Nuclear Weapons should be More Expensive

Hail said:

second: you should have full knowledge of what the program should be doing, otherwise it will not work as desired. if can write up an ai generic algorithm, that once in while will be deduced [by the ai] to paradroping units behind enemy lines, without explicitly coding it in, then my respect.

Click to expand...

I can see how I would code a program that could "deduce" that, fwiw.

(Short version; multiple copies of AI set up to play against each other at all sorts of points and positions in the game. Randomise their initial behaviour, give them the capacity to assess the returned benefit or drawback of any given action, and adjust the likelihood of doing it again accordingly, iterate, let Darwin happen, keep a human eye on it for falling into pathological modes.)

P.S. that i cannot multiply 5-digit integers by fraction of a second: does that make me dumber than a Core2Duo cpu?

Click to expand...

If and only if you define "dumb" as "less good at multiplying 5-digit integers".

rysmiel said:

(Short version; multiple copies of AI set up to play against each other at all sorts of points and positions in the game. Randomise their initial behaviour, give them the capacity to assess the returned benefit or drawback of any given action, and adjust the likelihood of doing it again accordingly, iterate, let Darwin happen, keep a human eye on it for falling into pathological modes.)

Click to expand...

life is short.

why not make nukes more devastating and costly? why not strive for pure realism once in a while!?

Hail said:

life is short.

Click to expand...

You'd be surprised how fast it's possible to implement something like this on a problem of this kind of scale.

Particularly if the algorithm testing and so forth was decoupled from the interface so you could do the strategy-evolving without actually having to generate all the graphics all the way along.

why not make nukes more devastating and costly?

Click to expand...

I'm not actually opposed to that...

why not strive for pure realism once in a while!?

Click to expand...

..just to doing it for this reason. (Because "pure realism" is an argument that does not front-and-centre make whether it improves the game the #1 consideration.)

rysmiel said:

You'd be surprised how fast it's possible to implement something like this on a problem of this kind of scale.

Particularly if the algorithm testing and so forth was decoupled from the interface so you could do the strategy-evolving without actually having to generate all the graphics all the way along.

Click to expand...

i was not referring to the time to code something like that, but to the time such game will play out. graphics are only needed if you want to watch the action.

Hail said:

i was not referring to the time to code something like that, but to the time such game will play out.

Click to expand...

For testing the AIs to the point at which they have decent strategies ?

I would reckon you are talking on the order of weeks for a first pass, and it's something that can be sitting in the background churning away to include better AIs whenever you get around to an expansion pack.

What do you mean by 'programming' in the higher difficulty levels? Can't you just play the game how it is?

Also: Civ3 (I'm guessing civ4 too) AIs are not good planners at all. They just go for one or two of your cities, and are pretty predictable.

And: The AIs in diplomacy are also very predictable. Your foreign advisor (in Civ3) will tell you if the deal that you are about to make is never going to happen, insulting, probably not going to happen, close, or acceptable. I find this very helpful, as far as I've seen, I don't feel like looking now, this is not in civ4.