Re: ::scr semantic

["Alaric B. Snell" <alaric@xxxxxxxxxxxxxxxx>]

On Monday 05 August 2002 12:17, you wrote:
> > Small semantic weblets (intrasemanticwebs? :-) might work out, since they
> > then work like distributed Prolog systems, but nobody's yet found a way
> > to make it scale to a global system.
>
> that begets the question, is an integrated global system desirable, rather
> than is it feasible? there are interesting ways of doing search, discovery
> and navigation in distributed metadata storage; i'd especially refer you
> to http://www.neurogrid.net/Decentralized_Meta-Data_Strategies-neat.html

I think it's desirable - when searching my disk for things to do with cheese, 
the search algorithm would benefit both from my private knowledge base's 
record that I am designing my own cheese called 'Brob' so that references to 
'brob' count as cheese references and from exploring the larger Internet to 
find names for other cheeses.

[lack of an agreed shared dictionary]
> this looks like what RDF, DAML and a host of small-scale RDF parsers and
> spiders, already do: the negotiation between namespaces and resolving how
> they might interconnect and alter each others implications. the
> kendall/bijan articles on xml.com i posted two mails ago go into a lot of
> this, and they're both over a year old.

Yep.

> i'm learning prolog now, and if you are into it, i think you'd really like
> http://logicmoo.sf.net

I learnt Prolog when I was a wee lad, when I first started trying to design 
global knowledge bases and failing :-)

> > perhaps some kind of relationship synonym directory
> > maintained by volunteers, or a central registry that encourages people to
> > find an existing relationship rather than searching for a new one...
>
> here i think scope and i think decentralisation. i also think bots; this
> stuff can be done with dynamic bot arbitration pretty successfully, and
> for me the bots also provide a good hook to hang semweb applications from.

Define 'bot'. From where I sit, 'bot' means 'computer program that interacts 
with a system through the same interface a human would' - IRC bots, Quake 
bots, etc. Which is probably a bit of a lower level definition than you're 
referring to here!

> > But anyway. What will actually pull together all of these assertions to
> > come to conclusions (and how will it deal with contradictory or
> > untrustworthy information)? I see something like Google being the only
> > thing that can actually make semantic inferences with much utility; for
> > everyone else it'll be limited to their corporate intranets or they will
> > explicitly subscribe to information services - "sign up to get hold of a
> > huge knowledge base about cats", rather than the more peer to peer model
> > of publishing we currently have.
>
> so you've read http://www.ftrain.com/google_takes_all.html and, possibly,
> not thought about it hard enough ;)

That's a fictional account that skips over many of the hard parts!

The model they're predicting is that Google would just parse all that RDF 
into a single centralised (if heavily replicated for load balancing and 
reliability) indexed knowledge base which the unwashed masses submit their 
queries to and wait for the response.

Is that enough? Can we leave such a crucial part of the future infrastructure 
to a company that holds a monopoly? Would Google not just become the next 
Verisign / Network Solutions / Microsoft, adding their own 'proprietary 
extensions' to RDF to lock competitors out and to take control of the 
publishing process?

The distributed knowledge systems I designed as a child didn't require any 
centralised authorities as part of the routine query process. But it did have 
an advisory group who would produce a symbol dictionary to agree on symbols 
for such concept as 'is-a' and 'danger', and to hammer out libraries of 
'common sense knowledge' - UK, GB, England are interchangeable in most cases, 
London is-in UK, etc.

Also is used a form of temporal logic to include historical facts rather than 
just current affairs, without having to define seperate past tenses of all 
the relations (X will-be-a Y if Z, X is-a Y, X was-a Y from T1 to T2, etc) - 
not sure if RDF has that, but I've never heard it mentioned.

> > So I'm skeptical!
>
> well, i'm wildly optimistic :) i'm also running out of steam a bit, and
> have bot glue to cook up before dorkbot wednesday

Have a good time!

> zx

ABS