Actions

Ontolog Forum

Revision as of 14:06, 12 June 2019 by imported>KennethBaclawski (→‎Proceedings)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Session Symposium
Duration 3 hour
Date/Time May 08 2019 14:00 GMT
8:00am PST/11:00am EST
4:00pm GMT/5:00pm CET
Convener Ken Baclawski

Ontology Summit 2019 Symposium Session 2

Agenda

Conference Call Information

  • Date: Wednesday, 08-May-2019
  • Start Time: 8:00am PDT / 11:00pm EDT / 5:00pm CEST / 4:00pm BST / 1500 UTC
  • Expected Call Duration: 3 hours
  • The Video Conference URL is https://zoom.us/j/689971575
    • iPhone one-tap :
      • US: +16699006833,,689971575# or +16465588665,,689971575#
    • Telephone:
      • Dial(for higher quality, dial a number based on your current location): US: +1 669 900 6833 or +1 646 558 8665
      • Meeting ID: 689 971 575
      • International numbers available: https://zoom.us/u/Iuuiouo
  • Chat Room

Attendees


Proceedings

Panel Questions

  • Can computer-based explanations be made as informative and useful as an explanation from a well-informed person?
  • Are there disconnects among researchers, industry and media, or between users and investors with respect to what constitutes an acceptable or successful explanation?
  • Is there general ignorance about what is already possible vs what is well beyond current capabilities?
  • What role do ontologies play?
  • How can one integrate ontological and statistical approaches?
  • What is required for an ontology to support explanation?

See prior day's panel discussion on these topics.

Discussions

[11:09] RaviSharma: I agree with janet, it is too fast!

[11:16] RaviSharma: Mark and Mike - Is it not true that financial regulatory understanding for compliance can be viewed to be unambiguously (almost!) interpretative and driven like logic thus business transaction or contracts have fixed explain-ability as compared to where context, narration (varied) is more variable for that track

[11:37] Mark Underwood: Ravi It's only relatively unambiguous within a specialist domain e.g., an auditor speaking to a regulator. That works in highly constrained contexts, but when speaking with consumers where the regulatory language is not canonical, it doesn't work so well

[11:43] Mark Underwood: Earlier this year I attended a conference by https://edmcouncil.org/ where several live ontology-driven fintech applications were deployed, e.g, one being developed by Thompson Reuters

[11:43] Mark Underwood: EDM is behind the FIBO ontology

[12:23] John Sowa: Context and provenance are far more than "almost necessary". They're absolutely necessary.

[12:26] John Sowa: Common sense should be considered an ontology. In fact, WordNet can be treated as the hierarchy for accessing the Commonsense ontology.

[12:27] John Sowa: The definitions for the Commonsense ontology are given by the WordNet synsets, but they should be supplemented with any or all dictionaries that may be available for all the languages of the world.

[12:28] John Sowa: Of course, no two dictionaries have exactly the same set of word senses and definitions.

[12:28] ToddSchneider: What is the intent of providing this 'cross track correlations'?

[12:29] John Sowa: But "C'est la vie". But pardon my French.

[12:30] ToddSchneider: What might be more interesting is commonality or overlap (or lack thereof) of mechanisms used.

[12:36] RaviSharma: Thus I tend to agree to John's cross track comment, CO is common to many AI solutions?

[12:37] TerryLongstreth: The linguist's term for the different affects adopted between people for different conversations is Register. People switch registers unconciously in the course of the same interaction or between/among different interactions.

[12:38] Gary: In the context of our discussion of people interpreting things, say in the Med field, I think that there is a great deal of varied interpretation going on in the complex task of finding the relations between the rack topics. Unless we had a great deal of time to standardize our ideas thru discussion we should stay away from any simple qualitative view but we can shoe the influences of these on each other.

[12:38] RaviSharma: I also like the comment that CO is upper Ont that will link down

[12:39] RaviSharma: not only to ontologies but to other factors such as NLP, DL, ML, Neural and other models used in explanation based solutions

[12:39] Gary: As an example, my Q o the use of XAI/ML in Finance could be handled by citing some of the Apps we know of and perhaps conclude that they may be less than in the Med area.

