Union AI totally broken

[Mickey3D]

Chess, Checkers, Go, and Bridge have small rulesets but are very complicated games (i. e., they are very 'subtle').

You can't compare complexity/subtility of Chess, Go, Bridge, AACW without defining what subtility is ? Number of possible move ? Something else ?

Moreover, I don't see the subtility in a game that has been solved by a piece of silicium running a dumb algorithm (just after "4 in a row" and "othello" ... just kidding :neener .

In fact, comparing computer and human brain has no meaning : both of them are not "wired" the same way. Human brain will see complexity in game with high number of possible moves because it can't handle it in a straight (crude ?) way but is far (far, far, far...) better at adapting itself to unknown and seeing patterns where a computer would only be lost without specifically developped program (even neural networks are designed to solve a specific problem).

AACW has no solution because there are randoms introduced and the info is imperfect. AACW and its ilk are vastly simpler than CCG&B, much easier to play at a high level of skill.

Being able to play a game where randomness occurs and information is imperfect seems to me requires more "intelligent" design for a computer program than exploring the tree of possible moves.

[Gray_Lensman]

deleted

[gekkoguy82]

did anyone even know what electrons were during the civil war?

[Pat "Stonewall" Cleburne]

Yeah Gray. The number of possible combinations and moves in AACW is damn near infinite (many orders of magnitude larger than chess) as you can stack units on top of each other and have them in any of those 1500+ positions. There isn't enough computing power to consider every possible move and make the best decision so you have to make an AI that actually "thinks". I don't know what this guy is trying to prove.

[Jarkko]

Here's strong advice for playing the AI.

Check the "Activation" button in options and set it to "if checked, land forces with non activated commander can't move at all".

A very good suggestion which I have been using for a couple months now, even in a my test-MP. Cuts much of the silly ahistorical raiding, in addition to forcing the player to consider movements for the worst case scenarios ("What if this commander doesn't get activated for thre turns after this, where do I *really* want him to be?"). Gives a much more historical feeling with the stop-go action, and at least the game feels much less like Vietnam War.

[arsan]

Do you recommend this activation setting for both human and Ai or only for human??
Cheers

[GraniteStater]

You gentlemen do not understand what I wrote.

First of all, there is no need to use phrases like 'cute formulas', or 'what this guy [please use my handle, not 'this guy'] is trying to prove'. I wasn't proving anything, I was making a distinction, I was describing something. If you don't understand what the other fellow is saying, please ask for a clarification and not assume that he's trying to "prove" something.

The only reason I provided some numbers and acknowleged what seems to be a correction of my understanding is so that readers could understand that I was illustrating that 32 pieces on 64 squares obeying a small ruleset can generate some very large numbers, indeed.

Secondly, if I'm understanding what I read correctly, you folks don't get the point. Is anyone here going to seriously argue that AACW is a more difficult game than top-flight Chess, Go, or Bridge? If you are of that persuasion, then go to USCF, get a membership and start playing. Heck, just log onto Caissa and start playing some games. Or buy Chessmaster and set it to an 1800 rating and see how you do. As someone once said, it's amazing how much hot water even a Grandmaster can get into in ten moves.

Then go look up Steintiz vs. von Bardeleben (Hastings, 1895): this (link below) is the Wiki for Wilhelm Steinitz and it has a depiction of the game just before his 22nd move as White. Work your way through the permutations of the combination and keep in mind that he demonstrated a 10 move mate to the audience which would have been forced after the resignation on Black's 25th move.

http://en.wikipedia.org/wiki/Wilhelm_Steinitz (the game is on the right sidebar about halfway down)

There is no commercial game that approaches, even approaches, the intellectual endeavor of something like seeing from a dozen moves ahead what Steinitz saw and executing it while every important piece was 'en prise' AND forcing every move by the opponent while one slip meant an instant back-rank mate.

