Soren Johnson: The Chick Parabola

Well, it started out as being about the pure randomness of a CiV environment vs. chess.

Though, I wonder if there isn't some sort of "opening Library" or rather a finite, but expansive series of scripts that could be complied to "help" the AI in CiV.

No, humans can make computers that beats humans. Computers itself are dumb silicon.
And yes, opening libraries are crucial for a suc6 in middle/late game. As i can never win from a club-player which have a better "librairy" in his head then mine. Before you notice it, you are at the loosing end. Middle/late game "brilliance" can't help you then, only a crucial mistake from your opponent. I am stronger in the 5/10 minutes quick games, as i often manage to out manouvre the other players, even them with higher rankings.

YoHo said:

Though, I wonder if there isn't some sort of "opening Library" or rather a finite, but expansive series of scripts that could be complied to "help" the AI in CiV.

Click to expand...

If you took seriously the problem of creating a strong AI for a civilization-type game, you would certainly create some different scripts or rules for it to follow in the "opening". Humans do this too, after all. An opening script doesn't have to be a precise set of moves, it can be a set of rules that only applies during a certain phase of the game.

DaviddesJ said:

If you took seriously the problem of creating a strong AI for a civilization-type game, you would certainly create some different scripts or rules for it to follow in the "opening". Humans do this too, after all. An opening script doesn't have to be a precise set of moves, it can be a set of rules that only applies during a certain phase of the game.

Click to expand...

Quit hijacking the chess thread!

If they had the battles on a tactical map it should simplify it for the AI. The AI could have a set list of opening formations/postures depending on the units/terrain involved, and a list of behaviors.

I think the tactical map concept would solve some issues and is in general a good idea with 1upt. I think two of the major drawbacks would be that it's completely "unCiv", and that it takes the player "out of the game" during the battle. Actually, I guess that's one drawback since they're almost the same thing.

neptune2000 said:

I think the tactical map concept would solve some issues and is in general a good idea with 1upt.

Click to expand...

I don't understand what you mean. Are you suggesting that every time my spearman unit attacks your archer unit, we set up a tactical map and my spearman unit becomes a bunch of individual spearmen moving around on the tactical map and your archer unit becomes a bunch of individual archers moving around on the tactical map and we take turns moving and shooting at each other until the battle is resolved? That would take damn near forever.

DaviddesJ said:

I don't understand what you mean. Are you suggesting that every time my spearman unit attacks your archer unit, we set up a tactical map and my spearman unit becomes a bunch of individual spearmen moving around on the tactical map and your archer unit becomes a bunch of individual archers moving around on the tactical map and we take turns moving and shooting at each other until the battle is resolved? That would take damn near forever.

Click to expand...

Just the opposite way.

On the strategical map (the normal map at which you are viewing anyway, you would have the stack system, may it be limited or unlimited), but the battles would be resolved on a "zoom in" tactical map, automatically generated from the information about the attacker's hex, the defender's hex and the adjacent hexes. On this tactical map the 1upt rule would be valid.

That way, moving any units through your empire would be easy, without any blocking due to an friend's scout sitting somewhere, but the sieges on enemy cities would be played like they are played at the moment.

You would have strategic decisions (where to locate your troops in general) on the strategic map, and tactical decisions (how to fight the battle) on the tactical map.

MadRat said:

Look, I believe a modern chess program can beat a GM; they regularly do so.

Click to expand...

Nowadays, that has become a widely known fact, not a believe any more...

MadRat said:

My point is simple : an AI without a database library will not be able to beat a GM consistently.

Click to expand...

That can not be true nor be proven in anyway.

MadRat said:

Let me put it to you this way - a modern computer a with say 8 Gig memory can probably store database which would include every published game match in the last 20 years as well as an opening library.

Click to expand...

BS!
1. An openning library stores only states which are considered as within the scope of chess openning, it does not store all steps of the whole match.

2. Bigger memory only serves a bigger buffer to load openning library content. Again it is a matter for better performance. With bigger content is loaded, the program can therefore match faster against states stored in the library
i.e. Bigger memory helps to speed up seek action, but keep increasing memory size will soon become meaningless (unlike keep increasing the number of CPU cores)

MadRat said:

My hypothesis is that a pure algorithm based chess program would not win consistently versus a GM.

Click to expand...

Trust me, you don't need a hypothesis in this case.

MadRat said:

This is NOT the same as saying a computer cant beat a GM, which you seem to imply that I am saying.

Click to expand...

I am not impying that way.
My point is:
1. It is technically possible now for a computer program (a good one) to keep winning a human GM without openning library.

2. If program A always beats a human GM, it can beat him even faster (it takes less time for openning moves) with a openning library.