[12:40] John Sowa: Commonsense is the foundation for everything that anybody learns from nursery school to a PhD, MD, JD...

[12:41] John Sowa: In fact, that is the foundation for explanation.

[12:41] RaviSharma: The intention and context also have place in these considerations and tracks.

[12:42] John Sowa: Every explanation must begin at the level of the commonsense ontology -- and then drill down as the participants in the dialog get into the details.

[12:45] John Sowa: But those cryptocurencies are evil. The blockchain *must* include a record that can be used as evidence in a court of law.

[12:47] John Sowa: That is why bitcoin is a disaster -- it's the currency of choice for ransom payments, smugglers, thieves, tax fraud, money laundering, and the Mafia.

[12:50] MikeBennett: I've finished a draft of the Finance section - is there a wiki page to put it on? (I see we are working in a Word doc)

[12:56] Mark Underwood: @Mike / all - I need to head to another meeting. Catch up with you later. Thanks for joining

[12:57] MikeBennett: I've popped the Finance Communique draft / synthesis material under our heading in today's agenda, if you refresh. I haven't reviewed so maybe a bit rough.

[12:58] RaviSharma: Cross track presentation by Ravi lead to many new avenues and observations - for example John Sowa saying that Commonsense is an ontology in most AI solutions and that links to detailed ontologies for deeper use (practioners) etc can be used for deeper explanations. Janet's comment on importance of tracks in future AI and XAI directions. Mark, Mike Gary and Todd also had views relating to Finance and also questioned this particular way of depicting Cross Track Matrix and Grading. Thus the purpose of engaging the Co-Champs and participants in Cross Track analysis and future solutions was useful, most likely not many deliverables may be there from X-track discussion for the communique. Related to this is Ram's waves notion i.e. in which aspect of future waves of AI would be impacted in what way by influence of these tracks / domains.

[12:59] RaviSharma: Communique discussion started

[13:00] Gary: Grice elaborated on from his Cooperative Principle. Take the 4 categories (with several submaxims) called Gricean maxims. Speakers and writers should consider the:

1. Maxim of relevance: Utterances should be relevant to a topic

2 Maxim of quality: Make your contribution as informative as is required (for the current purposes of communication). So assertions should be true and you are not saying what you believe to be false or do not say something for which you lack adequate evidence

3. Maxim of quantity: Assertions contains all information that might be useful, and no information that is useless/superfluous.

4. Maxim of Manner: Be perspicuous, that is avoid obscurity & ambiguity. https://en.wikipedia.org/wiki/Cooperative_principle

[13:04] MikeBennett: I wonder if it would make sense to bifurcate the Communique in terms of techniques versus industries - with Finance and Medical as the only industries represented time around, we can introduce it as being indicative of the use of the various techniques in industry. That would also let us de-duplicate a lot of what is in finance that is covered elsewhere in more general terms.

[13:05] John Sowa: In the work on computational linguistics, the researchers found that the *example sentences* in dictionaries were often *more* useful than the definitions.

[13:08] John Sowa: The examples show typical ways of *using* the words. That is the context, which is often more important for understanding than the word definitions.

[13:10] Gary: Material to be added to the CSK section -Make the point that CSK and explanation can complement one another (But a suite of CSK ontologies may be needed as shown by Michael Gruninger's work)

CSK can be particularly valuable for ML/DL efforts:

Commonsense aware models may:

help to create adversarial learning training data

generalize to other novel situations and domains

compensate for limited training data

be amicable to explanation e.g. with intermediate structures.

[13:12] John Sowa: A counterfactual (sentence containing "would be") always shows that there is some implicit law, rule, or principle that determine why something "would be" in that context or any similar context.

[13:13] Gary: The section should make the point that: There remains confusion about what is meant by explanation as well as commonsense concepts & how their quality should be measured. Sometimes a summarization is as good or better than a complex "complete" technical explanation that does not take into account human assumptions.

[13:14] John Sowa: That implicit rule or principle (implied by the "would be") is an important point that an explanation should present or be ready to present in an answer to any follow-up question.

[13:15] BobbinTeegarden: @John what about the 'could be'?