That's the point I'm making. Large rulesets, e. g., AACW, do not tell you anything about the complexity of the game. OTOH, even though the intellectual complexity of a commercial computer game is not in the same league as Chess, for example, it is probably more difficult to program, because of the size of the ruleset (more SQA to run) and the fact that you have extraneous factors involved (modelling of a universe or behavior, commercial considerations, audience expectations, etc.).

Open ended NxN games are solvable and lend themselves to sophisticated, but relatively straightforward techniques for implementing them in a program. The straight poop is that the best chess players are programs. In ten years they'll be beating the Anands of the world with depressing regularity.

AACW is not an 'OE NxN' game, by definition (neither is Bridge). That's all I'm saying; although AACW, HOI2, Railroad Tycoon, Pirates!, and a host of others are much simpler games (as a game, i. e., in order to play it well; very, very well for sustained periods in your life), interestingly enough, they're probably harder to program succesfully than Chess & its cousins.

Which is my way of saying that the North's AI is not 'broken.'

FURTHERMORE:

Just 'cuz something is an OE NxN game does not mean it's complex. Tic-tac-Toe is simple. Bridge, when played at the highest levels, is quite exacting, as difficult as Chess at times, and is not OE NxN. It's ruleset is somewhat small, maybe smaller than Chess's.

If I were rated in Chess and did my studying, I'd probably be around 1900 - short of an Expert. I played in some strong company at one point; my very good friend was around 2400, and his buddies were all at least 2100. I beat a 2200 player once - straight up. I play Bridge casually and can appreciate what really good players can do.

And I stink at Checkers.

And none of this was meant to denigrate the finest strategic ACW game on the market. It's just not as hard as Chess, that's all.

[Gray_Lensman]

deleted

[Generalisimo]

And none of this was meant to denigrate the finest strategic ACW game on the market. It's just not as hard as Chess, that's all.

I must say that I do not really see your point in what you are saying, nor the relation with AACW AI... even I have done an AI for a chess game, but I will not even know from where to start to make a new Athena AI with all the variables I will have to manage ...
But we are probably going way off-topic... so I will just stop here.

[Franciscus]

I will not argue with anyone. I just would like to say that IMHO Chess is probably the best game invented by Man, and even one the greatest achievements of the Human mind. No other game has a ruleset that fits probably one or 2 pages, can be understood and played by less-than-10 year old kids, and yet can also transform itself in one of the greatest intelectual challenges in the gaming (and not only gaming) world.

[Jarkko]

Do you recommend this activation setting for both human and Ai or only for human??

Don't know what Bigus meant, but I give it only for the human. In my experience the AI is usually way too aggressive for its own good, but as it finally seems to have figured it needs to move troops to winter-quarters, I think the attrition would be again too bad for the AI if the troops had to stop moving.

However, I do *not* give the AI any activation bonus and I do *not* remove the fog of war totally (but I give the AI a small advantage in the FOW, just don't remove it in full), as the AI does in my experience do stupid things when it sees un-garrisoned towns far behind the front (but it seems to act smarter when it sees just a little bit further). Thus the AI can move even when not in command, but it isn't in command too often to be able to do the suicide attacks. Meanwhile the player has to take very safe movement orders so that troops don't get stuck in the middle of nowhere (city hopping, or in worst case stick close to railroads and waterways, and for sure no daring movements through wilderness in autumns (if you fail the command roll a few times in row, those sorry bastards sitting in the mountains will sit there even when snow starts to fall)).

[GraniteStater]

Which is the difference that I'm pointing out. Chess programs, as powerful as they are, have not solved Chess, and they possibly may never do so because the search-space is so immense. They are powerful enough to beat the best humans now, though. IOW, the programming is 'finished', so to speak.

"Computer games", for lack of a better term, are a different animal. You could, theoretically, play AACW on your table, with dice and combat-results tables, etc., just like the old AH games of yore. You could play it with more than two opponents; it would take much longer to play. The program is the game, essentially.

A Chess program is not Chess; it is a program that is executing instructions to "solve" the "problem". AACW is not defined that way at all.

