo, wait for Global Conflict?  I'll be posting my mods at the Battlefront repository after release, and I use the plural because I now have working 1939, 1941, 1942 and 1944 campaigns.  OK, shameless plug mode OFF.  [;)]

Seriously, I support what Steve's doing with WiF and I'm anxiously waiting for it to get released.  I offer my perspective on AI as just that, a perspective.  AI programming and scripting and all the rest isn't trivial.  It's hard.  But it's also doable!  [8D]

You have me confused about Global Conflict as I have not played SC anything for awhile, is it a stand alone mod an add on or what? Or will it be in the Repository? What game do you have to have loaded if it is an add on?

Bo

[pzgndr]

Bo, Global Conflict will be stand alone game offered by Battlefront soon, and my mod will be a freebie posted at their file repository soon after release.  I've been focused on the ETO gaming and not so much on PTO and global campaigns; that's just me.  There will be some obvious similarities and differences between WiF and Global Conflict.  I spoke up here about the AI and don't want to go off on a tangent.  So again, the AI issues I've discussed in my Wargamer article are likely the same issues Steve is wrestling with and I'll be interested to see how it all turns out.  My one recommendation would be to develop AI vs AI playtesting capability for WiF to flesh out and resolve AI issues as soon as practical. 

[Skanvak]

Thanks steve I heard a good chess player thinks 6 to seven moves ahead, I have trouble with more than two and also a chess programmer only has to worry about 64 squares not thousands. My offer on Tony Lukes still stands

Bo

Actually, even chess has too much possibility for a computer to calculate. Because chess is a finite game (ie we know from start who will win, that is a mathematical theorem. According to Kasparov the white win). Once the computer can calculate all move, you will always lose against it. It is more that thinking 6 moves ahead; it is writing a mapping of ALL possibles moves in the game and going back ward from final situation where you win (try with the X O game to see how it goes).

The problem with wargame (increase in multi-player and wego system) is that you have to program a guessing algorythm and there is no mathematical model for that (the game theory only give a random strategy). Of course with game as big as WiF even in 2 players, the possibility are huge...

[bo]

ORIGINAL: pzgndr

Bo, Global Conflict will be stand alone game offered by Battlefront soon, and my mod will be a freebie posted at their file repository soon after release.  I've been focused on the ETO gaming and not so much on PTO and global campaigns; that's just me.  There will be some obvious similarities and differences between WiF and Global Conflict.  I spoke up here about the AI and don't want to go off on a tangent.  So again, the AI issues I've discussed in my Wargamer article are likely the same issues Steve is wrestling with and I'll be interested to see how it all turns out.  My one recommendation would be to develop AI vs AI playtesting capability for WiF to flesh out and resolve AI issues as soon as practical. 

thank you pzgndr, I went to the game sight maybe some time in jan or feb, now your mod is it based on 3rd Reich, meaning is it a 3rd Reich map with 3rd Reich rules?

Bo

[vonpaul]

I hope i'm not just repeating what has been before....and i'm definetly no Troll

I check in on the forums every couple of months and have seen we are still looking at quite a while till release (partly because of the AI).

I think AI is a 'bridge too far' at this junction. It will take ALONG time to get a even decent AI and a bad one will just damage the games reputation. Its either a release in 2011(12?) or concentrate on getting bug-free solitare and kick-arse multiplayer experience for release this year. Then develop & implement AI as an expansion (deluxe) release. The hardcore will buy without an AI (along as it is better than Vassal) and I wouldn't mind paying for an expansion to then include an AI (if makes a good opponent). Just look at Paradox (HOI3) so see how NOT to release a game, they have done permanent reputation damage with their attempts at developing AI on the cheap (I will never buy their games on release again). Know your limitations and work to them i reckon.

I think a game this niche needs to have a very good reputation to get many non-WIF players to buy. Building a community of fans of Matrix WIF (as opposed to just the boardgame) will help achieve some buzz about the game again.

Then again maybe Steve's contract says must have functional AI and its all a moot point

Happy to hear how you think a basic AI will be anything more than a hindrance in a game of WIF's scope and whether you think it will do more harm than good releasing with a 'tutorial AI'?

[pzgndr]

I think AI is a 'bridge too far' at this junction... The hardcore will buy without an AI (along as it is better than Vassal)

I'm curious, since we have this same silly debate over on the Empires in Arms forum. What exactly is wrong with Vassal and its game modules, such as World in Flames? If all the "hardcore" wants is pbem software without a computer opponent, then it's already available. If Vassal is unsatisfactory then where is the customer demand on their forums for improvements and enhancements?? [&:]

Then again maybe Steve's contract says must have functional AI and its all a moot point

And maybe Matrix Games being in the computer wargame business and not the pbem software business recognizes a customer demand for AI and computer opponents. And Matrix Games seems to be doing pretty well for itself, yes? So rather than criticizing Matrix Games for supporting AI development, why not plant yourself over on the Vassal forum and criticize them for not offering better pbem software features? Just a thought. [:)]

[BallyJ]

It wasn't criticisem. It was an inquiry.
I have tried vassel.
I know I am a computer hopeless.
I can not get it to work!~@#$$%^
My failure I know.
THAT IS THE REASON I AM WAITING FOR THIS GAME TO COME OUT

[paulderynck]

ORIGINAL: BallyJ

It wasn't criticisem. It was an inquiry.
I have tried vassel.
I know I am a computer hopeless.
I can not get it to work!~@#$$%^
My failure I know.
THAT IS THE REASON I AM WAITING FOR THIS GAME TO COME OUT

Vassal should work for you unless your computer is old and has very little memory. I think I'd be surprised if a computer could run MWIF but not Vassal and vice versa. And I'd expect you'd get help from the forums if needed.

But Vassal simulates playing the game FTF. So players won't get any help from the computer setting up, regulating movement, calculating odds, etc., along with a myriad of other things computers are good at.

One really neat feature of Vassal is that you can switch back and forth between online real time play and PBEM.