[13:16] TerryLongstreth: @Gary: It's probably too late to mention this but an explanation can only succeed when it's approachable by the intended recipient.

[13:16] John Sowa: Bobbin, 'could be' is a possibility, 'would be' is a necessity. Both versions depend on the same principle.

[13:17] John Sowa: If the principle allows options, any one is a 'could be'. If it's a requirement, it's a 'would be'.

[13:18] BobbinTeegarden: @John but in AI systems, the could be is what draws you forward to iterate towards the possibilities, no?

[13:20] John Sowa: Bobbin, that depends on the context. It's one more reason why follow-up questions and a dialog are necessary.

[13:21] BobbinTeegarden: @John Thank you, totally agree.

[13:21] John Sowa: When I'm thinking about explanations, I always remember the many frustrations I've had in dealing with telephone systems that are driven by a fixed script.

[13:23] John Sowa: Even when the script has options, those options are usually provided by the person who developed the system. If they forgot some feature, they would certainly forget the need to explain how to get around it.

[13:25] BobbinTeegarden: Regarding section 7 first paragraph: if 80s was dominated w knowledge-based systems, it seemed that one of the presentation indicated the next/emerging wave was based on 'ontologies', so isn't that also based on knowledge based systems? Are we in a cycle?

[13:32] BobbinTeegarden: Overlap in tracts: isn't that because each tract was a 'context' in itself, and there's an 'and' set with a bunch or 'or' sets when you combine contexts like that? Are we looking for the common 'and' set?

[13:33] ToddSchneider: Findings should come first.

[13:34] ToddSchneider: Depending on how 'Findings' are presented, they could lead into 'Challenges'.

[13:36] RaviSharma: Sections findings could be Summit 2019 and related Findings

[13:37] RaviSharma: Challenges and opportunities are for all tracks for important future directions

[13:53] BobbinTeegarden: If OWL doesn't work, can we encourage tools vendors to supply some good tools based, say, on CL? Why is that not happening?

[13:56] RaviSharma: Everyone is requested to provide bullet points to Ken for section 8 and especially section 9 based on challenges from speakers and track summaries

[13:57] ToddSchneider: Bobbin, many people have problems with OWL. Use of Common Logic is leap too far, unfortunately. More importantly, currently there's no commercial support for Common Logic. Another aspect of the more general problem is the expectation that the 'technology' should be 'easily' understood and used.

[13:57] RaviSharma: Medical inputs from Ram and David if available

[13:57] MikeBennett: OWL 'works' - it is a decent application language. What it doesn't do is reflect all the kinds of meanings of things that we care about for e.g.explanations.

[13:58] RaviSharma: Financial track inputs to be provided by Mark and Mike

[13:58] MikeBennett: I've completd the Finance piece as noted earlier.

[13:59] MikeBennett: (Finance piece includes need for upper ontology)

[13:59] John Sowa: Re OWL and CL: The DOL project has a very nice diagram of the many different logics are related.

[14:00] TerryLongstreth: I guess no one could hear me.

[14:00] RaviSharma: Everyone to add to communique or subtract (suggest) to Ken in next 1-2 weeks.

[14:00] John Sowa: The reason why OWL hasn't gone further is that the W3C is dominated by people who's learning stopped in 2006.

[14:00] TerryLongstreth: I will review the communique but not today.

[14:01] RaviSharma: Post Martem session will hopefully converge to publishable communique.

[14:01] RaviSharma: Also issue of Overview and tracks.

[14:01] AlexShkotin: Thank you All! Great Symposium!

[14:02] RaviSharma: I meant full publication (such as Washington Acad of Sciences)

[14:04] RaviSharma: Again we hope to see usage stat on our pages and downloads

[14:06] ToddSchneider: Have to go. I look forward to this communique.

[14:06] RaviSharma: Peter Yim was good at marketing.

[14:07] RaviSharma: people are joining our forum everyday

[14:08] janet singer: Need to switch calls - thanks Ken

[14:08] RaviSharma: 1000 past and present participants and active members

[14:09] RaviSharma: Ken and all many thanks.

Resources

Previous Meetings

... further results

Next Meetings