I hope this distinction is clear; it's been a long time since I dealt with Game Theory. I'd have to do some intense study to really be precise about the distinction and also see where I might be unclear in my own understanding.

Anyhow, you can't expect too much outta AIs, unless you have something that is extremely well defined, like Chess (not to say computer games are sloppy; mathematically well defined), which is much more amenable to programming.

@Franciscus: and Go is supposed to be harder than Chess.

[tagwyn]

Jarkko: "and, those who don't/won't count." That is me. t

[MFogal]

GraniteStater:

Unfortunately, I know quite a bit of game theory. I think that some of your statements about the complexity of chess are correct, but that you underestimate the difficulty of AACW. I'll try to address your points without boring everyone else:

1) Mathematically, a full-information, limited space game is solvable -- think tic-tac-toe. Chess is (mathematically, when we're modeling it as a game) identical to tic-tac-toe: limited number of moves, sequential moves, full information, etc. Therefore, chess has been proven to have a "solution" -- there exists a chess strategy where the first move is guaranteed at least a draw.

1a) Note: the solution for chess has not yet been found, although apparently, checkers has indeed been "solved."

2) Chess is a finite game -- extremetly large (more than all the electrons in the universe??) but still finite . Even so, humans are not cognitively capable of dealing with such a complex game -- certainly not with just 2.5 hours/side. Perhaps an unlimited computer somewhere can.

2a) From my perspective, this is what is fun about playing chess (or any game) -- I've got to come up with mental shortcuts because I'm just not smart enough to accurately compute all the possible permutations. More on this in a later post.

3) Bridge becomes much more difficult to model (in formal game theory) primarily because the model has to incorporate "luck of the draw." Similarly, AACW must incorporate the luck of the draw. e.g. Sometimes, the militia brigade will stop Stonewall Jackson's Corp go figure.

3a) To see the mathematical difficulty, imagine that you can describe an optimal bridge strategy given that you know where all the cards lay. Then, you've got to describe a different bridge strategy for each of the possible permutations of where the cards lay (given full information). Then you've got to optimize your strategy across different probability distributions of different permutations of where the cards lay (imperfect information).

3b) Note that game theory at all levels requires full rationality -- you assume your opponent is incredibly smart, which means that he will not make any dumb mistakes, which means you can compute what he does and does not have in his hand based on his past play and bidding ....

3c) As you can see, incorporating uncertainty makes the game theory modeling much more difficult. It is the rolling of the dice which makes AACW more difficult to "solve" (from a game-theoretic perspective.) To be honest, I don't know if there is an existence theorem for equilibrium in games of imperfect information (that is, if such games have "solutions"). You see, in the presence of imperfect information, sometimes "playing dumb" is a smart move ....

4) The misinterpretation you seem to have made is that, since 0-sum, finite, perfect information games have been solved and that since 0-sum, finite, imperfect information games have not been solved, that somehow perfect info games are "more complex." I believe the opposite is true.

4a) The formal complexity in both AACW and bridge is not just about the size of the board, but is a function rather of (a) not observing your opponent's move (unlike chess, checkers, go, etc.) and (b) rolling the dice (or dealing the cards).

You make a strong case for how difficult it is to create a good AI to play chess. All I said (way back when) is that, given how difficult it has been for lots of chess geeks (and I say "geeks" with affection) to come up with a good chess AI, it is difficult for me to imagine that the (relatively few) AACW modders (howsoever well-motivated and brilliant though they surely are) would have been able to do better.

I have some ideas about heuristics that might be programmed into the AI, but I'll leave those for another day.

[Gray_Lensman]

deleted

[GraniteStater]

We're way off topic here, but I'd like to clarify something.

...and GL. please. I'm not beating a drum here, just making one simple point - see my first post.

4) The misinterpretation you seem to have made is that, since 0-sum, finite, perfect information games have been solved and that since 0-sum, finite, imperfect information games have not been solved, that somehow perfect info games are "more complex." I believe the opposite is true.

