[Freeciv-Dev] Re: AI

[Raahul Kumar <raahul_da_man@xxxxxxxxx>]

> The current AI is largely in the state that it was
> in a
> year or so back when Syela (Trent can give you the
> real
> name) stopped working on it. Basically, no one else
> can 
> figure out how to change this without the whole
> system 
> coming unglued.
>

I noticed this. I sent some patches a long while back
to Thue trying to fix the movement code, because so
much of the code has majic numbers in it. 

> Currently the "neural net" (because that is the best
> 
> analogy for the current code - layers of complex
> decision 
> weights and code that mixes certain weights to make
> a 
> decision) is hardwired for a despotic sort of
> xenophobic 
> play mode. The AI has two states, "No Contact" and
> "War".
> 

A neural net? Where the hell is the learning through
negative feedback? I think it is much closer to an
exceptionally bad expert system, with a lot of hard to
find rules hidden in various places. It doesn't even
do 
much in the way of chosing which particular rule to
apply in a given situation.

> There are proposals to develop client-side AI
> through
> "Advisors" that humans can use to offload some of
> the 
> mundane strategy aspects. Raimar is laying some of
> the 
> infrastructure for this.

I am interested in this. Does this mean people could
compete to produce the best AI? Can you give me
further details. 

> I am in the process of hacking the server AI code
> left by
> Syela to try and first cleanup and document what it
> does,
> and then try to localize and parameterize some of
> the
> characteristics, so that the AI instances can
> develop a
> bit more "personality", i.e. to want things other
> than to
> beat up on any other civ it knows about.

Then in that case have you noticed the way the AI
chooses city squares is pretty bad. There is no check 
for how close other cities(friendly or not) are, so
the 
AI is always a default ICS player. 

> The step after this is to teach the AI the concept
> of 
> "not enemy", and then maybe move on to "trade with
> not
> enemy is good", where trade values may be influenced
> by 
> the personality goals.

Excellent. This sounds very useful to me. I am
specifically only going to put in sufficient code,
 to handle relations with known civs based on
states(peace,war no contact).
(If at war, the AI will demand tribute/and or techs
for peace, will probably decline or accept ceasefires
or peace treaties based on your personality traits).

> >I would also like to know exactly how Civ II ai
> behaved,
> >i.e which techs to offer to other players, whether
> to
> >demand tribute or not, and how exactly it kept
> track
> >of karma. 
> 
> This answer to this is, whatever you want to make of
> it.
> Currently you have pretty much a blank slate to
> write code 
> on.

I wanted to know if there are any existing ideas about
how to implement a better AI. With the diplomacy
system in place, we can finally have a competitive AI
player.
With the AI trading techs and making alliances the
human player will finally have some difficulty beating
the AI.
With someone else hopefully fixing the AI problems
with handling planes and ships a competitive late game
AI could be here.  
 


> 
> It is not bad ... it is non-existent.
> 
> Cheers,
> RossW
> =====
> 


__________________________________________________
Terrorist Attacks on U.S. - How can you help?
Donate cash, emergency relief information
http://dailynews.yahoo.com/fc/US/Emergency_Information/