SV: [RDF] Non-fact statements

[Stefan Andersson]

> > > Non-fact statements from a non-specified model could just be placed in
> > > some special model that defaults to untrusted.
> > 
> > Well, actually, I've always found the division between 'fact' and
> > non-fact a bit silly. There is no such thing as a 'fact', only trusted
> > or non-trusted statements/data.
> 
> Yes. But you just have to believe in something before you can start
> mistrust.  

Hmm.. I would change that to 'without trust there can be no communication,
but without communication, there is nothing to trust.' So I guess you're
right. I think that starting point would be 'the things I know I know' - a
single trust statement. Or set of trust statements. ROOT statement. Hmmm..

> The system base logic has a whole lot of RDF implicit and
> explicit. Those base statements must be trusted.  My first thought was
> to consider fact statements the unquestioned truths and mark all
> "user" and imported statements as non-facts.

Actually, what we do is optimizing thru faith. There is nothing 'must' about
it. And I don't always trust my own memory. I could import faulty or
fraudulent data.

> > A model can be judged trustworthy based on its internal consistency, or
> > its backing by an authority. There is _always_ an issuer of the
> > statement, implicit or explicit. The trick is wheter the system
> > _remembers_ (is aware of and has retained the knowledge of) who the
> > issuer was.
> 
> Wraf ties a model to every statement. That model will hold that data.
> i don't know how much has been done in modeling this in RDF. Maby it
> would be better to wait with the details.  But the main functionality
> is to get a level of trust for a specific statement. that could maby
> be calculated from the statement content and the model.  That is: the
> trust can differ acording to the type of statement.

Hmmm. I'm not entirely with you on that last statement. But I can quote
Gerck - 'there are no degrees of trust. Either you trust, or you dont.' -
and I think he is right.
 
> > And - these are different levels of indirection.
> > 
> > If my system were to access the model
> > http://www.w3.org/1999/02/22-rdf-syntax-ns
> > the _primary_ _implicit_ issuer of the model would be the actualized
> > http daemon on www.w3.org.
> > 
> > Now, if the returned model would include some sort of standardized
> > 'attributed_to', that would be an _secondary_ _explicit_ statement of
> > issuer. 
> > One could imagine a _secondary_ _implicit_ statement of issuer based
> > upon the OS 'last_updated_by' or by file system rights.
> >
> > Now, if the _primary_ issuer wasn't trustworthy, there is no way you
> > could trust the issued model, and thus not the attribution in it. That's
> > pretty banal. But the opposite is true as well. Just because the primary
> > source is trusted, that does not mean the secondary has to be.
> 
> Yes. That will be contained in the model metadata.

I think the 'model' should be treated simply as an optimization by concept,
and not as a built-in prerequisite. But you already know my view on this.
 
> > If I was to try to summarize, it would be by pointing to a
> > thought-provoking paper on trust published by E Gerck and the MCG:
> > 
> > http://www.mcg.org.br/trustdef.htm
> > 
> > Read it. I mean it. Then read it again.

> and you call that a summary?!?  It's long!  I have printed it and will
look at it later.

Ah, sorry, sorrryyy! No, my summary was the act of pointing to the document,
not the document in itself. The document is long. And hard to penetrate.
But, I think, rewarding.

Is there anybody else on the list that know anything about Gerck?

Kram,
/Stefan