*sigh* I was afraid of this. It's not a misinterpretation. I'm afraid you're reading too much into my statement. I don't beleive I implied what you're saying above. As a matter of fact, if you read carefully, you'll see that I brought up Tic-tac-Toe for that very point: OENxN are not necessarily more complex. I don't know if imperfect info and/or random chance games are necessarily more complex, either. It could be a case by case basis. It could be neurological; maybe some alien intelligence would find Chess to be a snap and Cribbage immensely difficult.

All I know (again, see my first post here - I was responding to a statement) is that Chess is much, much harder to play well than any computer game I've experienced. Magnitudes of order harder to play well.

Your Bridge discussion I can't address, it's be much too off topic, except to say that one could perhaps treat a hand as perfect info for programming purposes, although the game is not played that way.

And please, I'm not trying to 'win' or 'prove' anything here. It's really simple: Chess, at least, is much harder than AACW. AACW, though, is probably more difficult to program.

[Generalisimo]

And please, I'm not trying to 'win' or 'prove' anything here. It's really simple: Chess, at least, is much harder than AACW. AACW, though, is probably more difficult to program.

Again, no.
What is the "function" you use to define what is "a hard game" to be able to say so many times in this thread that "X is harder than Y"?

Do a simple exercise... try to play AACW like Chess, thinking in advance X turns of your opponents moves to "plan" your best strategy, it is just imposible to think ALL the posibilities you have in any given turn...

If you want a more real example, try to play against one of the best PBEM players we have around... (the spaniard) Manstein... and again, try to solve the problem the same way than in chess to be able to beat him.
It is just imposible.
That's why, it is WAY harder to make an AI for AACW than in Chess... the posibilities are endless... (well, maybe there is some enormous finite number, but I doubt any of us can calculate it in any way :neener .

[Generalisimo]

Anyway, going back on topic to what it is important to the AACW community... like Pocus said before, the community of WIA have made a HUGE effort using the tools at hand provided by the engine to be able to help the AI beat the player (AI special events, AI focus, AI priorities, etc).
Sadly, that road hasn't been taken/explored by many people yet in AACW... let's hope someone does it in the near future...

[Gray_Lensman]

deleted

[Generalisimo]

It really doesn't matter what functional definition GraniteStater might use to try to justify chess being more "complex" than AACW. The point is whichever one he wants to try to emphasize if applied to AACW would show AACW being just as complex or more so, to wit: thinking moves ahead, board position combinations, randomness, etc.

Exactly... that's why I really do not understand his position... and that's why I am trying to try to understand why he is "saying" that (Check GS, I didn't said "defending", because you didn't like that ).

Saying that AACW is more complex and/or harder than Chess doesn't mean in ANY way that Chess is easy to play... I am sure I KNOW how to play chess, but that doesn't mean I can beat Kasparov.
Also, I can play chess with my nephew that is 9 y/o... on the other hand, he cannot even understand what AACW is trying to represent...
Even less when I start talking about things like supplies, army organization, leadership bonuses, seniority, conscripts, etc, etc, etc...
For him, they are just "some pretty soldiers on the map"...

That's also one of the "beauties" of chess... you can "play" it with anyone... but it is WAY hard to master it....

Anyway, if some moderator can split this discussion from here it will be great... this really belongs to the General Discussion forum... we were talking about the Union AI and AACW AI.

[enf91]

Granite:
Sorry, man. It looks like you're outnumbered or, at least, there's some miscommunication going on. If the second is true, then either everyone else just doesn't understand you (been there) or you haven't found a way to communicate your point (also been there). How about we say this discussion is over and return to talking about the original point, if there is anything left to say about it?

[Gray_Lensman]

deleted

[Reverend Zombie]

We're way off topic here, but I'd like to clarify something.
Chess is much, much harder to play well than any computer game I've experienced. Magnitudes of order harder to play well.

+1

[GraniteStater]

Thank you, good Reverend.

Last and for all:

That's also one of the "beauties" of chess... you can "play" it with anyone... but it is WAY hard to master it....

