[MUD-Dev] Player run reputation system

[Ola Fosheim Grøstad <olag@ifi.uio.no>]

"Koster, Raph" wrote:
>> From: Sean Kelly
>> From: "Koster, Raph" <rkoster at verant.com>
 
>>> Then you don't need a reputation system at all. :) You can size
>>> them up in the traditional way, by getting to know them and
>>> talking to people who have interacted with them.
 
>>> What else would you use it for?
 
>> NPC interaction.  I really like the idea of NPC behavior towards
>> a player being at least somewhat dependent on that player's local
>> reputation.  Thic could vary from shopowners varying prices,

> For that, you're probably better off having just the server
> tracking actions rather than having the above NPC behaviors be
> based on a player-entered reputation. Or so it seems to me...

Actions by themselves do not reflect intentions.  For a role acting
system you want to measure intentions, but to that you need to infer
context.  Which might be an "AI complete" problem (i.e. you need to
solve all other difficult AI problems related to context and more
subjective issues that often crop up in relation to human behaviour
in order to address it)...

Sean seems to be onto something here. At least for higher level game
play (serious gamers) where you've got a sufficiently rich and large
set of data with the desirable statistical properties that could
make a system somewhat reliable... Lower level players could be
classified according to their affiliation with higher level
groups. I.e. in order to improve your standing with a city you would
improve your standing with a higher level group of players or a
group that is affiliated with a higher level group.

The best solution is probably to mix objective and subjective data,
system provided and user provided meta information input... I.e.:
Provide systems for defining legal actions: "my friends are allowed
to pick apples from my garden", formalize affiliation and
positioning between groups, basically let users define their borders
and the exceptions; let them express their perception of other
people... Let the system do cluster analysis and map the most
significant features to multiple dimensions which can be used by
NPCs. (maybe some kind of principal axis type of analysis,
regression lines... (I've forgotten all about that stuff)) Group
NPCs and let them form "internal preferences/alliances" which are
changed at certain intervals in order to make sure that the system
is dynamic. (You could force this through a rich set of well thought
out heuristics "this city is no getting fed up with <the dull
elves>, we now switch the preference to <a group that is separate
from elves in the cluster analysis> that possess the following
features... etc")

If you suggest actions as the only data source, then you sort of
make the claim that you will be able to classify human personality
based on actions alone. I don't even believe much in the more
quantitative personality classification schemes which you can find
in psychology, which makes it difficult for me to believe that game
developers are capable of making a hard formal system that solves
the issues involved at a reasonable (useful) level of
significance. (Useful to role acting, that is. Any system can turn
out to be useful as a target for strategic game play.)

Maybe one could base such as system on Bayesian inference. (i.e.
combining feelings and empirical evidence). Unfortunately, I don't
know enough about it.

--
Ola  -  http://www.notam.uio.no/~olagr/
_______________________________________________
MUD-Dev mailing list
MUD-Dev at kanga.nu
https://www.kanga.nu/lists/listinfo/mud-dev