Revolutionary Thoughts

[Thresh]

The inclusion of dice does not mean luck and chance area part of the game, it means there are additional  probabilities to take into account.  There are several layers of probability in the game, each adding a layer of complexity that takes newer (and sometimes older) players more than a few games to get through, and someof us still struggle.

Luck is probability taken personally.  If you don't take the time to consider all the odds involved (and in EiA that can be quite a few), the you shouldn't be surprised when you lose when you "think" you should win.  More often than not, the odds of you actually winning were a lot less than you think they were.  Yes, you may have had a chance to win, but the odds of success were not in your favor, making that particular course of action bad.

Army composition, leader effects,terrain effects, and chit picks are all things that need to be taken into account, and IMO chit picks are the hardest to judge, especially in PbEM games because it's harder to gauge an opponent that way.  They add a level of probality thats hard to quantify because they are not set instone (like die rolls are).  Sure Mack be a 1.3.4 leader, but what if he does pull the outflank "just because?"  The probability of that happening is quite low, but its these elements that make such probabilities  harder to take into account.  Die rolls are easy, I can guarentee the odds on those every time, but the additional elements of chit picks and potential morale changes add to thelevels...

But to simply make a move, roll some dice, and then explain things away as bad luck is simplistic inthe extreme.

Todd

[NeverMan]

Dice are a part of the game. You cannot fully account for dice, I don't care how good your strategy is. I've seen VERY expereinced and GOOD players that have good strategies lose a war against a newbie because of dice rolls. Its utterly amazing what a 1 vs. 6 will do a few times in a game.

[MarshallEllis]

ORIGINAL: NeverMan

ORIGINAL: Marshall Ellis

Well for my 2 cents, from a programming standpoint EiANW was VERY complicated! Don't get me wrong, I think the EiA design is awesome and very well balanced but balance comes in different ways.

For example:

Turkey is poor BUT she has the feudals.
Great Britain has few troops BUT she has a huge navy.
Spain is lower middle class but has a decent navy, Spanish gold and guerillas.
bla bla bla

My point is that this game is not a linear balanced system like Risk where everybody starts with the same amount of armies and nations. This creates all sorts of special code for a specific nation i.e. Turkey can take a ONLY a feudal unit on the board and move it to its home province and raise it to full strength BUT only Turkey can do this and ONLY in December LOL! I'm surprised I didn't get a drug problem from those type of requirements!

As far as complexity? You bet! No one can tell me that doing a supply step or eco phase is easier than Risk. This is simply a complex game but these are 2 different things IMO. The game flow is complex because of all the rules but this is not what makes it difficult to master that is where the design comes in. It's simply a great game design.

Anyway, my unsolicited two cents.

Marshall, the whole Risk analogy was purposefully silly (since Risk is such a silly game) for guys like Guy and Jimmer who believe that simply because a game has more rules that it is harder to master, which is just absurd.

Please don't compare Risk to EiA, as I have not. Risk is a VERY simple game that I have had no desire to even look at since I was 12 (started playing EiA at 15).

It is also obviuos that niether Guy nor Jimmer have played Chess even at the most intermediate level.

Neverman:

Don't take this wrong, I was not using (or trying to slam) your Risk analogy but using my own Risk analogy to demonstrate linear balance versus EiA balance. We're on the same page.

[MarshallEllis]

Hey guys:
 
I honestly think there ar multiple aspects that we are debating.
I don't think this is apples to apples but:
 
Chess is a VERY simple game to play BUT extremely difficult to master.
 
EiA is difficult to play and really cannot be mastered. This is no slam to Chess since this is just not a good comparison. The discussion about dice is spot on. Chess totally relies on your brain to win. In EiANW if my 8 year old rolls sixes the entire game (Understanding this is a mathematical near impossibility) then he will smoke anybody that does not roll sixes the whole game LOL!
 