My point, exactly.

As I said above (again, see my first post, it was a response to a statement), is anyone going to try to seriously assert that AACW is a more difficult game to master than Chess?

C'mon, think about it, willya? Has anyone here ever lost in AACW? I haven't played CSA yet, but in general, my experience with computer simulations is that there is no serious prospect of losing, at least not compared to, say, Chessmaster at an Expert or Master setting.

Can you lose in computer games? Sure, play some weak backwater in historical sims. Or the original Shogun:TW - it was quite common to end up with The Horde of Ungodly MegaStax at the end (which made you see why the Geisha was so important). Or Railroad Tycoon Platinum, an excellent game. Trying to get the Gold Medal for a scenario could be challenging; many times I had to settle for Silver or Bronze.

But in most game forums, a common theme is 'the AI ain't hard enough.' Most mods exist for (a) greater historicity, or (b) ramping up the challenge.

There is no "function", or yardstick that I know of, that measures the 'difficulty' of a game. It is subjective, but subjective is not a synonym for irrelevant or untrue. I know from my own experience, and I'm absolutely sure most people with similar experience would agree, that Chess is a much, much harder game to play consistently well than almost any commercial game published.

[Generalisimo]

C'mon, think about it, willya? Has anyone here ever lost in AACW? [color="Red"]I haven't played CSA yet[/color], but in general, my experience with computer simulations is that there is no serious prospect of losing, at least not compared to, say, Chessmaster at an Expert or Master setting.

No, really, no... you are comparing apples and oranges and reaching a completelly weird conclusion...

That's also one of the "beauties" of chess... you can "play" it with anyone... but it is WAY hard to master it....

My point, exactly.

If you just said that, I would have agreed with you. The problem is exactly that, you didn't said ONLY that.

As I said above (again, see my first post, it was a response to a statement), is anyone going to try to seriously assert that AACW is a more difficult game to master than Chess?

Yes, go and play a PBEM game against Manstein... then you tell me how easy it was...

C'mon, think about it, willya? Has anyone here ever lost in AACW? I haven't played CSA yet, but in general, my experience with computer simulations is that there is no serious prospect of losing, at least not compared to, say, Chessmaster at an Expert or Master setting.

Not in AACW, I haven't played it much really... but for example, I have lost many times in NCP.

Can you lose in computer games? Sure, play some weak backwater in historical sims. Or the original Shogun:TW - it was quite common to end up with The Horde of Ungodly MegaStax at the end (which made you see why the Geisha was so important). Or Railroad Tycoon Platinum, an excellent game. Trying to get the Gold Medal for a scenario could be challenging; many times I had to settle for Silver or Bronze.

But in most game forums, a common theme is 'the AI ain't hard enough.' Most mods exist for (a) greater historicity, or (b) ramping up the challenge.

That's what we have been telling you since the beggining: because of the COMPLEXITY of the game, it is IMPOSIBLE to create an AI that "thinks" like an human player in AACW.
On the other hand, thanks to the SIMPLICITY of Chess, you can make an AI that is smarter than the player (Deep Blue vs Kasparov).
Really, take your time to think about it.

There is no "function", or yardstick that I know of, that measures the 'difficulty' of a game. It is subjective, but subjective is not a synonym for irrelevant or untrue. I know from my own experience, and I'm absolutely sure most people with similar experience would agree, that Chess is a much, much harder game to play consistently well than almost any commercial game published

No, obviously, you haven't been playing MP games too much.

[GraniteStater]

I truly think you don't understand what I'm saying. I mean that literally.

I never said AACW was simple and avoided the term 'complex' as much as possible.

I did say that it's child's play next to Chess and will maintain that. I described one ruleset as being smaller than another's. Perhaps to some, appearances mean a lot, but don't be deceived by the elegant and spare and conclude that it's simplistic, or simple (not synonyms, BTW). AACW is not necessarily more complex than Chess just because it has more rules or more lines of code - I don't think you've been reading the posts or understanding them; I already covered that. I am distinguishing between games, as games, and the programming required for a particular program. I certainly hope that would be clear to the average reader.

