Dumb AI

[LDiCesare]

You dont have any local

There are local problems in chess, for instance check and mate are local. You don't care if the opponent could mate you in 4 if they moved if you can force mate them in 2 on the other side of the board. Taking a queen with a pawn is a local situation. If it doesn't lead to a bad situation within 10 turns, it is very probably a good move. Actually what I ment is more the difference between tactical and strategic approaches.

[Urban Ranger]

Originally posted by MightyTiny
My point was NOT that computers could be expected to be "naturally" good at chess, but rather that it is orders of magnitude easier to program a computer to play a "deep game" like chess well (or at least adequately), than it is to program a computer to play a "wide game" like Civ4 well.

Yes, no, maybe.

Take the silicon estate of, say, an AMD64, and use that to make a bunch of 8 bit processors. Then connect them in parallel and run separate threads on them. That will be a configuration that is good to tackle "wide" problems.

[Urban Ranger]

Originally posted by johnmcd
Chess is not something computers are good at. I can't believe people are saying it's about computation. It's the fact that it's emphatically not about computation that means it's taken so long to build computators that can play it to a high standard.

Chess is one of the games that is very suited for computers to handle because it is very "well defined" and relatively small.

I use "well defined" here to describe the following:

1. The game has clearly defined and rigid rules.
2. There is no hidden information in the game.

As a result, it is rather easy to devise algorithms to play chess. The main problem for a chess program can play at the International Grandmaster level is speed. Even though the algorithm is not complex the look-ahead tree explodes in size quickly. The program bogs down quickly in looking for the best move. Sure, you can apply certain pruning methods, but that will knock out certain combos that give up stuff (such as material) initially for a bigger gain later.

This can be handled in a few ways. First of all, you can put in faster hardware. Secondly, consult expert players (International Grandmasters) when creating the program. Third, create a massive database to "short circuit" the lookahead process.

This third method is very similar to how we humans do things. We are very good at pattern recognition, which is a fuzzy search and match of our internal "database." OTOH, we are rubbish at rote computations. This is exactly the opposite of digital computers. That's things come easy to us (e.g. recognising faces) are so difficult for computers.

Originally posted by johnmcd
Playing chess on computation is like recognising your friends by measuring the distance between their eyes and the circumference of their skull. It's the least effecient way to do it when you can 'just' recognise them. That 'just' is something no computer can do.

Only part of the game is recognition, and this is very basic recognition that a computer can do quickly. So if you have a massive database you can make a chess program much better.

[Urban Ranger]

Originally posted by padillah
Big Blue beat Garry Kasparov 3.5 to 2.5

To be fair, Deep Blue (I think it's called Deep Blue - Big Blue is IBM itself ) had a whole team of people tending it, incuding some expert chess players IIRC.

[padillah]

Originally posted by Urban Ranger


To be fair, Deep Blue (I think it's called Deep Blue - Big Blue is IBM itself ) had a whole team of people tending it, incuding some expert chess players IIRC.

First off: Thanks. You're right, it's Deep Blue, but I worked for IBM for a year so it got more than muddled.

Second: They helped but only inasmuch as they helped write the look-ahead algorithms and code the piecs weights and position weights and such. This is the same preparation Mr. Kasparov had available to him and, indeed, used. The difference being exactly your point: Deep Blue could remember and calculate all the information given to him, Mr. Kasparov had to settle for 30+years of pattern recognition and study.

Tom P.

[padillah]

Originally posted by Urban Ranger


Yes, no, maybe.

Take the silicon estate of, say, an AMD64, and use that to make a bunch of 8 bit processors. Then connect them in parallel and run separate threads on them. That will be a configuration that is good to tackle "wide" problems.

This would work moderatly well... then, again, this is exactly what we are trying to say: The standard home computer is ill equipt to process the amount of data to full make an informed decision in each move of Civ.

The problem is two fold: the processing power itself, be it deep or wide or whatever, and the algorithms that weigh and decide which move is better, worse, or indifferent.

We've mentioned Deep Blue, that had a cadre of programmers, mathematicians, and no less than three other Int. Grand Masters helping decide the weighting and processing and movement database and so forth that went into each move.

Civ does not have those kinds of resources.

Tom P.

[padillah]

Originally posted by Loader2k
That's true, a strategy game is only so difficult as your
opponent. But there are strategy games that are easier to
master than others. Chess and Go are hard, some game
for children with little strategy in it is easy to master, cIV in
my opinion is definitely easier than Chess or Go.
Now the game of cIV apart from the AI is still demanding
enough to be very entertaining - as i said the game
balance is very good. But without decent oppenents...
well.

Actually, my point was the opposite of this. Chess is a very easy GAME. The defenition of the game is very succinct, and very specific. There are 6 pieces and they each move in a very specific and unique manner. The game surface is open knowledge and never changes or gives advantage.

The same cannot be said for Civ. Heck even just piece movement can change based on various wonders you have or feats you've accomplished. So even saying "Triremes move 2 squares" is not good enough.

This level of abstraction adds difficulty, not takes it away.

Which is UR's point: to us, humans, we think this way in the first place. We don't have much to change.

I disagree. In Chess it's not easy. You dont have any local problems in Chess. Moving a piece anywhere on the board affects the whole board. The existence of local effects should in fact enable a better AI since local problems are independent of each other and can be solved
independently, reducing the problem size *dramatically*.

No. In chess every problem poses both a local AND board wide problem. Taking a piece, especially a pawn or knight, is a rather local problem. They can currently move one or eight limited places. We compound this with the knowledge that we also will not have that piece in the long run but the same can be said for Civ.

In Civ it may even be magnified, since piece construction costs time, or tempo in cheess terms, and a loss of tempo can almost never be reclaimed. ( Hence the ongoing debate on whether to build a setler first or a warrior. You can only build one and that tempo will probably never be recovered.)

Tom P.

[Strudo]

I think it's also worth considering that in chess both players have the same goal to achive. Only victory condition is the take the opposing king. Both players know thats what the other player is attempting. Civ on the other hand has many victory conditions and you can't possibly know which one your opponent is attempting to execute.

On the note of standard player vs hardcore player, I consider myself a standard player as I play the game for what it is. So the A.I. cheats to make the game more difficult... humpf... so what? Then the A.I. does stupid things... so? I know people who do stupid things. And yes, they do them repeatedly. I start a game of civ, I play the game and I finish the game. Then maye I'll play another game of civ, or maybe I'll play something else.The hardcore players are those on this site who can go into combat knowing the odds of winning to 3 decimal places. They look at what happens in the game and analyze it until they understand it. And by understanding the game, they can offer suggestions to improve it. Standard players cannot offer this degree of information, and frankly probably don't care. If the game improves 10 fold in some area, I'll just start a game, play the game and finish the game as I always have. Maybe I'll have more fun with the 10 fold improvements, but it's still just a game.

Thats what I reckon anyway.

After reading all these posts it's getting too late to start a civ now. Might just play some X3 and let the A.I. crash my ships into stations

[Akaoz]

meh... I was a week to slow... ignore me.

[potoroo]

Originally posted by Akaoz

Originally posted by padillah
I also know that a chess board has 64 x 64 = 4,096 squares and 16 + 16 = 32 tokens.

That's one hell of a Chess Board...

It's the new Huge chess board. Games now last up to a week.

[padillah]

Originally posted by potoroo

That's one hell of a Chess Board...

It's the new Huge chess board. Games now last up to a week.

I'm working on a Marathon speed... give me some time, I'll get it.

Tom P.