- From: dehora <bill@dehora.fsnet.co.uk>
- Date: Wed, 18 Apr 2001 00:35:55 +0100
- To: "RDF-Interest" <www-rdf-interest@w3.org>
: TBL speaks of signing metadata to provide provenance (the source of the : information). Wouldn't sources of spoofed metadata eventually (rapidly?) : develop a ruputaion for unreliability? : : Fundamentally, the reliability of metadata is an issue, and ideas of : provenance provide a means of assessing the source of the metadata. Well there's a few things worth pointing out: -We need code to generate such provenance. -We need code to evaluate such provenance. -We need web architecture for the code and the relating of it to metadata. I understand Tim Berners Lee believes digital signatures to be pretty fundamental to furthering web architecture. They may well arise due to commercial and political pressure to target and legislate people and behaviour in the first instance. -It's not always possible to determine the difference between what JL Austin called "misfires" and "abuses" of assertions/informations (more generally speech acts): best you can do is identify "infelicities". This a bit like trying to determine why exactly the server never responded to my request. -How would we separate infelicity from the way people drift over time in the way in which they classify things? -If there exists signed metadata, or metadata with a provenance of signatures (two careful owners, 10,300 requests on the clock...) and that metadata is evaluated against a graph of reputations or past metadata, it won't take long to figure out that what might be good is the web equivalent of something like certainty factors, epistemic probability or symbolic argumentation (where we don't have any numbers to crunch) to help our machines manage uncertainty about metadata. Does that imply less of a role for reification? -What does it mean to sign metadata? My word is my bond? -Possibly the kinds of technologies used for outlier analysis in machine learning, data mining and fraud detection may have a role to play here also. Of course they're largely statistical in nature. One way or another, validating uncertain metadata seems like a lot of work. regards, Bill de hÓra
Received on Tuesday, 17 April 2001 19:36:12 UTC