- From: Graham Klyne <gk-lists@dial.pipex.com>
- Date: Wed, 01 Nov 2000 12:15:59 +0000
- To: pat hayes <phayes@ai.uwf.edu>
- Cc: "Hart, Lewis" <lhart@grci.com>, www-rdf-logic@w3.org
At 03:53 PM 10/23/00 -0500, pat hayes wrote: >Lewis Hart: >>..... >>A search engine, for example, must keep track of what is >>believed where/by whom. Keeping track of this could be tedious. >>If a search engine imports every ontology it comes across, the search >>engine's own, internal ontology will eventually be quite a mess. Rather, >>the engine needs to build articulations among it's internal ontology and >>the external ones on the web. Essentially, meta facts about what is believed >>by external systems, with reference to its own beliefs. This will allow >>a representation of a set of sites that believe daml:Thing isa ec:Computer >>without the search engine believing it internally. >> >>The good things about this are: >> >>1) Anybody can still say anything. >>2) There is no central authority. >> 2a) Authorities will arise out of usage. >>3) The scope of bad a ontology is limited. >> 3a) Search engines can cut them off at the root if desired. >>4) Information can still be found. >> >>The Bad things are: >> >>1) Systems will have to delineate between what it knows about >> and what it believes. >>2) Keeping track of this could be resource intense. >>3) Some sites will think everything is a computer. > >I agree this is the right vision. As long as every assertion can be >'traced' to its source, then it will be possible to disentangle who has >been saying what about who. However, if we allow inferences to be drawn, >stored and then retrieved, things could get a little more complicated. It >might be that something that bad: said allows one to infer something which >does not itself refer to anything bad#. To keep track of things properly >then would require not just tracking the URI's in the sentence being >examined, but also those in every sentence that was used to derive that >sentence. This means that (barring some new security mechanism) one should >never archive a bare conclusion, but only archive a conclusion together >with its derivation (logical proof). > >This isnt an issue yet, but I think it will be very soon. I thought that was all part of the DAML/SWeLL plan. I think that introduces some scalability concerns (assuming 'proofs' are complex), but I'd suggest that might be partly overcome by having assurances by trusted parties that assert some statement has been inferred by 'safe' means (for some value of 'safe', presumably linked to a policy of the assuring party). I think this approach mirrors the way that people deal with information and trust: we don't insist on knowing the justification for everything we believe -- many things we take on trust if it is assured by 'sufficient' authority. #g ------------------------------------------------------------ Graham Klyne Content Technologies Ltd. Strategic Research <http://www.mimesweeper.com> <Graham.Klyne@mimesweeper.com> ------------------------------------------------------------
Received on Wednesday, 1 November 2000 12:51:06 UTC