Slide 1

BOF Goals

	Identify needs for explanation/proof work from the daml community
	Identify who has plans to work on DAML+OIL/OWL explanation/proof work
	Generate list of actionable items.
	Significant issues
	Recommendations/plan of action
	Discussion of “good” proofs (explanations)

Motivation

	Trust disclosure – trust inference rules, premises, recency, inference engine, …
	Interoperability – multiple owls interacting, proof composition, …
	Proof reuse – individual reuse, individual refinement, group reuse/refinement….

Issues

	Variable granularity   (lcf, pruning, etc)
	Degree of annotation for human readability  (human paraphrase in addition to machine readability)
	Agents should be able to verify proofs
	Proofs should be “nestable” and “queryable and/or reexecutable”
	Proof language should be ubiquitous
	Proofs should be incremental
	Confidence in proof steps should be expressible
	Daml-compliant inference engines should respond to client requests with “reasonable explanation” in the daml language
	Identifying rules   (naming,…)

Issues, continued

	System needs to be extensible with respect to inference rules, …
	Should include black box algorithms with trust annotation on black box
	What is trust?   Trust of inference rules, agent (might have additional granularity), source.  Look at solutions such as delegated trust in n3
	Proofs with true but not useful information- need techniques for pruning
	Are there techniques like Google’s reverse links that can help?
	If you want a “good explanation” that may impact the proof spec.  And what is a “good explanation”
	Where do ground facts ground out  (what granularity)
	Provenance or other annotations on information

Plans/People

	W3C – Contact:  Berners-Lee, Connolly,…
		Cwm will handle explanation and validation sometime
	Stanford – Contact:  McGuinness
		http://www.ksl.stanford.edu/projects/daml/Proof/
		DAML+OIL/OWL specification of proofs, examples, challenges…
		Implementation of explanation/proof browser for proofs/inference webs
		JTP reasoner is being made compatible with proof spec
	Cycorp –  Contact:  Steve Reed
		Explanation implementation of Stanford's design initial test subsumption,
		why assertion NOT assertable and make recommendations
	Agfa -   Contact:  Jos de Roo

Plans/People  cont.

	Teknowledge - Contact – Adam Pease
		Proof pruning, coordination
	UWF/IMHC - Contact - Pat Hayes
		Designing proofs for good explanation
	Northeastern University - Contact - Mitch Kokar
		Ontology for inconsistencies in DAML
	get pointer from pat on lcf….
	McGuinness will maintain list – send mail to [email protected] to update.

Actionable items

	Build and maintain list of contacts on explanation work on RDF-compliant systems – McGuinness
	Build a test ontology and set of test cases
		Possible domains – wine ontology, ….
	Draft DAML+OIL/owl spec for shareable proofs  and architecture
	Obtain comments on draft spec for shareable proofs - - Karlsruhe, RKF(SRI, KM, Northwestern, Boeing, …), Cycorp, …
	Interoperability tests  (at least Stanford and Cycorp)
	List of heuristics for pruning/presenting explanations

Slide 9