3. So, the different of having an openning library is to beat a human GM with less time or more time. That it is.

The most important fact I want to point out is, if a program has relatively weak algorithm (which is used after openning stage), it will always loss to a human GM regardless of how good its openning library and how well it can use the library.

No offence, you simply don't know the architecture of computer chess software and have a wrong idea of the role of openning library in that matter, that then cause you to keep on saying it is damn important.

In brief, (allow me to repeat):
You have over-emphasize the openning library of a chess program.

Seems to me that "symetrical" games are always beaten when the system is understood. Nothing new here...

lschnarch said:

On the strategical map (the normal map at which you are viewing anyway, you would have the stack system, may it be limited or unlimited), but the battles would be resolved on a "zoom in" tactical map, automatically generated from the information about the attacker's hex, the defender's hex and the adjacent hexes. On this tactical map the 1upt rule would be valid.

Click to expand...

Well, that's not 1upt, that is just the opposite. A stack of 30 axemen would completely crush a stack of 20 axemen on the tactical map, so you're going to force everyone to pile all of their troops into one huge army to have any chance in the tactical battles.

The whole idea of 1upt was to get away from the huge stacks of death. (I'm not saying it was a good idea, this whole thread is about why it's not a particularly good idea, but that is the idea.)

If by "limited" you mean Nupt, where N is greater than 1, then you force everyone to fight in stacks of exactly N units, and place a super-premium on unit quality because that's the only way to cram more strength into one hex. And you also completely hose defenders who have more than N units, yet their attacker can target a single stack and assault it with N units and then the survivors with N more units, while the neighboring defenders just sit there.

MadRat said:

I call BS on this. I haven't heard any evidence of this; if these so called "top programs" were beating High level players so regularly we would have heard about it. Without an opening book or DB library of games an AI will get creamed by most GMS and IMs.

BTW Deep Blue (some version) and Turbo Fritz are the top programs ATM I think.

Rat

Click to expand...

Your last sentence provides very clear evidence that you really don't know what you're talking about on this topic.

Deep Blue was quite literally dismantled after the May 1997 Kasparov match.

And Rybka has been the top program for years.

MadRat said:

Without an opening book and opening such as 1 h4 a5 is just as valid as 1 e4 c5 without the context of opening theory.

Click to expand...

No, of course it isn't. I don't know what you mean by "the context of opening theory", but basic principles such as mobility, space, control of the center, are central to chess AI evaluation functions and apply to all positions including these. Any chess program, even the most basic, has no problem identifying the latter moves as weaker than the former.

It's sort of bizarre that this thread has been taken over by arguments about chess and yet the two most vehement arguers both know so little about chess.

Perplexity said:

Your last sentence provides very clear evidence that you really don't know what you're talking about on this topic.

Deep Blue was quite literally dismantled after the May 1997 Kasparov match.

And Rybka has been the top program for years.

Click to expand...

I admit my samples were dated as I still use Fritz 10, my apologies. That doesnt disprove my point. Rybka 3 is essentially a huge database (http://www.chessbase.com/newsdetail.asp?newsid=4772). All of these programs are databases using hash algorithms on top of a good chess engine. If these huge databases are so minor why are they essential to every top program?? Even the HIARCS engine (fritz) which has stronger algo. still uses a hash sort. Indeed most improved program versions constantly refer to "faster, improved search" during game play.

I would honestly like it if someone would provide this forum with evidence that the innate chess playing algorithm in these programs is sufficiently strong enough to beat 2500 Elo (FIDE) rated players consistently WITHOUT ITS DATABASE OF OPENINGS AND GAMES. If it is such a minor thing, these books, why are they so essential to the AIs gameplay?

Rat

Dale said:

Brian Reynolds is responsible for Farmville and Frontierville.

Using him as an example of getting design right is not good for your argument.

Click to expand...

And yet Alpha Centauri was crap, right?

DaviddesJ said:

Any chess program, even the most basic, has no problem identifying the latter moves as weaker than the former..

Click to expand...

That is because most chess programs use opening books and games libraries logic to help evaluate the position. I dont think the AI is evaluating those opening principles as much as it is weeding out bad lines using a database.

Rat

MadRat said:

That is because most chess programs use opening books and games libraries logic to help evaluate the position.

Click to expand...

No, it just isn't. I am amazed that you know so little about how chess programs work and yet you are posting here about them. Like I said, chess programs use evaluation functions that incorporate well-known heuristics like mobility, space, and the value of controlling the center of the board, among many other factors. This has nothing to do with opening books, first principles are sufficient to immediately tell any chess program that the latter moves are much better than the former moves. If you deleted the opening library completely, chess programs aren't going to start opening with 1. h4. Even weak ones. That is just wrong.

