November 3, 2004
Center of Excellence Proposal
à
White Paper on Incident Information Orb
Architecture (IIOA) à
Adi Structural Ontology Part I
à
ARDA rejection notice
(Friday, October 29, 2004) à
Possibility from
Knowledge Processors
BCNGroup Founding Committee October 27, 2004
Note from Dr. Richard Ballard
Founder, Knowledge Foundations Inc
Paul and BCNGroup colleagues,
Mark 1 and 2 have 20 year histories and can cite over 50 nationally important projects with over 30 NASA, DoD, and commercial industries so that often makes government agencies more receptive. But my experience with AQUAINT was that the ARDA mind set is most particularly closed. Their Fort Huachuca intermediators recommended HARPA (Homeland security's ARPA) as a better, less inaccessible funding agency.
Still NSA and intelligence agencies are perhaps the least informed and use their secrecy to hide it and the identity of their reviewers.
Give me a scientist, astronaut, or uniformed officer and you will get more forward looking honesty every time. At the same time the NSA types have been the most ripped around by AI and Linguistic Webs posing as semantics.
The old hands are bitter and eager to classify everything as pure bullshit without reading any further.
To them they have only to know what storage system you use and then you must be like some related con they once believed and supported for years before. They can cite dozens. Some stick with it simply to humiliate the new pretenders -- a psychology of old men with high hopes gone bad.
Secrecy sounds really sexy and well funded, but it is really bad science and worse business. All are empty rooms with no windows and one or two government faces. Successful or not all your efforts are guaranteed to be compartmentalized and made invisible.
We hold general secret and top secret clearances (now all lapsed) when necessary, but religiously avoid maintaining secure facilities or holding and classified documents or meeting notes. A secret demo has the same financial value and future worth as write-only memory. We model only from declassified sources. We prefer charts and graphs with all the numbers cut out. Our industry partners with secure facilities fill in the names and numbers. Typically we use commercial simulation games -- flight simulators, respected tactical games like Harpoon, etc. for progress demo deliverables and to produce plausibly deniable budgets, scenarios, and tool demos. That means we and they can both demo our tools to unclassified audiences as representative of their efforts and interests.
Most agencies have declassified sample data sets. Find out which your client uses. The AQUAINT Proposers Information Pamphlet (PIP) at Ft Huachuca ARDA site BAA 03-06-FH we responded to listed the AQUANT Newpaper/Newswire Collection (including New York Times newswire and AP Wordstream English newswire June 1998-Sept 2000); and structured data by Center of Non-Proliferation Studies (CNS) and membership materials available through the Linguisitc Data Consortium. You should be dropping those names or at least comparing your demo sources to them.
I would be more interested in gaining some insight or intelligence on Ron Brachman's Developing Cognitive Systems effort at DARPA (IPTO). It looks from afar as more representative of my interests, but having been around for a while it may have narrowed into a closed circle of self-limiting expectation. No way to ask or investigate casually from this distance.
Washington is more an important sales center than a technology hotbed.
Our own leading edge work is on rational path models and formalizing these within the ontology of mediating structures and our budding sourcebook of models. In this context the best criminal or intel applications deal with matching criminal (hostile) action profiles to rational "how" path bundles, then using future constraint browser functionality to build criminal act self-awareness into autonomous agent or news report scanning and storage. This kind of capability might be plausibly fielded in late 2005 -- a push to start selling it now makes sense.
This prototype construction has a profile modeling component, a report scanning and modeling component, Mark 3 template grammar defintion, full path structure classifications, n-ary feature set extractions, and early interactive experiments in constraint browser design and automation. That would be as close to "HAL 2001" trying to figure out "Dave's hostile intentions" as any real machine has ever gotten before. The Grammar notes and Class lectures notes could get you all within reach of these ideas.
The main linguistic components within Mark 3 are centered on the Concept finder and its inverted word-concept list n-aries and on a special grammar variant of the constraint browser with a superset of WordNet type functionalities. This was first designed in our 1992 project called BookReader for Media-Share, a Joestens Learning Company spinoff in San Diego.
Our more immediate need is for an English Language custom wordlist generators -- accepting user English language concept titles. It would automatically make custom dictionary entries and encode these new word-concepts with machine generated part-of-speech (POS) model-instance codes. This assumes the use of a pre-existing English Language Spelling Checker and a POS tagger. In general Spelling Checker and POS taggers are available academically or commercially for most world languages. Our interest is in a sponsored experimental project that would test the suitability of using such generative custom list components until standardized POS word encodings were commercially available and licensable in every country offering a suitable knowledge tool market. We will need the custom English Language list component by Summer 2005.
Companies interested in developing such lists internationally are to be potential strategic partners for eventual international tool rollouts.
That is my best profile on where we are and the kinds of collaborative efforts that match our early needs.
Dick
See related:
New bead thread on Generative Methodology for production of Structured Ontology