Ok, word: The AI gets cost cheats at nearly every difficulty level:

[opaque]

I find it quite surprising how easy some people seem to think writing a good AI is.

I've almost finished making a very basic text-based strategy game with cities and defenders and attackers with 1/10000th of the complexity of Civ and I've only just managed to create an AI that can challenge the player. To make it hard, I still have to let it cheat...and it keeps getting stuck in infinite loops!

Seriously, non-programmers, it is NOT easy!

[player1]

I don't think AI knows where are your units (exemp if maybe having a spy of line of sight from religion).

The reason I say this is that I had a game where my closest city to AI has more then those units, while other, not to far away had just a few.

AI still attacked the closest city (with lots of troops I may add).
In Civ3, it would avoid it and centre on the other less defended cities.

[Todesengel]

Originally posted by opaque

Seriously, non-programmers, it is NOT easy!

That is true, but I think it is not a job a programmer can do alone. (If (s)he is not one of those great engineers or scientists...)

[kgsan]

Originally posted by Soren Johnson Firaxis


well, it may be BAD AI. Not the same thing as cheating AI.

No complaints, so far in my limited playing, the AI seems much improved over earlier Civ games. In addition, I know past practice has shown that Firaxis will certainly patch any major problems which come to light. However, there have been some comments on the forum that Firaxis plans AI improvement in a future patch as well. Is there any chance that the "man himself" could let us know if there is anything to this?

[Kuciwalker]

Originally posted by Todesengel
So, according to this I´m a top player after my third game...
However, you are wrong. Ever heard of simulated neural networks? Do you think the cia have hundreds of thousends people reading every email and listening to every phone call they intercept? There is no need, that the programmer knows, what the "ai" is doing. But as the capaceties of home pcs is limited, it might be a good idea to combine different approches. Furthermore, "ai"s or "es" have certain strengthes compared to the human mind. Why not exploit them? Look at what the stardock guys have done!

The NSA (not the CIA) has insane amounts of processing power. A desktop PC does not.

[Todesengel]

Originally posted by Kuciwalker


The NSA (not the CIA) has insane amounts of processing power. A desktop PC does not.

Well, whoever spies on me...

"the capaceties of home pcs is limited" is already noted.

[Urban Ranger]

Originally posted by Krill
he AI can't learn, and so can't improve naturally, like all other players.

This particular AI, maybe. However, it is possible for an AI to learn from mistakes and improve.

[Urban Ranger]

Originally posted by nbarclay
What is really needed is (1) a list of the advantages AIs get on all levels, (2) a table showing how the bonuses that vary by difficulty level differ where it is practical to present them in tabular form, and (3) a description of how things like hut outcomes and starting techs that aren't so easy to fit in a table compare. It would have been nice if Firaxis had done that so all players can have the information readily accessible whether they visit places like Apolyton or not. But since you haven't, I guess someone from the player community will need to do it. (I'm not volunteering at present, but I might do it at some point if no one else beats me to it.)

Excellent idea. Hm, I am not volunteering you, either

[padillah]

Originally posted by nbarclay
Developing a great AI is a whole lot easier said than done. ......
When you compare the games' complexity, Civ is enough more complex than chess to make Deep Blue (the computer that could beat the world's top chess players) look slower than an Apple II. Using anything resembling a brute-force approach, a Civ-type AI cannot possibly consider more than a tiny fraction of all the different combinations of things that could happen over the next several turns.

So with anything resembling current artificial intelligence technology, it's just plain unrealistic to expect AIs to be able to compete with the best human players with both sides playing by the same rules.

This is not so. There are several types of learning algorithms that could be applied to a Civ stratagy with differeing results. For example, multi-perceptron analysis with back propagation. Here's a link with several examples written in JAVA (to guarentee they are running on today's hardware )

http://diwww.epfl.ch/mantra/tutorial/english/

Some of these examples use suprvised learning, some, unsupervised.