PBEM is irrelevant - are you seriously going to maintain that PBEM AACW is in the same universe as Chess when it comes to difficulty of the game? C'mon, gimme a break, willya? If you seriously think that is so, you haven't played very much Chess, maybe none. I'd almost have to think that if you are going to assert that AACW is as hard as Chess.

And think about something for a minute - you are asserting that a game (AACW - which could be replicated on your kitchen table with cardbourd counters) is soooo hard, that a human being for an opponent is going to give you a more challenging experience than the AI.

Whereas in Chess, the opposite is true.

Willya listen to yourself, for the love of Mike? Think.

I'd say that demonstrates that the game which is almost impossible to beat when played by a computer (i.e., shall we say, close to 'perfect', 'flawless' play) and is easier when played by mere humans, is more difficult to play well than the former case.

Wouldn't you agree? Hmmmmmmmmmm?

That's it, I'm done. Feel free to assert there's little gnomes inside your computer, I don't care.

[Franciscus]

Well, at least I think you guys can all agree that AGEOD would not mind if AACW had at least 1% of the popularity and longevity of Chess...

[Generalisimo]

...

Really, I do not like your tone. So, for me, the discussion with you is over.
Have fun!

[Mickey3D]

I must say that I do not really see your point in what you are saying, nor the relation with AACW AI...

I think I understand a little bit more Granitestater :

He is saying that, from a human perception point of view, chess is more complex. I agree with him as human brain is not naturally "armed" to searched through deep and wide tree of possibilities but is more inclined to deal with incomplete information.

Inversely, from a programmer point of view, chess algorithm required to beat a human player is easier to design than AACW one.

Yes, Granitestater is not defining the function to evaluate complexity. But defining this function means you would be considering complexity from the programmer point of view and I think he is more interested by the human perception of complexity. How could we evaluate this perception ? Don't know. Perhaps should we look at our brain activity ?

But we are probably going way off-topic... so I will just stop here.

+1

[Generalisimo]

I think I understand a little bit more Granitestater :

He is saying that, from a human perception point of view, chess is more complex. I agree with him as human brain is not naturally "armed" to searched through deep and wide tree of possibilities but is more inclined to deal with incomplete information.

Inversely, from a programmer point of view, chess algorithm required to beat a human player is easier to design than AACW one.

Yes, Granitestater is not defining the function to evaluate complexity. But defining this function means you would be considering complexity from the programmer point of view and I think he is more interested by the human perception of complexity. How could we evaluate this perception ? Don't know. Perhaps should we look at our brain activity ?

When you can model the "intelligence" needed to win a Chess game with a simple algorithm that uses Brute force and a simple decision rule like the Minimax, it clearly shows you that the complexity of the game is not that big.
The problem is that a human brain just can't process all that info togheter and think in advance 20 turns, like a machine can while exploring all the posible solutions.
So, what does a player do to "win"?... think in advance X turns (2, 4, 10, etc depending on the ability of the player) and make the move knowing in advance the posible outcomes of that decision.

In AACW, that is just plain imposible. There is no way to explore posible solutions because the complexity of any type of solution to the game is just enormous when you look at the entire universe of posibilities that you have at hand.
So, what do you do?
Most of the time, you take decisions based on already proven success ... if you did X and win, you try to do X and win again...
That's a very simple strategy, but most of the people do it because there is no way to plan ahead the combinations of posible solutions to the "problem" you have in front of you.
So, because the problem is so big and complex, you just simplify it to take a decision.

In case you are interested in the programmer point of view of this problem, a good and simple book that will give you an insight of how difficult it is even the problem to "decide which way to go", is "AI Techniques for Game Devlopment", by Mat Buckland.
It is not new, but it is one of the few that covers Genetic Algorithms and even Neural Networks.
There are many books around, but I found that this one is quite simple if you already have some background on the area.

+1

+2

This should really be outside this subforum