[Freeciv-Dev] Mind-States (Was: Re: AI Generally)

["Sam BC" <sambc@xxxxxxxxxxxxxxxxxxx>]

With what people have been bandying about regarding 'adaptive learning'
AI's, that could be part of a stored seed which stores the AI's personality
as well - so a particular installation of an AI can grow, develop, and
become a pacifist, a philanthropist, or a right b*stard...

Sam BC

> -----Original Message-----
> From: freeciv-dev-bounce@xxxxxxxxxxx
> [mailto:freeciv-dev-bounce@xxxxxxxxxxx]On Behalf Of Erik Sigra
> Sent: 02 June 2000 12:37
> To: Thue Janus Kristensen; freeciv-dev@xxxxxxxxxxx
> Subject: [Freeciv-Dev] Re: AI Generally
>
>
>
> ----- Original Message -----
> From: "Thue Janus Kristensen" <thue@xxxxxxx>
> To: <freeciv-dev@xxxxxxxxxxx>
> Sent: Friday, June 02, 2000 11:57 AM
> Subject: [Freeciv-Dev] Re: AI Generally
>
>
> > On Fri, 02 Jun 2000, Massimo Campostrini wrote:
> > > "Sam BC" <sambc@xxxxxxxxxxxxxxxxxxx> writes:
> > >
> > > > Correct me if I'm wrong, but the current code is pretty much opaque
> (or at
> > > > least only translucent) and hard to build on. Maybe after the next
> version
> > > > release it might be worth thinking about completely re-coding the AI
> in an
> > > > experimental version, even including both in source with a
> compile-time (or
> > > > run-time but the executables will be big) option.
> > >
> > > The current AI code looked completely opaque to me as well.
>
> > > If you plan re-coding the AI, consider writing it as a client-side
> > > AI.  Then we can have AI tournaments!  The main (dis)advantage is
> > > that it can't cheat.
> > >
> > > I don't know if anything is missing from the server to support
> > > client-side AIs.  As a minimum, a command to launch one or more
> > > client-side AIs on the server's host would be nice; maybe just
> > > modify "create" to accept a second optional argument:
> > >
> > > create <player-name> <external-AI-program>
> > >
> > > Regards,
> > > Massimo Campostrini
> >
> > I too agree that the AI code is opaque, though some of it work quite
> > well, that doesn't help if it is impossible to modify. Though I did
> > already "save" parts of ugly AI-related opaque code in
> > handle_unit_move_request() and the goto. (not accidentally the parts
> > needed in any case)
> >
> > I also think the right war to code a new ai is to make it client side.
> > That would accomplish:
> > 1) You very easily get clear of the ugly legacy AI code
> > 2) The AI will have it's own thead, making it possible to make it use
> > much more CPU without completely stalling the player (the server will
> > currently not take new input while the AI is running). For single player
> > games you could just disallow the human player to move units
> until the AI
> > had finished, but he could still look around and manage cities
> in the mean
> > time.
> >
> > -Thue
> >
>
> I'm wondering how the saving of games will be done with client side AIs.
>
> With server side AIs, the `AI-mind´ (long time plans, strategies and
> knowledge, the kind of things that human players save in their brains
> between their playing sessions)  can be stored in the standard savegame
> file.
>
> But with AI-less server, the standard savegame file only contains
> `the state
> of the game´ and nothing else. Do you plan an approach where every
> AI-clients saves a `Mind-File´?
>
> I can't think of any other solution. Not saving a Mind-File could maybe
> result in an AI-player after a reload wondering "What was my
> intention with
> this transport full of hoowizers? Where did I plan to unload
> them? Where to
> send them after that?" or "How was my civilisation organized into regions
> and what were the priorities of thoose regions?" or "Wich of the
> players was
> mean to me so I started building these nukes? They are ready now!"
>
>
>
>