Re: Documents, Cars, Hills, and Valleys

On Fri, 19 Apr 2002, Dan Brickley wrote:

> representation. Whether the representation is of a physical thing or not
> is irrelevant to this problem.
> 
> http://www.w3.org/Icons/WWW/w3c_home
> 
> If you say that an HTTP header (or http:-URIref) identifies a visual-work,
> and then a GET returns an image/jpeg or image/png bag of bytes, how do we
> distinguish between these three things? A partial answer might be
> reference-by-description: 'the image/png representation whose sha1sum is
> yaddayadda'.
> 
> Why make work for ourselves? What worldy benefit is there in coming up
> with criteria for splitting the world into two huge disjoint categories:
> 'things that can be named with http:-uris, and things that can't'? Sure,
> the HTTP spec has woolly words about network / informational / resources
> and/or services, but what inferences do such vague categories buy us?
> Whatever happened to minimally constraining architectures...?

Dan,
the meat of your argument seems to be that because content negotiation
means that we can't tie any one http URI to a particular bit stream, then
we might as well give up and say that an http URI can identify anything?

My concern is not really with the rights and wrongs of using http URIs to
identify non-http accessible resources, but that if we live in a world
where the same http URI can identify both 'the car' and 'the digital image
of the car' then we have a situation that is so non-intuitive to the
average user of RDF that the whole thing is doomed.

This discussion seems to me to lie at the heart of the problems with RDF.  
Here we have a system that allows us to make clear, simple statements
about resources identified by URIs, yet this list can't even agree what
kinds of resources are being identified by particular kinds of URIs!

Of course, many people will be quite happy to use http URIs to identify
whatever suits them at the point they are creating the RDF description.  
The real problem will be in 2 or 3 years time (assuming that RDF is still
around then) when we will have many hundreds, thousands or even millions
of descriptions out there, but no way of knowing whether those
descriptions refer to the car or the image of the car :-(

The only real solution lies in encouraging people to uniquely and
unambiguosly identify the particular thing that is being described (the
car or the image of the car, the person or the home page of the person,
etc.) using appropriate URIs.  I would argue that an http URI is often not
the most appropriate identifier for this, since (as you point out above)
an http URI often doesn't uniquely identify a resource in a non-ambiguous
way.

Andy
--
Distributed Systems, UKOLN, University of Bath, Bath, BA2 7AY, UK
http://www.ukoln.ac.uk/ukoln/staff/a.powell       +44 1225 383933
Resource Discovery Network http://www.rdn.ac.uk/

Received on Friday, 19 April 2002 19:35:43 UTC