Table of Contents

AI PTF Interim Virtual Meeting on 4 May 2020

Notes by Claude Baudoin

Attendees

What can we do offline before the June meeting?

OMG’s Q2 Technical Meeting will be virtual again. The AI PTF will be on June 23, 24, or 25 (t.b.d.)

Three actions seem to be possible offline between now and then:

  1. Finish the analysis of the 98 NIST RFI responses – almost half of them are done
    • Claude showed the Google Sheet used to summarize and rate the responses. Each entry is hyperlinked to the actual response PDF file. Read-only link. Do not share this link outside of the AI PTF, as the document contains candid feedback about some of the responses.
    • Volunteers to help: René, Alan. Claude will share with them another link that allows them to edit the document.
    • Derive a catalog of relevant standards mentioned in the various responses.
    • Try to derive a set of priorities from the level of interest by respondents.
  2. Mutual introduction document
    • René suggested “a small document for all to introduce ourselves and give our topics of interest to coordinate/align activities.”
    • Claude will create a simple template and invite people to fill it.
  3. Taxonomy of AI
    • A taxonomy or ontology was suggested back in December.
    • Claude: it probably won’t have rules or relations other than hierarchy, so it’s really a taxonomy rather than a full ontology (with apology to ontologists in the group).
    • There’s one in the search menu for the AAAI’s journal, it was reproduced in an Appendix to the OMG AI Strategy paper. There’s also one on Wikipedia. So do we really need to create one, or is it useful to synthesize multiple such sources, ,and perhaps publish the result using a standard such as SKOS (or OWL if it turned out to be a real ontology)?
    • This could be handled by a breakout Working Group rather than the full PTF. The WG would present its work for review and approval.
    • Alan says this could help feed some of the other activities, such as alignment with other domain-specific standard work.

Ongoing activities (beyond June 2020)

Alan: we need to coordinate and avoid redundancy between OMG and his activities (diagnostic and prognostic for industrial systems) in ISO/IEC SC 4/TC 184/WG 6, including the ISO 18101 standard, “interoperability for asset-intensive industries,” digital twin, analytics. We need to ncorporate each other’s work by reference instead of replicating it.

Suggested AI PTF Deliverables / Work Items

This section is about formal deliverables such as RFIs, RFPs, RFCs or Discussion Papers not otherwise mentioned.

  1. Karl: An AI system should have a sample and a “sample lifecycle object” – auditors talk about “sampling risk.” This is to verify the reproducibility of the system’s results.
    • Holger asked for clarification, but then said this would also be useful for bias avoidance. How could this be formalized?
    • Karl says it’s doable.
    • Holger says the sampling process is not always formalized.
  2. Karl recently contributed to the Association of German Industries (BDI) position regarding the European Commission’s White Paper on AI of 19 Feb 2020. The BDI comments will be finalized soon and then translated into English, upon which he will share them with us. He will be able to commit in a subsequent meeting to giving us a presentation on this.
  3. Karl says we need some “domain-specific objects.” The suggestion needs to be explained better to this group at a future meeting.
  4. René (via chat) : deterministic algorithms formalization - think along the lines of OMG’s SACM (Structured Assurance Case Metamodel) and SBVR (Semantics of Business Vocabulary and Business Rules) specifications.

Next Meetings

Frequency of conference calls

Agenda of June meeting