RE: does RDF require understanding all 82 URI schemes?

: > Trust isn't a requirement.

: Ross Judson:
: I believe it is.  Unless the protocols by which trusted relationships can be
: created and destroyed are well-known, they can't be engaged in on a
: widespread basis.  This is the equivalent of web browsers that don't support
: http, where each browser has its own protocol.

I'd suggest that any Trust protocol is an application of the semantic web, not a
requirement for its base  architecture. It belongs on a different layer, so to
speak. What you're saying sounds to me liking insisting that a grammar only
produce semantically meaningful sentences as opposed to merely syntactically
correct ones, but I suspect we're talking at cross purposes.


: Anything can be modeled by a set of peers; the question is whether the
: protocol is of sufficiently general use that it belongs in a well-defined
: standard.  Trust is the wheel of the internet; it should not be re-invented.
: Let's get it right (or at least good).

The network should allow a Trust or set of Trust protocols to arise naturally.
Essentially the network and its hosted communities should learn (and if needs
be, unlearn) any Trust protocol.

Why? Modeling Trust and getting agents to make best effort decisions based on
inferred Trust is very hard with systems of computationally bound asocial agents
holding only partial information (essentially it involves mind reading another
agent's intent). It's not at all clear how to architect such a system. One could
certainly have a crack at building a legislated subsystem or space and see how
may people want play. Might even be a fun semantic web app, like building City
States or Trading Ports on the web.

But, no it probably be can't be got right ahead of time and that's my point.
Sure it hits a certain kind of interop for now, but if Trust is badly specified,
then the network itself becomes partially crocked.

I heard a speaker from a large online bank say recently that Trust on the web is
best bootstrapped by branding, not legislation: that's an interesting
non-technical approach. There's a good amount of work being done on contractual
style discourse/negotiation systems for resource bound multiagent systems, where
game theoretic models aren't always applicable. Sourceforge.org, Ebay.com and
Advogato.org use various forms of trust metric rating systems (Advogato's system
is well worth a look). There seems to be lots of best-effort ways to skin this
puppy, why commit ahead of time?

As a strawman suggestion, my immediate preference for modeling peer to peer
machine Trust would be to build it out of language primitives, possibly using
discourse centered on promissory style speech acts and possibly as a part of the
DAML initiative.

 Bill de hOra

Received on Monday, 12 February 2001 19:36:15 UTC