Re: Showing the Semantic Web

> The reasoner has no choice but to believe that statement, unless it  
> runs up against an ontological constraint, right? That's what I mean  
> by absolute truth. If you give a reasoner a set of assertions, it  
> must assume them to be true unless it conflicts with the related  
> ontology. On the web, this is a ridiculous assumption. It makes every  
> reasoner gullible.

I think of reasoners being even more dumb than that.  They believe
everything they see, and deduce what they can.  Of course they should
keep track of the sources which led to each conclusion, and one should
be able to control what they see.  If they reach a contradiction (and
notice, instead of deducing the world) they can say, "Sorry, but my
current sources [_details_] contradict each other."

I do expect some defeasible reasoning around using web pages which have
expired.  

And I tend to imagine limitting reasoners to a combination of trusted
sources and sources reached by following certain kinds of links out of
the queries themselves.   The kinds of trust control needed depend on
the application, of course.

> It would be better if every single assertion was implicitly reified  
> by its containing URI, with the *option* of attaching even more  
> metadata to the statement-about-a-statement (ie, which individual or  
> entity in particular is making these assertions, when they were made,  
> etc). Then the reasoner would be able to parse many identical  
> assertions, and rate their validity based on any number of factors.  
> For instance,  you could ask such a reasoner (although probably not  
> in English): "do French people believe that Jerry Lewis is funny?"  
> Even with no such attributes attached to a statement, the results of  
> any query you make could be qualified based on their popularity.

I think you can do that with the approach I'm suggesting.  I'm just
saying that every time you want to point to an assertion, you have to
put it in a file by itself on the web.   I imagine typically you'll want
to group them anyway.

   -- sandro

Received on Thursday, 23 February 2006 22:44:02 UTC