As a matter of fact, Civ, with it's multiple inputs and outputs for every problem would be a wonderful use of neural networks.

NOT to say this would be easy. Heck I've been doing this for about a year and I still don't understand the math. But it's not nearly as unrealistic as you state. The problem comes with understanding, application and time.

The developers have to realise that neural networks are out there and are not science fiction anymore.

The developers have to find a quantatative input and output meathod for Civ moves and strategies. (What's the binary representation of "build more Axemen to take Paris unless the Romans break their treaty with us first"?)

And, last but not least, the developers need to be left alone long enough to do this kind of development. This is not something you can figure out overnight, it's actually very difficult and you can't implement this within a one year time frame.

And then, after all that, how do you cripple it so novice players can play it and enjoy themselves?

(The rest is not directed at you, nbarclay)

People act like this is the movie WarGames. It's not. You can't just tell a computer to act more human and have it listen.

If the game is given an AI that can think and react, learn and adapt... three people in the world will beat it and everyone else will hate it and never buy another game again.

You cry and moan but you don't really want a good AI, you want one you can beat, but just barely. Well, this is how you get that. Unless you can think of a way to make the AI smarter than it is but dumber than you but not make stupid mistakes but makes enough mistakes that you win but it shouldn't be too easy...

Asking for much?

Tom P.

[padillah]

Originally posted by Kuciwalker


Go however has a larger board, and I believe in general it has possible moves per turn than chess. In addition, it takes a lot more processing power to calculate the result of a move (at least AIUI - Go is similar to Othello, isn't it?). These combine to make search infeasible, unless we get massively more powerful computers.

Not really. There are neural networks using temporal difference learning and backpropogation that are making strides right now. On the hardware we have in our homes.

http://www.cgl.ucsf.edu/go/Programs/neurogo-html/NeuroGo.html

Hell, search is only feasible for chess with a lot of past games stored in memory.

Actually, no. Not even remotely. They are currently done using a system of Bitboards. nbarclay mentioned earlier that there were 64 squares on a chess board. Does that tickel something in the back of your brain? That's right! using a 64bit number you can represent every space on the board.

So, if I have one number that represents where all the white pieces are, and one number that represents all the rooks, and one number that represents all the moves each rook can make and I XOR them together I get all the moves a white rook can make. In 3 64bit numbers. Below is na article that explains it better.

http://www.fzibi.com/cchess/bitboards.htm

Tom P.

[-Ab-]

i am indeed not a coder so i may well be speaking rubbish but i feel that a good ai is not that hard to make,after all it is from a certain point of view just a flow chart, however turning that basic plan into code that runs likely takes some skill
for anybody thats played chess to any level you will know that you look at the board and you sit and calculate every single possible move before you choose the one you think is best
why cant this be applied to a civ game?
instead of makeing things cheaper and easier for the AI why cant it just be made to make use of what its got ?
and if that isnt enough then it must fight (something civ AI doesnt do alot of)
as far as military AI its a case of getting a group of units together and sending them off to fight,if it sends enough reasonably advanced units its going to win

the AI can't learn, and so can't improve naturally, like all other players

i think i posted this b4 somewhere, but galciv will ask you to upload your game every time you finish it so your strats can be stored and accessed by the game to improve the AI (i think that was the theory anyway i dont play galciv much)

Progam a better one then. Or post the rules for a better AI. Make sure to use exact programmable terms, for example, you can't say attack city x, or pillage y, you have to say attack city x with z units with a relative strength which is q times the defenders, etc.

i think i probably could write you a plan for an AI on paper and its something i have thought about for years now,however it would probably take me months and months and i am lazy and unmotivated and frankly cba sorry

When you compare the games' complexity, Civ is enough more complex than chess to make Deep Blue (the computer that could beat the world's top chess players) look slower than an Apple II. Using anything resembling a brute-force approach, a Civ-type AI cannot possibly consider more than a tiny fraction of all the different combinations of things that could happen over the next several turns

Civ4 is not more complex than chess..
I could go on about why but this post is already getting too long
its just not and to try and compare civ4s AI to deepblue is laughable

[padillah]

From a person that has spent man-years in design meetings talking about AI's and learning and backpropagation, etc. The trick does not come in making a "good" AI.

