[Freeciv-Dev] Re: your mail

[Raimar Falke <hawk@xxxxxxxxxxxxxxxxxxxxxxx>]

On Thu, Apr 19, 2001 at 06:43:17AM -0700, SimFlyer wrote:
> From: SimFlyer <gamer@xxxxxxxxxxxxxxx>
> To: rf13@xxxxxxxxxxxxxxxxxxxxxx
> CC: freeciv-dev@xxxxxxxxxxx
> In-reply-to: <20010419130222.A29447@xxxxxxxxxxxxxxxxxxxxxxx> (message from
>       Raimar Falke on Thu, 19 Apr 2001 13:02:22 +0200)
> Subject: Re: Client AI development
> Reply-to: gamer@xxxxxxxxxxxxxxx
> References:  <20010419130222.A29447@xxxxxxxxxxxxxxxxxxxxxxx>
> 
>    Date: Thu, 19 Apr 2001 13:02:22 +0200
>    From: Raimar Falke <hawk@xxxxxxxxxxxxxxxxxxxxxxx>
>    Cc: SimFlyer <gamer@xxxxxxxxxxxxxxx>
>    Reply-To: rf13@xxxxxxxxxxxxxxxxxxxxxx
>    X-Location: Dresden, Saxonia, Germany, Europe, Earth
> 
> 
> 
>    Cell based client structure
>    ===========================
> 
>    ...
> 
>    What to you think? Overengineered? Do you think it is realizable?
> 
> Single Input and Output mean that separate chains have difficulty
> communicating.

No. I think it will be only one chain. The lower part consisting of
the NetworkCell and the HistoryCell. Certain high level functions form
the middle part. The upper part is the real "ai" which decided and it
most missing currently.

> Allowing multiple inputs and outputs provides cross-communication,
> which allows an AI to act globally.

This is AFAIK not necessary.

> You need to add a "Timer" and a method to handle "best move(s)".  The
> issue is that units have several possible moves in any turn, and a
> limited time to issue that move.  There is a best move for Offensive
> purposes, a best move for Defensive purposes, etc.  You cannot spend
> forever deciding what move should occur, but you should only issue
> move/action orders at the last possible moment, in order to maximize
> that time.  These suggestions are not simple to achieve, as every game
> AI programmer has learned.

I would like to address the problem of inter-turn-timing later. Some
upper layers (put a role on the unit) or the unit by itself (if it is
autonomous) has to decide which move it's best.

> Realizability:
> 
> These plans really come down to the dream of AI, something that has
> been elusive for some time now.  I wonder if the correct way to
> achieve something useful is to modify the goal.  Rather than create a
> complete stand-alone AI competing in games with player at a
> challenging level, create an AI helper for a player.

I has always both of this in mind. The player should be able to use
certain high level functions. Like the current "auto-settler" there
should be a "auto-city goal=attack units".

> I know that part of the challenge of the game is management and that
> AI aides are generally regarded as 'cheating', but not everyone
> plays for relaxation.  An AI helper is a stepping stone to a full AI
> player.  Freeciv contains several low level "AI" helpers currently,
> expanding their roles is a tough problem, but it is easier than the
> "better AI" problem.

Ack.

        Raimar

-- 
 email: rf13@xxxxxxxxxxxxxxxxx
 "This is Linux Country. On a quiet night, you can hear Windows reboot."