Re: WebID-ISSUE-13 (bblfish): Develop a logic of trust

On 30 Jan 2011, at 15:22, Dan Brickley wrote:

> FWIW Google's Social Graph API has some basic pieces of this logic: a
> notion of transitivity, following "rel=me" across sites (and some FOAF
> version too, though possibly buggy), plus a kind of skepticism
> structured in terms of reciprocal claims. If my homepage claims a
> youtube page as mine, and vice-versa, that's taken as being somewhat
> more reliable than an unverified link.

Thanks for that reference.

> This kind of 'claim graph analytics' is well worth investigating. Most
> of the formal attention around RDF so far (OWL etc) has focused on
> simple flat collections of triples, rather than on figuring out how to
> deal with different parties and documents making competing and
> supporting claims.

Yes, fully agree. OWL is a very important logical tool but it deals with what 
one could think of as Object Oriented inference. It is reasoning within a 
body of truth. The inferencing we need to look at is prior to that, in
how we come to trust something as true in the first place, given that we already
trust some things - usually within some context or for some purpose.  Of course 
OWL can still be very important in  this reasoning because 

 - it could help deduce from what is written in a graph, what that graph implies, and
   so help confirm that it confirms the other graph
 - help find that two graphs are contradictory: in which case one has to chose one or the
   other graph, or make a decision of which parts of one of the graphs one accepts
   (something like coming to a compromise)
 - we can use owl in the graphs of facts we know for sure: such as when and from where
   we fetched a graph
 - ...more surely...

That does make things sound quite advanced, and we can certainly get far 
with a lot less. 

So a rule that would give access to a resource would read something like this

Give access to group of resources R to whoever is the union of
   { J }
   the set of J's close friends as he declares them in his profile
where you can take as evidence for other names of those friends information 
from their profile

Now one has to be careful. Depending on whether you take the claims of J's 
friends altogether or just for each friend what that friend says on his 
profile you will get a  different result. So imagine that two close 
friends of J are Mathew and Judas. If Judas claims in the anonymous
profile which he used to login with his WebID that 

<http://anonymous.example/p/123#me> owl:sameAs <> .

And if he then places in his profile
that same statement as above, then a simple engine could end up 
authorizing him as Mathew.

(There will be a log of that but it may be too late by that time)

So even though this is somewhat outside the direct scope of the WebID group,
and closer to the ACL group, I think we should work on some of the most useful 
rules like this as it will help explain how this can have wider consequences.
WebID just by itself may be a bit too dry for some.

Of course if another group does a great job of that, then we can just point 
to there work.


> cheers,
> Dan

Social Web Architect

Received on Sunday, 30 January 2011 15:51:55 UTC