It comes in making, for CivIV anyway, NINE good AI's.

I suck at Civ. I do. I have yet to approach beating it even on Nobel. Velocyrix (not to pick on him but...) however can routinly beat Nobel with little effort. And there are others that make short work of anything except Diety.

I would not want to play against Vel's AI (and I wouldn't even want to turn on the PC if Sirian or one of the "Big Guns" had their AI on.) So how do you make an AI that is everything to everyone?

You don't. And that's pretty much the end of the conversation. Good, Bad, or Indifferent... this is the AI in Civ IV. The SDK will be released in the near future and then we can all poke around and see what we can see.

(I personally am going to try and add some decision weighting if not a true neural network of some degree.)

Tom P.

[-Ab-]

You cry and moan but you don't really want a good AI, you want one you can beat, but just barely. Well, this is how you get that. Unless you can think of a way to make the AI smarter than it is but dumber than you but not make stupid mistakes but makes enough mistakes that you win but it shouldn't be too easy...

you have a fair point there,however when i see the AI cranking out units at a speed that i know isnt possible without cheating (and i dont need to use a calculator to check) or the business with diplomacy and refusal to trade tech even when their your best friend and building times and barbs and blah blah blah... and strats you can use to advance thru the game seem to have been nerfed (particulary the way your not allowed to expand) it all just gets terrably boring

after all it is from a certain point of view just a flow chart

i have no doubt there is a way to beat civ4 with all its AI bonuses even on the hardest setting and people will say "oh your only moaning because you cant win" but there comes a time when you not playign a game anymore, instead your following the correct path in the "flow chart" of the AI
for me when that happens its time to put the game down and prolly uninstall it

[SpencerH]

Originally posted by WolfEyes

, the barbs get progressively better units and can even have cities.

I have at least 3 barb cities in my current 'noble' game that are now spitting out axemen against my quechas. I took one city with 6 promoted quechas (losing four in the process). The barbs have just taken one of my cities with one axeman. It may be a very short game.

[padillah]

Originally posted by -Ab-
i am indeed not a coder so i may well be speaking rubbish but i feel that a good ai is not that hard to make,after all it is from a certain point of view just a flow chart, however turning that basic plan into code that runs likely takes some skill

Sorry, but you're right... you are speaking rubbish.

It is not, under most circumstances, a flowchart. If it were designed this way than it would have to take into account every possible action any player would take and have an appropriate response predetermined.

Can YOU think of every possible action some of these guys playing CIV will take in EVERY possible situation? I can't.

So we are left with finding a method that doesn't depend on the opponent to accel. So, now design from there.

Civ4 is not more complex than chess..
I could go on about why but this post is already getting too long
its just not and to try and compare civ4s AI to deepblue is laughable

OK, let's just use pure number theory.
Chess: 16 pieces per side. 64 squares per board = 1024 possible moves at any one time. But some pieces have two types of move (i.e. pawns) so eight of those are duplicated but the knights have four moves and the rooks and queens have several. These can (and have been) quantified and can be represented as bitboards. (see above link)

Now, Civ: There are over 40 types of pieces (I tried not to UU's because they replace units, they are not extra) and over 10,000 squares = 400,000 possible moves with only one unit per type. Now, with no restrictions on squares to move to or pieces to capture this remains almost infinite throughout the length of the game.

This does not include trying to develop technologies, building units, spreading religion, cultivating culture, building a space ship, or any diplomacy what-so-ever.

Yes, CivIV is more complex than chess.

Tom P.