W3C home > Mailing lists > Public > www-archive@w3.org > March 2004

Re: more text / Trust bipolar decision or continuum

From: Chris Bizer <chris@bizer.de>
Date: Thu, 18 Mar 2004 11:09:58 +0100
Message-ID: <002b01c40cd1$2e294010$1f12fea9@named4gc1asnuj>
To: "Pat Hayes" <phayes@ihmc.us>
Cc: "Patrick Stickler" <patrick.stickler@nokia.com>, "Jeremy Carroll" <jjc@hpl.hp.com>, <www-archive@w3.org>

> >
> >See quote from one of my documents below:
> >"There is a wide range of different trust definitions known from
> >sociology, economic science and computer science. There is no general
> >agreement on the term "Trust". A comparison of the different views on the
> >topic and the different trust definitions can be found in [Marsh94] and
> >[Gambetta00]. Following Joseph M. Reagle [Reagle02], we define trust as:
> >Trust (worthiness) is the degree to which an agent (human or machine)
> >considers information to be true for a given time and context.
> Is trust then always and only concerned with truth of information?
> That seems rather restrictive as a general theory.

It wasn't meant as a general theory but as a domain specific definition
aiming at domains like the Semantic Web, where trust in information is kind
of relevant.  The definitions used today are usually domain specific,
because general definitions tent to get so abstract, that they aren't
operational anymore. If you want more action orientated definitions you can
look at the ones used in game theory. If you want definitions which don't
oversimplify the complexity of trust you can look at the ones used in
psychology and sociology (I can guarantee that you will not like them

>For example, there
> are people I know who I trust to be rational, or to be objective, or
> to be honest: but none of this means that I necessarily believe what
> they say to be true.
> >Our definition refers to the following aspects of trust:
> >-There is uncertainty in trust situations, which cannot be eliminated. It
> >only possible to minimize uncertainty.
> One can make similar claims about any situation involving
> propositional attitudes of any kind. Such philosophical havering is
> easy, but not convincing unless it is backed up by some deeper
> analysis. (IMO). Is there anything special about trust that makes it
> inherently continuous? If anything, the potential link to action and
> agency usually makes things more polar, not less polar, since its
> impossible to act in a continuum: you either do the action, or not.

Agents and their decision making is one thing. Ranking information found on
the Semantic Web is the other.

An agent might find 500 contradiction assertions about a specific topic. He
may prefer to rank them using a trust policy and display the ranked results
to the user, leaving the final decision to him.

> >-Trust is subjective. Different users have different views of the world
> >different subjective trust requirements in the same situation.
> >-Trust depends on the context and changes over time.
> >-Trust in a continuum and no bivariate property.
> None of this seems convincing to me. It reads like pop sociology.

You should be convinced at least by the first two, because we are proposing
them in our paper. Our agents use subjective trust policies. One agent might
trust a graph, another agent might distrust it in the same situation because
he uses a more paranoid trust policy. The policy used depends on the
specific task (context). If you buy something worth a million you are more
paranoid than if you are searching for my papers :-) When an agent gets
access to new background information (e.g. role of an authority or
certificate revokal) his trust in a graph changes.


> >"
> >
> >This conflicts with what we write in  Chapter 4:
> >
> >"The meaning of a set of named graphs depends on a separate decision
> >which of the graphs to accept. We represent this decision as a set A of
> >nodes naming the accepted graphs. The meaning of a set of accepted named
> >graphs hA,Ni is given by taking the graph merge Sa2A N(a), and then
> >interpreting that graph using the semantics of RDF[?]. Any extension
> >semantics of RDF can be used; in this paper we uniformly use those of OWL
> >Full[4]."
> >
> >I totally agree with Jeremy, that for practical reasons we should have a
> >bipolar view on trust in the context of the Semantic Web at it's current
> >state of development. But the interesting question is: Is this bipolar
> >required for RDF semantics or OWL to work or can we be open to agents
> >continuum based trust models (modal logic ??) ?
> OWL and RDF are inherently bipolar (boolean 2-valued) in their basic
> assertional semantics. However, one can make assertions about a
> continuum within a boolean language, provided that you refer to the
> values somehow (which would likely be a good idea anyway). You can
> say that (it is true that) my trust in foo is 0.1637, or that my
> trust in foo is greater than my trust in baz. You can have a
> multivariate-valued trust datatype if you want :-)
> So I don't think we are *excluding* anything by having a
> truth-functional language to say it all in. The logic is one thing,
> the content another. Messing with the logic is usually not as
> practical as inventing an ontology of your own favorite continuum.
> Pat
> >
> >I don't mandate, that we should include this stuff in the paper, but it
> >would be nice if our approach wouldn't exclude future agents that might
> >continuum based trust models (and will have the unlimited processing
> >to reason based on these models :-). In addition it would make our work
> >attackable for the trust community.
> >
> >Chris
> --
> ---------------------------------------------------------------------
> IHMC (850)434 8903 or (650)494 3973   home
> 40 South Alcaniz St. (850)202 4416   office
> Pensacola (850)202 4440   fax
> FL 32501 (850)291 0667    cell
> phayes@ihmc.us       http://www.ihmc.us/users/phayes
Received on Thursday, 18 March 2004 06:09:01 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 22:32:25 UTC