To me, expectations on IA's (intelligent agents)
for OFW seemed dangerously over-ambitious and naive from the view of practical
software R&D:
- Scale and time frame: nothing remotely similar has yet
been done in the IA world. It will set new international standards
(hopefully not for folly).
- Host Network: IAs deploy on a fragile, tree-like (non-web)
network based on chain-of-command, line-of-site, and hardware-encryption.
- Client-side IAs: 100-1000 will likely run on the person
of an OFW soldier. And the set must adjust - rapidly - to role or mission
changes.
- Server-side IAs: they will "fuse" data - somehow
- into shared situational awareness more effectively than today's human-based
Intel services.
- Embedded Training: this and pre-battle "rehearsals" imply
that global "what if" powers exist for each IA in the hierarchry.
My pessimism declined only after I started to imagine a distributed web-based
"factory" at which IAs might be bulk-developed to meet OFW's specific needs.
It would comprise a secure web site (or ten) at which tools, standards, and
training existed to aid and support all future IA developers.These key features and tactics
seem central to the factory's successful operation:
- Simulate the IA network from the view of each node.
This enables IA unit testing under all sorts of bizarre, emulated
comm-networks. Simulators have aided other big US software R&D efforts
since at least the pre-Apollo space missions. This one would let IAs
function soon under a digital facsimile of OFW's future host network (see
"B", above).
- Create an OFW-specific "IA Developers Tool Kit".
The Factory would feed data to each IA in part by reacting to its outputs.
It independently hosts each IA's developing rule-base and sets of facts, standard
inference engines to run them, and other shared development tools. These common IA
resources would much simplify meeting goals under "A" and "E".
- Declare OFW-standard IA ontologies. IAs interact
best - with each other, people, and their host network - when they all use one common
language. By centrally selecting the core ontologies they use,
OFW leadership can limit confusion, cut costs of training and IA integration, and
lower the complexity barriers for goals under "C" and "D".
- Share the factory with FCS and the Intel world.
The vital need for OFW-FCS integration was often stated. What that means in
practice remains to be seen, but I expect on-going debates among
800-lb gorillas over IA standards.
A well-run factory could become the arena in which competing standards
might co-exist, or might battle objectively versus politically.
As outlined on our Company Profile Abstract, Lexikos Corporation has created
software which can be part of an IA Developers Tool Kit and aid
the IA Factory's creation. Specifically, using J2EE as a COTS platform
also ideal for hosting the OFW-standard simulator and inference engine(s), we can
add an idealized English case-frame grammar as our own core ontology, and an
"avatar" paradigm (based on our interactive English parser) to aid (1) the bulk
entry of IA-compatible metadata;
(2) voice-based UI channels with OFW personnel; (3) language translations; (4)
embedded training tools.
I was pleased to get positive feedback on such plans from several influential
sources, including Lockheed (who deemed our 1-hour Saturday chat "unusally
engaging"). After such encouragement, I now plan to initiate dialogs on specific
proposals with potential LSIs, government labs, and others who might share a
common interest in creating a prototype IA factory, and/or in using the particular
software technologies above to help advance specific FCS/OFW goals.
If you are among them, please contact me.
- Dan Corwin, Chairman & Software Architect
|
