Anders Sandberg wrote:
>
> On Wed, May 02, 2001 at 04:35:21PM -0700, Adrian Tymes wrote:
> >
> > I'll take a whack at it, if only because the first thought that crossed
> > my mind upon reading this was, "So how does one hack it?"
>
> Always a good design method.
>
> I wonder how resistent it is to societies for mutual admiration? If
> Spike and I always rate each other very highly, that ought to be able to
> increase our ERA. It would become even more effective if we could get
> more people into the fanclub. This doesn't need phantom people.
>
> I guess this is part of the problem Natasha pointed out: how much we
> like each other has little to say how much good stuff we actually
> contribute, and is far too one-dimensional. Clique formation can
> certainly happen when we deal with actual content (for example by the
> formation of the "Singularity now!" gang vs "AI is hard and slow" gang
> when discussing bootstrapping SI) but is more likely with the default
> social behavior of humans.
>
> BTW, a mailing list ought to be a wonderful source of information for
> automatically making a sociogram (a graph of who responds to whom,
> showing the topology of discussion). It would be fun to see how this
> list looks as a sociogram.
>
I'm sure Javien has thought of all this, and it'll be interesting to
see what they release... anyway, it sounds like the raw reputation
number should be augmented by a second number - "breadth" - to try
to give an idea how widely that "opinion" is held. If a few people
get together and rate each other to the max, but everyone else doesn't
particpate then that small group would have a low breadth. Or high
"uncertainty". I'm sure there are better terms to use, but you get the
idea. One number is probably not enough to provide a really usable
reputation system. We need stats stats stats.
-- Brian Atkins Director, Singularity Institute for Artificial Intelligence http://www.singinst.org/
This archive was generated by hypermail 2b30 : Mon May 28 2001 - 10:00:02 MDT