Trinity, AMFV, world representation
Phil Goetz
I'm fairly certain A Mind Forever Voyaging requires 128K.
I'll buy through US mail from individuals.
Lest Adam Engst, David Graves, & co. despair,
let me talk about Int Fiction...
I may spend a lot of time building what is actually intended as a tool
for AI research but would incidentally be a darn good IF platform.
My question is about world representation.
The usual tactic is to divide the world into objects (i.e. rooms, people,
things in the rooms), & change the world by discrete actions.
I would like to know as many problems that anyone can think of with
this scheme. For instance, what advantages could be gained by a
continuous 3D representation of the rooms instead of discrete different
rooms? What problems would this introduce (i.e. deciding when to list
objects on the display, how close you need to be to use different objects)?
What problems result from similarly dividing time up into discrete turns?
Also, if you can think of any reasons why an agent needs {a simulated world
to act in} rather than just {a collection of stored knowledge to answer
questions about} to demonstrate certain abilities (i.e. in plan
formation & execution, ambiguity resolution, various affects related
to the existence of a present physical context), that would also be
helpful.
We have art in order not to die of the truth.
- Friedrich Nietzsche
Jim Edwards-Hewitt
I may spend a lot of time building what is actually intended as a tool
for AI research but would incidentally be a darn good IF platform.
My question is about world representation.
The usual tactic is to divide the world into objects (i.e. rooms, people,
things in the rooms), & change the world by discrete actions.
I would like to know as many problems that anyone can think of with
this scheme. For instance, what advantages could be gained by a
continuous 3D representation of the rooms instead of discrete different
rooms? What problems would this introduce (i.e. deciding when to list
objects on the display, how close you need to be to use different objects)?
What problems result from similarly dividing time up into discrete turns?
The problem I've always had with "rooms" is their restrictiveness in
representing spaces large enough to move around in, such as outdoor areas
or rooms bigger than a small office. Outdoor representations tend to
consist of a set of "important" locations and a set of paths through the
"unimportant" areas. If I am at location A in the woods, and a path leads
to location B and then to C, there is generally no way for me to hack my
way through the woods straight from A to C. The only actions and
combinations of actions are those anticipated by the designer.
Indoors, where discrete spaces are realistic, the problem is more subtle,
but can be worse. If the game defines "East End of Hall" and "West End of
Hall", and I want to go to "Just down past the middle of the Hall, peeking
around the edge of that doorway into the corridor", I'm usually out of
luck. Moving far enough to the side to see around behind that altar,
without *going* behind it, is beyond most world structures.
I realize that these things *can* be done with a room representation, but
it's kind of like doing structured programming in old BASIC; you have to do
it all yourself, working against the representation, instead of having it
work for you.
The problems of a continuous representation directly correspond to the
advantages. With the advantage of moving around and seeing things from
different locations, you have the problem of generating a scene or
description from any location. With the advantage of moving through large
outdoor spaces, you have the problem that most of the area actually *is*
unimportant, and the player needs guidance while exploring it to keep from
being bored and frustrated. With the advantage of having a complete,
continuous world comes the problem of generating a reasonable description
of any area of it, however close the player chooses to examine it, without
using an infinite amount of storage space and authoring time. And so on.
Also, if you can think of any reasons why an agent needs {a simulated world
to act in} rather than just {a collection of stored knowledge to answer
questions about}
I would be interested in hearing more about this. I hadn't thought of that
as an alternative, and it may be a better solution to some conceptual
problems I've had with removing restrictions on interactions of physical
objects.
-- Jim
j...@visix.com
Gavin Inglis
[lots of stuff about a room system being a pain in the ass]
>> Also, if you can think of any reasons why an agent needs {a simulated world
>> to act in} rather than just {a collection of stored knowledge to answer
>> questions about}
> I would be interested in hearing more about this. I hadn't thought of that
> as an alternative, and it may be a better solution to some conceptual
> problems I've had with removing restrictions on interactions of physical
> objects.
There's a lot of scope here for escaping from conventional
representations. I believe that if we can step back, and look at age-old
ways of doing things critically, like this, that's when we start to
think laterally and new ideas crop up.
It's not new, but one of my favourite games from long-ago was called
System 15000. The plot was: you had been ripped off by somebody to the
tune of a million bucks (OK, it wasn't _all that_ realistic) and you had
to break into some computer and transfer it back to your account. Your
screen was set up like a modem: you typed the number to dial and a login
prompt came up. The thrill of hacking, without the expense and legality.
Ahhh, how my pulse raced when I eventually cracked the bank's machine...
You started on UK numbers and eventually got some overseas ones, with
different dialling tones.
At the time it inspired me to do something similar, and include bulletin
boards with messages that changed, system police and YOU ARE BEING
TRACED flashing alerts. It was great fun and actually very enjoyable to
play yourself, even though you knew all the passwords. I did it in BASIC
(I was young, right?) and was surprised at how easy the coding was. It
let me concentrate on making it feel realistic.
The other one which stood out was The Fourth Protocol, based on Fred
Forsyth's book about nuclear espionage. The first part was the best; it
had you at a desk for the whole game. Memos came in, the phone rang. You
could access files, ring people, and set surveillance on suspects.
Things would happen like: a watcher would phone you up, and say he had
seen his target bury something. Should he dig it up? Some selections
were menu-driven, and I know generally menus are fairly boring. But it
was the free-form bit with the whole desk at your fingertips that was
really fun.
The reason I waffle so much about these games is that, for me, they both
contain a world-model that's different, and _fun_ to interact with, even
if you know them back to front. The Fourth Protocol is fun to play, even
if you know the solution and just run through it; the interaction is the
joy. Can you say that about Colossal Cave?
Go on, respond. Make a redhead happy.
--
Gav | "Life's too short to stuff a mushroom"
<ai...@uk.ac.ed.castle> | - Libby Thompson