Artificial Intelligence PTF

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.
===== AI PTF Interim Virtual Meeting on 4 May 2020 =====

Notes by [[claude@omg.org|Claude Baudoin]]

==== Attendees ====
  * Claude Baudoin (cébé IT & Knowledge Management)
  * Rachel Bauer (prostep ivip)
  * Roy Bell (Ratyheon)
  * Mike Bennett (Hypercube / IOTA)
  * Roger Burkhart (John Deere)
  * Fred Cummins (Agile Enterprise Design)
  * Karl Gosejacob (GOSEJACOB)
  * Alan Johnston (MIMOSA)
  * Dr. Holger Kenn (Microsoft)
  * Raynier “René” van Egmond (XFintell)

==== 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:
  - **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. [[https://drive.google.com/open?id=1WClmNsL2TeY29hjOaZWrbP9wEj_2KodKFtpkCXmNVA4|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.
  - **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.
  - **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.
  - 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.
  - 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.
  - Karl says we need some “domain-specific objects.” The suggestion needs to be explained better to this group at a future meeting.
  - 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 ===
  * Karl suggests just one more call in 4 weeks before the June meeting.
  * Fred: we should leave enough time between that call and “virtual Orlando” to fine-tune the agenda.
  * Claude: we can start an agenda and revise it later if we need to.
  * Claude: action item to do a Doodle poll for one date: offer May 26-29, June 2-5 (May 25 is Memorial Day in the U.S., June 1 is Pentecost Monday in Europe)

=== Agenda of June meeting ===

  * Preference for Tuesday, Wed or Thursday (June 23, 24, or 25)?
    * Select through another Doodle poll?
    * Alan says ISO TC 184 plenary is potentially on June 23-24. Even if it is virtual, it may be on those dates and Alan would not be available. This would lead to a preference for June 25, but we still need to check with others.
  * Presenters?
    * Karl on “sample lifecycle model” idea, and on domain-specific models.
    * Rachel Bauer: short update on the activities of prostep ivip, with a more complete presentation at the planned September “AI Forum.”
    * Claude can follow up with Labelbox, Torch.ai, JAIC, Mace Fusion, and others…: those are also candidates for longer presentations in September.
    * Alan: can provide an update on ISO 18101 and ties with ISO/IEC on interoperability (but needs to focus on the AI aspects, because his work straddles ManTIS, SysEng, and AI)
  * Breakout sessions on taxonomy, Ai ethics, explainability, ...?
Artificial Intelligence PTF

User Tools

Site Tools

Page Tools
