[Freeciv-Dev] Re: [RFC][Patch] AI can fly v2 (fwd)

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

At 10:04 AM 02/04/12 -0500, Tony Stuckey wrote:
>On Thu, Apr 11, 2002 at 07:26:49PM -0400, Ross W. Wetmore wrote:
>> The real problem here is that humans have persistent memory about what
>> they *have* seen (in now fogged out areas), or pretty good guessing skills
>> about what they are likely to find based on lots of subtle clues.
>
>       Starting map parameters are known to all players.  That should
>prime a lot of expectations right there -- Archipelago versus Single Large
>Continent versus Desert World, versus Earth Map, etc.
>       The AI should understand geopolitics at least as far as "There are
>an average of 150 map tiles available per player.  35% of those are land.
>I only know about 38 tiles.  Explore."

Except that exploration only sets known irrevocably while seen soon
turns off.

Unseen tiles are not targettable by the AI if you fix goto_is_sane()
to actually work, because its condition is tile_is_known_and_seen().

Thus humans can target anywhere their mental memory or intuitive sense
suggests is reasonable for moves.

The AI can only target squares (to explore or found cities) to the limits 
of its know AND SEEN space. This means humans cheat according to the rules 
forced on the AI, and it is a very powerful cheat.

The AI is also not particularly intelligent in exploring surrounding 
territory or continually refreshing its explored space so that it can
calculate effective moves, such as where best to site a new city or
send an explorer.

The current AI only works because goto_is_sane() is broken and it CAN
target terrain which is unknown or fogged out like humans.

Note, targetting squares and revealing information that is in them for 
other purposes are two different things. Allowing an AI (or mentally
handicapped human) a fuzzy look at tiles it would not normally know
about is not a bad strategy for providing an offset to high cognitive
abilities in some humans that provide them with an unfair advantage. 

Or conversely, the fact that you use the real map as a starting base 
and apply error fuzz vs maintain some complex ongoing guesstimate to 
build the mental map from zero is only an implementation detail that
has little effect on the final result except maybe usable computation
speed.

But at the least the AI should be allowed to query such tiles and be 
given an "average tile" value if it fails the known-and-seen criteria. 

Afterall, tourists, news services and/or bards have always providing 
these peeks into foreign locations in real life to offset the fact that
most people don't ever visit or receive personal ongoing mail reports
from every geographical district in the world on an ongoing basis.

>-- 
>Anthony J. Stuckey                              stuckey@xxxxxxxxxxxxxxxxx
>
>'Finally, the Navy stated that [...] "However, use of the area as a live
>fire range has the beneficial effect of reducing the negative impacts of
>human intrusion."' - Center For Biological Diversity v Pirie and Rumsfeld