 
 

[eske]

From the programmers perspective assessing the complexity of a game has nothing to do with his ability to play or even understand the game.
In games like chess and go, where two players alternately change the state of the game (makes a move) one undividable action at the time, the program only needs two parts to play the game: a gametree searching algorithm and an evaluation function. If the computer has time to generate the complete gametree the evaluation algorithm only needs to determine who won, and the program will play perfectly. And the programmer has a method of building this program without ever having played it. Not difficult. Just work. The more special situations the game contains, the more work.
Fortunatelty for the enjoyment of playing chess or go, the gametree for those games grows so fast for each move forward, that the program most of the time cannot build a complete gametree. A chess player has on average 30 or so possiible moves each turn. So two moves (I go, you go) generates 900 positions or nodes in the game tree. Two moves more, and it is 810000 positions. Go is a filling game with 361 possible opening moves, then 360, 359, 358 etc. From the beginning four moves generates 16702719120 positions. This is the way game complexity is measured mathematically. And go is thus more complex than chess. And to no ones surprise it has proven harder to make a strong go program than a strong chess program.
The mathematical game-complexity does however not tell much about how hard it is to play, since the human brain basicly is a pattern recognition device, which is not very fast at searching gametrees.
 
And to conclude this little lecture the game complexity of EiA cannot be measured this way. There are 7 players, each players phase contains multiple actions, that changes the game state (with 10 land corps each having at least 10 independant possible ending positions each player has at least 10 billion possible land moves!), a player does not have full information of the game state, part of the game is random, part of the game is diplomatic negotiations (And I could go on...)
So it is not possible to build a complete gametree and it makes no sense to do it. There is no way to compare the game complexity of EiA with that of chess and go objectively.
 
So what is the programmer to do, to address this hydra of a game. I can think of three tools: divide the game into a number of smaller interacting games, which is more or less what EiA is. The cost of this is local optimization, where decisions lack global purpose. Use rules of thumb (heuristics) to compare positions. And use rules of thumb to make near-optimal decisions.
 
All in all a totally different ballgame - for the programmer. What he needs most to build a system of thumbrules is lots and lots of game-experience. And we - the players - are the largest resource there is. So guys. If you want this program to play well, you better volunteer your little secrets to the benefit of the programmer, right !
 
Now I'm sure Marshall has plenty to do, so it has to be only the essence in its purest possible form, to increase direct usablilty, ok? And I'm sure the AI's gameplay will grow with each update to come. 
 
Now back to your revolution, Mardonius
 
/eske
 
 

[Mardonius]

Thanks Eske. But now I have had a few moments during this math storm to throw a few dice and sharpen my guillotines. Got the mob with pitchforks and torches too.

I would think that we might want to find some threshold that indicates AI fruition, lest -- for reasons noted above -- we chase those diminishing marginal returns down little rabbit holes until our halcyon days. I'd love to hear what an acceptable AI is.

Bugs, of course, need to be reduced to enhance playability. I fear, too, that at some point -- don't think we are there yet -- that these enhancements will become asymptotical as well and we will never quite get there.

And so my revanche, my rejoinder, my storming of the Bastille! We should soon dedicate some portion of our work on game options (OPTIONS) including some of those, perhaps I noted above and some of the classical EiA options/scenarios as well. I, for one, would love to see the French Revolutionary Scenario, maybe a Seven Years War Scenario etc.

Sic Semper Tyrannis,

Mardonius

[gwheelock]

ORIGINAL: Mardonius

Thanks Eske. But now I have had a few moments during this math storm to throw a few dice and sharpen my guillotines. Got the mob with pitchforks and torches too.

I would think that we might want to find some threshold that indicates AI fruition, lest -- for reasons noted above -- we chase those diminishing marginal returns down little rabbit holes until our halcyon days. I'd love to hear what an acceptable AI is.

Sic Semper Tyrannis,

Mardonius

I would suggest a variation of the "Turing Test".

