W3C home > Mailing lists > Public > public-prov-wg@w3.org > July 2011

Re: Agent Sub-Types

From: Paul Groth <p.t.groth@vu.nl>
Date: Tue, 19 Jul 2011 16:11:50 +0200
Message-ID: <4E2590A6.3000506@vu.nl>
To: Paolo Missier <Paolo.Missier@ncl.ac.uk>
CC: "public-prov-wg@w3.org" <public-prov-wg@w3.org>
Paolo,

I think what Reza is pointing to here is can provenance begin with a 
person. I think this is a fundamental thing to be able to make the 
destination between people and other stuff.

cheers
Paul

Paolo Missier wrote:
> Reza
>
> leaving (1) aside as we seem to agree on the specifics of Trust:
>> Paolo -
>>
>> In the case of "Trust", I think your solution works and it solves the
>> problem. There was another thread where someone suggested (I don't
>> remember who now, sorry) the same thing for system vs. human (this is
>> the user-agent vs. agent vs. etc.) discussion. In that case, it
>> doesn't work.
>>
>> So, here are my thoughts on your email:
>>
>>    1. Thanks for the solution of "Trust" being an attribute of some
>>       sort. This will work. Either you and/or I should send out a
>>       separate email as proposal to see what the greater group thinks.
>>    2. On profiles, I also responded to this, I'm very weary of
>>       profiles. There are lots of standards with profiles that have
>>       flopped. At least my strong preference is to be as specific as
>>       possible in the model and would rather risk being wrong on the
>>       specific versus generic side given examples like HTML, etc.
>>       where standards are more successful in gaining adoption when
>>       erring on the side of specific. I also realize we don't want to
>>       venture too far in there, but I think you get my point.
>>
> I do share your concerns on profiles. But you will agree that a line
> must be drawn somewhere in terms of our scope? the only criterion that
> comes to mind (not having implemented standards myself) is to stop when
> it becomes "domain-specific". Is that the idea? Even then, at some point
> you will have to give instructions for extensions, as your standard
> cannot be closed. The crucial difference with HTML is that we are not
> defining a formatting language, which we can confine to what we choose
> to, instead we are introducing concepts that seem to belong in some
> upper ontology, which by their nature are there to be extended.
>>
>> There is also another question that has come up that I do not find an
>> answer to that I believe is crucial to be answered in order for more
>> concrete definition on entities:
>>
>> Does the provenance of an entity begin beyond its physical origination?
>>
>> I'd like to get a binary answer from the perspective of this working
>> group. All the materials that I have studied regarding provenance in
>> the past indicates a "Yes" to above so I had assumed (perhaps
>> incorrectly) that the answer is "Yes". But I think it's crucial we
>> answer this question and document it on the wiki.
> Crisp off-the-cuff answer: provenance begins as early as there are
> observers that can make assertions to record it.
> However, you seem to refer to "when a thing assumes its current
> identity". That I don't know, but I have argued elsewhere that it
> doesn't matter. I can talk about the provenance of a car in terms of its
> components before they were assembled into a car, and the fact that the
> car had no identity before the assembly process does not bother me.
>
> atb -Paolo
>>
>>
>> On 7/18/11 10:50 AM, Paolo Missier wrote:
>>> Reza,
>>>
>>> it's worth trying to make progress on this as we are in the process
>>> of editing a document draft.
>>>
>>> I can see the dilemma:
>>> - the less you specify in the model, the more you risk
>>> incompatibilities as different implementations make their own choices
>>> to fill the gaps;
>>> - but by adding specific extensions to the top-level concepts in the
>>> model, you risk to make those choices arbitrary.
>>>
>>> But in this specific instance, I believe that "trusted" is one
>>> specific qualification of "Agent" that does not belong to the model,
>>> rather it belongs to applications that use the model (i.e., to
>>> /assess some measure of trust/).
>>> But I see your need for a "placeholder" where I can assert something
>>> about how trust for Agents. This is fine: let Trusted not be a
>>> sub-class of Agent, but let "Trust properties" can be properties of
>>> Agent. Would that be a problem? Any application that knows about
>>> trust would fill in into those properties.
>>>
>>> To repeat my proposal, I see Agent as a Role that any first-class
>>> entity in the model can take on when it is involved in relations that
>>> concern activities.
>>>
>>> The general lesson I see from this thread is that we urgently need to
>>> discuss how principled extension mechanisms ("profiles") make it into
>>> our proposal.
>>>
>>> atb -Paolo
>

-- 
Dr. Paul Groth (p.t.groth@vu.nl)
http://www.few.vu.nl/~pgroth/
Assistant Professor
Knowledge Representation & Reasoning Group
Artificial Intelligence Section
Department of Computer Science
VU University Amsterdam
Received on Tuesday, 19 July 2011 14:12:32 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Thursday, 26 April 2012 13:06:37 GMT