Re: PROV-ISSUE-203: Proposal to amend definition and usage of Plan in PROV-DM [prov-dm]

On Dec 15, 2011, at 3:21 PM, Luc Moreau wrote:

> Hi Stephan,
> 
> Some comments interleaved.
> 
> On 15/12/11 19:52, Provenance Working Group Issue Tracker wrote:
>> PROV-ISSUE-203: Proposal to amend definition and usage of Plan in PROV-DM [prov-dm]
>> 
>> http://www.w3.org/2011/prov/track/issues/203
>> 
>> Raised by: Stephan Zednik
>> On product: prov-dm
>> 
>> Dear all,
>> 
>> I would like to suggest an amended definition and usage for plan in PROV-DM.
>> 
>> CONTEXT:
>> 
>> 1. The concept of plan is currently related to activity via a plan link, which is a specialization of an activity association record; as a consequence all plans are agents.
>> 
>> http://dvcs.w3.org/hg/prov/raw-file/default/model/ProvenanceModel.html#record-planLink
>>   
> 
> It's important we understand why the consequence is that all plans are agents.
> The inference association-agent
> http://dvcs.w3.org/hg/prov/raw-file/default/model/ProvenanceModel.html#association-Agent
> forces us to consider anything associated with an activity to be an agent.
> 
> This inference leads already to unnatural agents in prov-dm.
> Look at the example which directly follows
> 
> It contains an entity --- the request to start an activity.
> It must also be regarded as an agent.
> I think this is equally problematic as regarding a plan as an agent.

Hmm, good point.  To me it reinforces the need to use some other relation than associatedWith with plan.

> 
> I feel that if we want to allow plan as entity, than we should also allow this request example to be an entity.
>> 2. This means that according to the definition of agent at http://dvcs.w3.org/hg/prov/raw-file/default/model/ProvenanceModel.html#section-entity-activity-agent, plans 1) take an active role in an activity and 2) can be assigned some degree of responsibility for the activity taking place.
>>   
> 
>> 3. The relation between agents and the plan that guides their actions is not explicit in PROV-DM.  We could infer that all agents involved in an activity act according to the activity's linked plan, but this prevents us from describing an activity were multiple agents act guided by different plans.
>> 
>> PROPOSAL 1: Amend the definition of plan in PROV-DM to:
>> 
>> "In the context of PROV-DM, a plan should be understood as the description of a set of actions or steps intended by one or more agents to achieve some goal."
>> 
>>   
> 
> +1
>> PROPOSAL 2: Amend plan link record such that it is not a specialization of an activity association record.
>> 
>> Comment: I do not think we should define all plans as agents.  By our existing definition a plan is a "set of actions or steps ... to achieve some goal."  It is a description (usually in the form of a document) of the action or actions an agent should take to achieve a desired goal.
>>   
> 
> If the intent is to allow plans to be entities and not agents, it is not the only approach.
> I would like to suggest that:
> 1. Constraint http://dvcs.w3.org/hg/prov/raw-file/default/model/ProvenanceModel.html#association-Agent
>   should not hold, since the notion of just being associated with an activity does not imply agency.
> 
> 2. Define agency independently from wasAssociatedWith.
> 
> If we do this, then I think we can keep hadPlan as a specialization of wasAssociatedWith.
> 
> My rationale is to try and minimize the number of distinct concepts/relations in the model.
> 
> 
> 
>> As a description of the actions an agent should take, it does not make sense for a plan to have responsibility.  The responsible party for a plan (e.g. owner, creator) could have some responsibility for the outcome of the activity, but the description of the "set of actions or steps intended to achieve some goal" would not.
>>   
> 
> Commonly, we say: we followed the plan, but it didn't work. It's the plan's fault. I see no problem with a notion of "responsibility".

In PROV-DM we define a mild version of responsibility in the from of the actedOnBehalfOf relation.  This is the only definition for responsibility I saw in PROV-DM - e.g. no mention of accountability, obligation, or duty, so I will use it to scope any discussion of responsibility.

Can a plan act on behalf of an agent?  Can an agent act on behalf of a plan (which implies the plan could have itself acted)?

I would argue that a plan does not have agency, and therefore cannot act on behalf of another or have another act on its behalf.


>> A plan without an agent is (generally) inert.  Something must follow/execute the plan for there to be action.  A recipe is inert - bread is made by the baker.  Driving directions are inert, a driver must control the vehicle from the trip's point of origin to its destination.  If driving directions are incorrect, it would be the source of the driving directions (e.g. Google Maps) that would bear responsibility for the failure of the activity to result in the desired objective.
>> 
>> Software and especially workflows are a fuzzy area; you can reasonably define any software as a document of steps that a microprocessor is capable of executing.
>> 
>> To allow users to record provenance in a way that is natural to them I suggest we not define plans and agents as disjoint, but I do not think we should define all plans as agents.
>> 
>> PROPOSAL 3: Amend plan link record such that it can express which agents in the activity are following the plan.
>>   
> 
> +1
>> PROPOSAL 4: We should define a relation to describe the responsible party for a plan.
>>   
> 
> +1.   Do you have a name?

Hmm, based on the current modeling of responsibility in PROV-DM as defined by delegation of action from one agent to another...  I am not sure how plan fits into responsibility.

I was originally thinking about accountability, and that the creators/owners of a plan would be accountable (e.g. responsible) for the content of the plan.  If the plan was faulty, the owners of the plan would be at fault.

--Stephan

> 
> 
> Luc
> 
>> --Stephan
>> 
>> 
>> 
>>   
> 
>

Received on Thursday, 15 December 2011 23:19:18 UTC