A recap for non-programmers - (taken from Wikipedia) :

[font="Arial"]The Turing test is a proposal for a test of a machine's capability to demonstrate intelligence. Described by Alan Turing in the 1950 paper "Computing Machinery and Intelligence," it proceeds as follows: a human judge engages in a natural language conversation with one human and one machine, each of which try to appear human; if the judge cannot reliably tell which is which, then the machine is said to pass the test. In order to test the machine's intelligence rather than its ability to render words into audio, the conversation is limited to a text-only channel such as a computer keyboard and screen (Turing originally suggested a teletype machine, one of the few text-only communication systems available in 1950.[/font]

(FYI - The Turing test hasn't [yet] been met except in very restricted conditions [some people have
mistaken a program called "Eliza" for a human who was making fun of them & got rather upset with "him"])

What I would suggest as a measure of an AI "good enough" to be labeled as "done" would be if you cannot
tell the difference between the AI's play or that of an AVERAGE (not EXPERT) human player; then I would
conclude the AI is done.

[MarshallEllis]

Interesting test. I like that. "Done" will be tricky. I don't think it will ever please all people so there will always be an element that thinks it should do more or act differently.
 

[DCWhitworth]

ORIGINAL: gwheelock

ORIGINAL: Mardonius

Thanks Eske. But now I have had a few moments during this math storm to throw a few dice and sharpen my guillotines. Got the mob with pitchforks and torches too.

I would think that we might want to find some threshold that indicates AI fruition, lest -- for reasons noted above -- we chase those diminishing marginal returns down little rabbit holes until our halcyon days. I'd love to hear what an acceptable AI is.

Sic Semper Tyrannis,

Mardonius

I would suggest a variation of the "Turing Test".

A recap for non-programmers - (taken from Wikipedia) :

[font="Arial"]The Turing test is a proposal for a test of a machine's capability to demonstrate intelligence. Described by Alan Turing in the 1950 paper "Computing Machinery and Intelligence," it proceeds as follows: a human judge engages in a natural language conversation with one human and one machine, each of which try to appear human; if the judge cannot reliably tell which is which, then the machine is said to pass the test. In order to test the machine's intelligence rather than its ability to render words into audio, the conversation is limited to a text-only channel such as a computer keyboard and screen (Turing originally suggested a teletype machine, one of the few text-only communication systems available in 1950.[/font]

(FYI - The Turing test hasn't [yet] been met except in very restricted conditions [some people have
mistaken a program called "Eliza" for a human who was making fun of them & got rather upset with "him"])

What I would suggest as a measure of an AI "good enough" to be labeled as "done" would be if you cannot
tell the difference between the AI's play or that of an AVERAGE (not EXPERT) human player; then I would
conclude the AI is done.

I think the Turing test would be an impossible target and in my opinion an unrealistic one. The fundamental fact is that humans do not think like computers so getting a computer to play like a human is even harder than getting it to play competantly.

Computers can play chess better than humans but they don't think about it the same way and they don't play like humans. To beat a human and a chess computer over the board requires different optimal techniques (when playing at a high level), but the computers eventually outpaced the humans, not by playing chess 'better' but by sheer brute force computing power.

In fact the entire computer chess development is a dead end as far as real AI is concerned. It doesn't give any sort of insight into true intelligence, it's just brute force computing in a limited environment.

[Mardonius]

Gents:

I think it would be a pretty poor use of time and resources to pursue the level of AI mentioned in the Turing Test. I am no genius, but if the test has not been met in 50 years, is it a smart level of AI to pursue? Hmmmm....

So, then, I posit: if this medium level of intelligence is not realistically obtainable, as the Turing Test pursuit reveals, why are investing in further AI development?

And if we do think we should continue, what is a realistic (REALISTIC) goal? When will be satisfied?

Mssr Withworth is spot on when he points out that the brute force calculations ala chess won't even come close to the diplomacy subtleties of EiA. So good luck with trying to get a HAL equal to kick its fellows out the pod bay door.

I think we should all set our expectation-phasers to "AI=teaching tool". This is, most of all, a diplomacy game. Some may differ, but they'd be wrong.

Just some thoughts...

