Re: off-topic on Beijing Re: BOF meeting on Semantic Web Search Engines at WWW 2008

On Friday 25 April 2008 2:36, Dan Brickley wrote:
> For 
> better or worse, this list has to stay focussed on on making technical 
> progress. Sorry if that sounds somehow callous...
[...]
> As chair of this (very 
> international) Interest Group I don't want to be in a situation of 
> saying which situations are urgent, awful or disgusting enough to use 
> this mailing list for emergency appeals.

Point taken.  I can't exactly apologize, but I understand, and I really 
appreciate the thoughtful replies.

> There are other, better ways to reach people, persuade people of 
> evidence, showing people what life elsewhere is like. And figuring out 
> how to improve those mechanisms is 100% on-topic for this list: the 
> Semantic Web is a project to improve the Web so that it better reflects 
> what is happening in the world around us, a world seen through layers 
> competing, interlinked claims and counter-claims. (Anyone who tells you 
> otherwise has got lost in the detail.)

Yes!  I took a look at debategraph.org, that's an interesting project.  Having 
something like that, even if they are not using RDF now, will show what terms 
are useful to make what they call debate maps.

> We do have pagerank, blogs, and so 
> on, but nothing structured in terms of evidential support for specific 
> claims. I think we can and should do better, and that the focus of the 
> Semantic Web community would be profitably spent on this area of work. I 
> firmly believe the Web will mature to give us a better claim-based, 
> provenance-based infrastructure for evaluating such claims

Sure - its a hard problem, but I think a very good one for the point where 
grassroots meets semantics.  Trust is critical too, a la Guha - as far as I 
understand a lot of people are basing trust networks on this work
 
http://citeseer.ist.psu.edu/guha04propagation.html
> 
>  > Ok, this is not RDF-related unless someone has the vocabulary to say it
> 
> I don't think the technical issue is exactly one of vocabulary here. 
> Rather it is one of being able to make an overwhelming case "this is 
> happening" grounded in documentary evidence published in the Web. Partly 
> a matter of weighing the credibility and authority of sources, of 
> providing a representation for the claims those sources make about the 
> world. 

Well - I know I tend to jump too fast to the concrete representation, but it 
seems to me that trust applies to at least three types of output from a 
person: 

	1st hand truthfulness/observation
	2nd hand quality of reporting
	judgement/analysis

There are many people who I would absolutely trust as to truthfullness, but 
not at all agree with their judgement and in some cases the validity of their 
analysis.  So I'd want to be able to say things like

	X observes Y
	X reports Y
	X thinks Y

and then say 

	Z trusts X.observation  with a value of 1.0
	Z trusts X.reporting  with a value of 0.8
	Z trusts X.judgement with a value of 0.5

Now given a group of statements by X, we can calculate from Z's point of view 
the likelihood that they are true.  

In an earlier post I tried to make sense out of the idea of applying neural 
networks.  Its a kind of learning system of weighted connections.  Instead of 
the system saying 'beep beep error, one of these has to be thrown out' when 
inferences conflict, it just goes back and corrects the weights until it gets 
into a sort of maximally stable state. This state might contain conflicts but 
with weights of less than 1.0.  
  Traditional neural nets are often envisioned as a pyramid trying to answer a 
single question from a bunch of input.   But they don't have to be that way - 
a neural net can be looked at as a multiply connected graph with various 
strengths of connection (though the connections aren't meaningful properties, 
usually only the nodes are assigned meaning).  We don't have to use 
traditional neural nets, but the idea of a statement having a weighted value 
other than 1 or 0 seems necessary to a model of evidence for real world 
assertions.

> But also a matter of user interface (something often neglected in  
> the SemWeb scene): how do we get from giving people access to the raw 
> facts, ... to getting them to care, and to act?

Its that last mapping step into people's minds...like PDM said, usability is a 
key.  I'm not sure we need a spec so much as some great examples!  

Back on the very first point.   I don't like making people uncomfortable and 
I'm sorry for that.  But it seems like those who are doing awful things 
depend on good people wanting to just do work and mind our own business.

Thanks again for letting this discussion happen - on a technical level!

--Golda

Received on Tuesday, 29 April 2008 14:30:42 UTC