[freeciv-ai] Re: Borg AI.

["Ross W. Wetmore" <rwetmore@xxxxxxxxxxxx>]

At 11:39 PM 02/07/02 +0200, Per I Mathisen wrote:
>On Tue, Jul 02, 2002 at 09:45:33PM +0200, Per I Mathisen wrote:
>> > Influence maps are very nice, as long as you don't have fog of war.
>
>On Tue, 2 Jul 2002, Raimar Falke wrote:
>> You have to cope with incomplete information (fog of war) in every
>> fair AI which doesn't cheat. You have to estimate the strength of the
>> enemy.
>
>In most games, most of the time, you don't see more than a tiny
>fraction of the enemy's forces, if at all. An influence map created using
>this tiny slice of the total picture will give you a very misleading
>influence map.
>
>So I don't see how an influence map that is rebuilt each turn can be
>viable.

Correct.

>If you can somehow make an influence map that isn't rebuilt but has some
>kind of decay function to represent decreasing reliability of past
>observations, then it might work since over time you see a lot more of the
>enemy's forces, if you do active exploring/patroling, but that'll be quite
>complex.

Both static and dynamic data needs to be considered. dynamic data like
unit observations should decay. This is a small part of the picture 
though the PF craze will probably make it a large part of the CPU cost.

Static data includes effects like 
1)  borders - constant or cumulative weight based on some definition of
    controlled territory (e.g. influence of local cities).
2)  terrain effects - certain features need to always be considered as
    a threat even if not currently enemy occupied, or a bonus if friendly
    occupied or under "sufficient" friendly control.

For static effects, you need to ignore fog-of-war. If you have seen 
terrain or an enemy city, you should always assume it is there until
updated, not have knowledge of it vanish with a real short term memory
effect. Similarly, you need to make its influence non-decaying.

If influence is decidedly non-local, you may need to bump the values
for known enemy static effects to account for the (known) total size
of the enemy Civ as reported by diplomats or general rumours like the
periodic reports. Assuming an enemy is tiny because you never penetrate
the Civ far enough to see it all would be a bad AI strategy, as you say.

>Yours
>Per

Cheers,
RossW
=====