But those options (OPTIONS) might not be so hard to achieve. Maybe we should put some of our efforts there?

On with the Revolution!!!

Semper Fi,
Mardonius

[NeverMan]

ORIGINAL: DCWhitworth

ORIGINAL: gwheelock

ORIGINAL: Mardonius

Thanks Eske. But now I have had a few moments during this math storm to throw a few dice and sharpen my guillotines. Got the mob with pitchforks and torches too.

I would think that we might want to find some threshold that indicates AI fruition, lest -- for reasons noted above -- we chase those diminishing marginal returns down little rabbit holes until our halcyon days. I'd love to hear what an acceptable AI is.

Sic Semper Tyrannis,

Mardonius

I would suggest a variation of the "Turing Test".

A recap for non-programmers - (taken from Wikipedia) :

[font="Arial"]The Turing test is a proposal for a test of a machine's capability to demonstrate intelligence. Described by Alan Turing in the 1950 paper "Computing Machinery and Intelligence," it proceeds as follows: a human judge engages in a natural language conversation with one human and one machine, each of which try to appear human; if the judge cannot reliably tell which is which, then the machine is said to pass the test. In order to test the machine's intelligence rather than its ability to render words into audio, the conversation is limited to a text-only channel such as a computer keyboard and screen (Turing originally suggested a teletype machine, one of the few text-only communication systems available in 1950.[/font]

(FYI - The Turing test hasn't [yet] been met except in very restricted conditions [some people have
mistaken a program called "Eliza" for a human who was making fun of them & got rather upset with "him"])

What I would suggest as a measure of an AI "good enough" to be labeled as "done" would be if you cannot
tell the difference between the AI's play or that of an AVERAGE (not EXPERT) human player; then I would
conclude the AI is done.

I think the Turing test would be an impossible target and in my opinion an unrealistic one. The fundamental fact is that humans do not think like computers so getting a computer to play like a human is even harder than getting it to play competantly.

Computers can play chess better than humans but they don't think about it the same way and they don't play like humans. To beat a human and a chess computer over the board requires different optimal techniques (when playing at a high level), but the computers eventually outpaced the humans, not by playing chess 'better' but by sheer brute force computing power.

In fact the entire computer chess development is a dead end as far as real AI is concerned. It doesn't give any sort of insight into true intelligence, it's just brute force computing in a limited environment.

This is not true regarding Chess AI. Modern chess AI incorporate genetic algorithms and neural networks, two very strong AI topics in modern computing. The "chess only uses brute force" is a computing myth and can be quickly dismissed with any "google scholar" search for whitepapers.

Does it use search functions? Absolutely, but don't you? Don't you think in steps when playing Chess? You do more than also and SO DOES the AI most of the time. Most AI's will search deeper when finding a line that is interesting and seems (where the AI comes in) to be advantageous, something a human player does all the time.

You can look at a line and know it's going nowhere but you can also see a line and then think "wait a minute, that might lead to something, I need to spend more time thinking about this" and SO DOES THE AI.

I don't think it's fair to dismiss an AI's thinking methods simply because you can explain them in some ISA and you fail to be able to explain your own. Their is no magic behind your thinking, we just haven't discovered the explanation behind the method yet.

[eske]

ORIGINAL: NeverMan
This is not true regarding Chess AI. Modern chess AI incorporate genetic algorithms and neural networks, two very strong AI topics in modern computing. The "chess only uses brute force" is a computing myth and can be quickly dismissed with any "google scholar" search for whitepapers.

Does it use search functions? Absolutely, but don't you? Don't you think in steps when playing Chess? You do more than also and SO DOES the AI most of the time. Most AI's will search deeper when finding a line that is interesting and seems (where the AI comes in) to be advantageous, something a human player does all the time.

You can look at a line and know it's going nowhere but you can also see a line and then think "wait a minute, that might lead to something, I need to spend more time thinking about this" and SO DOES THE AI.

I don't think it's fair to dismiss an AI's thinking methods simply because you can explain them in some ISA and you fail to be able to explain your own. Their is no magic behind your thinking, we just haven't discovered the explanation behind the method yet.

