Notes on general issues surrounding trust in the semantic web

Contents

1 Introduction
2 Importing trust from the web, pgp, etc
3 Problems with broad sem web (attacks)
4 Annotation, trust, etc.

1 Introduction

2 Importing t from the web, pgp, etc

Note: trust here is at a syntactic not nec. semantic level.

2.1 From the web

2.2 From PGP

2.3 XML / RDF signatures

Also: signature splitting: subset of information exposable? Maybe direct from the source (capabilities possible here)

3 Problems with sem web (attacks)

Motivations: accident, misapplication, bugs, maliciousness, "a fast buck". Eg, the mess that email systems are currently in: if apps (like FOAF) start to be seen as having concrete value, they will be subject to attack.

3.1 Explanatory view: what is the semantic web?

Note: cross-application interop. Application-specific semantics? How much can be captured using rdf, rdfs, owl, rdf-rules, ... ?

Taking a holistic view, how do technologies fare against attacks, misapplications, gaming? How does a sem web application operate? Analogy with a person: discovering new sources, new data, integrating with existing knowledge, in order to reach a goal. Where does it go? How does it rate stuff? When does it stop chasing links?

1. Contradictions eg, OWL construct. [ A0 = A1 ], [ A1 = A2 ] , ... [ An-1 = An ] , [ An != A0 != {} ]. Take any subset, they're fine. A contradiction can arise when all are considered. Using the introduction of contradictions isn't reasonable since ANY subset is apparently ok.

   Leads to: maybe large-scale inference or other mechanisms, must be proof against this (relevant logics?). Maybe strong inference on the document level - a document-centric approach.

2. FOAF, "everyone is the same person" using eg, inverse-functional-property
3. RDF, just making lots of spurious assertions.

Lead maybe to the notion of "reputation" - closely related with annotation. Reputation can mean many things: how reliable is this data? Can I trust the source? Also: how was it collected? How has it been processed? Also: provenance important. (Pointer to demo)

4 Annotation, trust, etc.

Example within RDF of simple annotation scheme. Context: FOAF + annotations. TU/TD. More fine-grained than simple "rating" but not overcomplex. [ Example of this: Hofstadter, Penrose both have expert knowledge in the subject of (broadly) mechanisability of human intelligence. Yet appear to draw opposing conclusions. So rate knowledge, opinions differently? Difficult problem in general ] Annotation of services? What about "service as data", eg mapping axioms.

Appendex A. References

[example]

Example reference This is not a pipe.