Trust in Information Sources as a Source for Trust: A Fuzzy Approach [Protected Link] Trust in Information Sources as a Source for Trust: A Fuzzy Approach


  author         = {Cristiano Castelfranchi and Rino Falcone and Giovanni Pezzulo},
  title          = {Trust in Information Sources as a Source for Trust: A Fuzzy Approach},
  year           = {2003},
  key-cites      = {castelfranchi-1998a, falcone-2001a, jonker-1999a, kosko-1986a, schillo-1999a},
  review-dates   = {2004-06-04, 2004-06-05, 2004-06-06},
  topics         = {fuzzy-logic, trust-decomposition},
  address        = {Melbourne, Australia},
  pages          = {89--96},
  read-status    = {reviewed},
  hardcopy       = {yes},
  publisher      = {ACM},
  value          = {dc},
  protected      = {no},
  booktitle      = {Proceedings of AAMAS '03},
  month          = {Jul},
  key            = {castelfranchi-2003a}


This paper did not help goals, but it does touch upon the decomposition of trust.  

The aim or the paper is to show that the "socio-cognitive approach" to trust allows the decomposition and analysis of trust in component parts.  Internal versus external is top/major separation.  The discussion of external components are related to trust in the way any external factors affect many decisions, doesn't seem these should be unique to a trust model.  The most interesting component that relates to current work is the notion of dishonesty/deceit/cheating/malevolence competence (which is further decomposed here, but not according to any particular principles).

The authors argue for Fuzzy Cognitive Maps (FCMs) to integrate trust information and allow for intuition and variation inherent in personality types.

The overall framework of trust is alluded to, but isn't fully defined; I suspect their previous papers do not define their model of trust rigorously either.  The paper does not address learning structure or parameter values for FCMs.  No clear re-usable trust framework is defined, just a shopping list of possible aspects that could be brought together.

An irritation, quotes like the following are content-free (given this paper, at any rate):

"In fact, our model introduced a degree of trust instead of a simple probability factor since it permits to evaluate the trustfulness in a rational way."

Key Factors

How placed in context (other work): Placed in context of their prior work and Kosko's FCM technique. References other trust papers {jonker-1999a, schillo-1999a}. Also references personality trait and shallow AI models of personality and emotions.  This is essentially a very limited illustration of their trust model using FCM.

Problem Addressed: Show how a trust model based on beliefs and their credibility is relevant in deciding whether to use a real doctor visit to the home or a automatice medical system in both an emergency and a routine situation.

Main Claim and Evidence: I guess it would be that FCMs can model trust effectively.  They don't really demonstrate a compelling case, as they appear to have hand-coded a bunch of values and got the answer one would expect.

Assumptions: FCM is best way (or sufficient way) to model trust.  They do not really address dishonesty (but mention it).  A lot of assumptions have gone into the setting of values.  

Next steps: Clear up how the FCM structure is to be created and how we learn the values and structure.  Really build an application for a domain.  Explain the framework in detail.

Remaining open questions: How does one go about choosing the correct features, model, and decomposition of trust?  How does one choose fuzzy values? 


Originality is average. Builds only slightly on prior work.  Somewhat novel in application to medical systems, but not very compelling.

Overall contribution is average.  Not clear how this really pushes forward the general ideas that were presumably expressed in the prior paper (may have to visit prior papers).

The quality of organization is good. Pretty normal flow, though a lot of sections for a short paper.

The quality of writing is good. Some serious problems in some sentences, but overall comprehensible.