Now you made me curious, Neverman. Which chess programs uses genetic algorithms and/or neural betworks ?
- and which chess programs uses modern chess AI ?

And - to make this discussion relevant to EiA - which part of modern chess AI can be applied to which area of EiA ?

/eske

[NeverMan]

ORIGINAL: eske

ORIGINAL: NeverMan
This is not true regarding Chess AI. Modern chess AI incorporate genetic algorithms and neural networks, two very strong AI topics in modern computing. The "chess only uses brute force" is a computing myth and can be quickly dismissed with any "google scholar" search for whitepapers.

Does it use search functions? Absolutely, but don't you? Don't you think in steps when playing Chess? You do more than also and SO DOES the AI most of the time. Most AI's will search deeper when finding a line that is interesting and seems (where the AI comes in) to be advantageous, something a human player does all the time.

You can look at a line and know it's going nowhere but you can also see a line and then think "wait a minute, that might lead to something, I need to spend more time thinking about this" and SO DOES THE AI.

I don't think it's fair to dismiss an AI's thinking methods simply because you can explain them in some ISA and you fail to be able to explain your own. Their is no magic behind your thinking, we just haven't discovered the explanation behind the method yet.

Now you made me curious, Neverman. Which chess programs uses genetic algorithms and/or neural betworks ?
- and which chess programs uses modern chess AI ?

And - to make this discussion relevant to EiA - which part of modern chess AI can be applied to which area of EiA ?

/eske

Modern chess AI incorporate genetic algorithms and neural networks, as I said, so the two are not mutally exclusive as you suggest in your question.

I suggest you start at "google" or "google scholar" to satisfy your own curiousity.

[eske]

ORIGINAL: NeverMan

ORIGINAL: eske

ORIGINAL: NeverMan
This is not true regarding Chess AI. Modern chess AI incorporate genetic algorithms and neural networks, two very strong AI topics in modern computing. The "chess only uses brute force" is a computing myth and can be quickly dismissed with any "google scholar" search for whitepapers.

Does it use search functions? Absolutely, but don't you? Don't you think in steps when playing Chess? You do more than also and SO DOES the AI most of the time. Most AI's will search deeper when finding a line that is interesting and seems (where the AI comes in) to be advantageous, something a human player does all the time.

You can look at a line and know it's going nowhere but you can also see a line and then think "wait a minute, that might lead to something, I need to spend more time thinking about this" and SO DOES THE AI.

I don't think it's fair to dismiss an AI's thinking methods simply because you can explain them in some ISA and you fail to be able to explain your own. Their is no magic behind your thinking, we just haven't discovered the explanation behind the method yet.

Now you made me curious, Neverman. Which chess programs uses genetic algorithms and/or neural betworks ?
- and which chess programs uses modern chess AI ?

And - to make this discussion relevant to EiA - which part of modern chess AI can be applied to which area of EiA ?

/eske

Modern chess AI incorporate genetic algorithms and neural networks, as I said, so the two are not mutally exclusive as you suggest in your question.

I suggest you start at "google" or "google scholar" to satisfy your own curiousity.

I certainly will. Just hoped you could give me a pointer to which programs you think of. You obviusly know more of recent succesfull chessprograms than I do. I sort of drifted to other fields when Deep Blue won against Kasparov. That was - as you probably know - based on the brute force approach.

I would be especially interested in programs that use both genetic algorithms and neural netwroks.

/eske

[Mardonius]

Hello Lads:

Besides wargaming and fomenting revolutions, I like to run. Road races, cross country or wherever. One thing I have learned from running races is that it helps to know where the finish line is. They usually even have a little gateway for you to run through and sometimes even a medal to make you feel a sense of accomplishment.

But here with the AI we are running a race where we don't know where the finish is. So can we rewind and ask a very simple question:

If we have not got that human level intellgence from fifty plus years of AI development (see above discussion), what level of AI would a reasonable REASONABLE person expect for EiANW, given that it encompasses a dynamic more sophisitcated than a brute force paradigm? What is a reasonable level of AI?