Chess programs have opening libraries because they are stronger with opening books than without them. That doesn't mean that they would necessarily be weak in openings without the book. That's just a logical fallacy. The best computer chess programs are now believed to be much, much stronger than the best human players---something like 200 Elo points. If you did away with their book entirely, they would lose some strength, but that doesn't mean they wouldn't still be better.

What does this have to do with the Chick Parabola, again? Being wrong is bad enough, and being irrelevant is bad enough, but why both at once?

if I remember correctly the chess state diagram is something like 10^38 states and until you get HW that can calculate through it in real time you will always depend on set number of turns predicted and evaluating function.

Opening library or library of played chess parties can severely reduce the number of states the computer has to walk through (the algorthims typically try to throw away trees that they already know outcome of), so saying that databases of chess parties is for computer is denying obvious fact, without it the computer can't go deep enough and will have not sufficient memory probably for such task.

DaviddesJ said:

Well, that's not 1upt, that is just the opposite. A stack of 30 axemen would completely crush a stack of 20 axemen on the tactical map, so you're going to force everyone to pile all of their troops into one huge army to have any chance in the tactical battles.

Click to expand...

Not really. It depends on map type, troop movement, upgrades, etc. Becuse this map would be played at 1Upt.
BTW, a huge army should beat a smaller one most of the time, except if something changes the balance, like a good general, weather, superior technology.

DaviddesJ said:

The whole idea of 1upt was to get away from the huge stacks of death. (I'm not saying it was a good idea, this whole thread is about why it's not a particularly good idea, but that is the idea.)

Click to expand...

The whole Idea was to avoid uninteresting combat. It failed because of the bad AI. If there's other way to implement interesting combat, I welcome it.

DaviddesJ said:

I don't understand what you mean. Are you suggesting that every time my spearman unit attacks your archer unit, we set up a tactical map and my spearman unit becomes a bunch of individual spearmen moving around on the tactical map and your archer unit becomes a bunch of individual archers moving around on the tactical map and we take turns moving and shooting at each other until the battle is resolved? That would take damn near forever.

Click to expand...

DaviddesJ said:

Well, that's not 1upt, that is just the opposite. A stack of 30 axemen would completely crush a stack of 20 axemen on the tactical map, so you're going to force everyone to pile all of their troops into one huge army to have any chance in the tactical battles.

The whole idea of 1upt was to get away from the huge stacks of death. (I'm not saying it was a good idea, this whole thread is about why it's not a particularly good idea, but that is the idea.)

If by "limited" you mean Nupt, where N is greater than 1, then you force everyone to fight in stacks of exactly N units, and place a super-premium on unit quality because that's the only way to cram more strength into one hex. And you also completely hose defenders who have more than N units, yet their attacker can target a single stack and assault it with N units and then the survivors with N more units, while the neighboring defenders just sit there.

Click to expand...

Yeah, I mispoke in my post (I'll blame it on the flu). I was thinking of a limited stack, a small number just to allow for combined arms. What I would suggest is that the flanking bonus for units/stacks be carried into the tactical map somehow, so your setup on the strategic map would still affect the battle. An individual battle would represent a slice of the larger war. Also, there would have to the ability to at least attempt to withdraw (although there currently is no way in any Civ for a defender to remove itself from a tile short of destruction).

I don't know how exactly it would be done or even if it's feasible wrt game flow. You're also correct that it would take longer. This is one of the reasons why I said it would be "unCiv". A "simulate" option would be necessary; perhaps running the tactical map would make sense if the odds were not overwhelming. Of course that would break down in multiplayer games if either player could choose to enter the tactical map (what? You want to manually resolve a warrior vs. 2 knights and a longbowman?)

DaviddesJ said:

What does this have to do with the Chick Parabola, again? Being wrong is bad enough, and being irrelevant is bad enough, but why both at once?

Click to expand...

Ok it is obvious you cant conduct a cordial discussion without juvenile troglodyte insults. Time to unsub this thread and add to my ignore list.

Rat

DaviddesJ said:

If by "limited" you mean Nupt, where N is greater than 1, then you force everyone to fight in stacks of exactly N units, and place a super-premium on unit quality because that's the only way to cram more strength into one hex. And you also completely hose defenders who have more than N units, yet their attacker can target a single stack and assault it with N units and then the survivors with N more units, while the neighboring defenders just sit there.

Click to expand...

I'm not deeply convinced by your argument here: assuming a Civ 4-style "best unit defends", then 1 UPT places more of a premium on unit quality, and hoses a defender more, than any larger N.

I largely support the goals of 1UPT, but I can also see some good arguments for, say, 2UPT.