Follow up question: How do we know we have got to this reasonable level, i.e. crossed the finish line?


best
Mardonius

[MarshallEllis]

ORIGINAL: Mardonius

Hello Lads:

Besides wargaming and fomenting revolutions, I like to run. Road races, cross country or wherever. One thing I have learned from running races is that it helps to know where the finish line is. They usually even have a little gateway for you to run through and sometimes even a medal to make you feel a sense of accomplishment.

But here with the AI we are running a race where we don't know where the finish is. So can we rewind and ask a very simple question:

If we have not got that human level intellgence from fifty plus years of AI development (see above discussion), what level of AI would a reasonable REASONABLE person expect for EiANW, given that it encompasses a dynamic more sophisitcated than a brute force paradigm? What is a reasonable level of AI?

Follow up question: How do we know we have got to this reasonable level, i.e. crossed the finish line?


best
Mardonius

Personally, I don't think I will ever be able to look at the AI and say "Finished!". At least from the non-tactical and more of the diplomatic side. I say this because human behavior / reactions are not constants. As situations change then so do human reactions. A good example of this might be How Turkey could be allies with GBR but then also allies with France (A lovely tactic that I have used before when GBR is hovering around 6pp away from instability). This tactic changes GBR to dang near hostile to Turkey but she is hesitant to issue a DOW since it would really cost BUT it makes France want to give you way more money than you will ever make in trade. Point is: In this scenario all three nations / personalities would change almost 180 degrees. The AI however might not understand how she needs to be hostile to an ally in this scenario.

I hope this makes sense. I'm trying to simply convey how current diplomatic settings do not always reflect how mad a human player may be with another player and yes you can approximate "mad points" if you will but this makes it more predictable and humans just are not predictable!

[Mardonius]

Exactly Marshall. There are so many dynamics in this game that go way beyond the brute force calculation that I too doubt that an AI will be even moderately capable when compared to a mid capability human.

And why do I keep asking about the AI and how much is enough? Glad you aked. Until now we have put the AI (and bug working out... which is another issue) as a higher priority than the OPTIONS or scenario editor. I think I know why... it probably has a bit to do with that survey you put out a few weeks back. Great customer solicitation but I am concerned that this pursuit of the deft AI will mean that you will never turn to the scenario editor and the OPTIONs similar to the ones I have been advocating so quietly.

At some point, you and Mssr Rutins (proud Vermonter) et alias should, perhaps, consider, puting one through the field goal posts and getting some more points on the board. A 1807 scenario should be a piece of cake ("slam dunk", in my hero, George Tenet's parlace)... other OPTIONs -- from both the SCRIPTURE and perhaps from apocryphal sources like those above -- could follow at regular intervals in your bug stomping campaign.

Vive La Revolution!
Mardonius

[Thresh]

Step back from the Kool Aid...

Marshall and Matrix have a plan.

That it is not your plan does not mean it is a bad plan, but Marshall has his priorities, and they are being worked on.  Much of what you pine for will be no good if the underlying issues aren't resolved first, right?

Todd

[Mardonius]

No Kool Aid here Todd. Sorry. You'll have to queue up elsewhere.

Please don't think that I am saying don't do any bug work or AI improvement. Just ask the question: how much AI is enough? I don't know the answer. I'd love to hear what is enough. What do you think Todd? I don't think asking the question is a Kool Aid thing.

And if the answer that comes out bespeaks that perhaps we should alter a plan after a certain metric is made, then that is a good thing.

Remember, it is a poor general who fights the war according to plan when the battlefield and the enemy indicate that a different course of action might be in mind. You're not from the Eastern Block, are you Todd?[:D]

The underlying bug issues, of course, do need a lot of work. To me that is less of a rabbit hole.

[fvianello]

mmmhh....no program has ever passed the turing test. I'd be surprised if EiA AI could make it