Cognitive Information Processing Technology DARPA BAA 02-21

I am attaching the text file as well, if you don't want to plow through this in the body of a text message. I haven't gone into detailed explanations, and I have asserted that we have work that isn't really done yet, or even started except in discussions, but funding would take care of that aspect, and this is just a broad overview anyway. However, in much the same way that the proposal Rob and I worked on sounded to me as if we were really the only group that could conceivably pull off the work topic as described, this one also seems to fit us in some particulars in ways that I honestly can't imagine any other group coming close to addressing. However, as we go through the point by point I want to encourage us all to cite specific ideas and achievable projects or experiments which can test our assumptions as we go. Basically, our core, primary base, is all that makes this work even conceivable, although I'm sure DARPA isn't aware of that, even when they ask for innovations which are actually necessary to doing this work at all.

SOL BAA 02-21
POC: Dr. Ronald J. Brachman and Mr. Zachary J. Lemnios, DARPA/IPTO

PROGRAM OBJECTIVES AND DESCRIPTION:

Can we propose innovative R&D on cognitive systems and technologies that outperform current approaches?

Yes, we can. The semiotic engine and the work of the Cognition and Context Subcommittee can be the basis for that.

What is called "reasoning" in the FBO is what we term "heuristics" based on self- and environment-monitoring and tasking priorities.

Can we build "learning" systems?

Yes we can. Rob has. However we need, at least among ourselves, to keep the distinction in mind that this is largely observed-behavior comparison-based adaptation, not intuitive conclusion-forming.

Can we build systems capable of "natural" interaction with humans?

As well or better than most, but this is a question to which the answer must always be understood as "not possible." A program is not a natural human and can only emulate humans as naturally as possible, even if we assign human abilities and attributes to it in order to emulate enough "humanness" for biological humans to be comfortable in interactions with these programs.

We are already working on cross-disciplinary approaches for stand-alone systems, mobile platforms and deeply embedded, integrated components, although the last is a stretch.

Computational Perception: We have yet to propose a model for perception, but since the basis for our model will be human perception, we are prepared in that sense to produce a model that will more closely resemble the human apparatus for interfacing with the external world and the inner workings of the human mind for correctly interpreting sensory data in terms of objects, events and situations.

We are particularly well-equipped to configure Representation and Reasoning based on mathematically well-founded knowledge representation formalisms and modeling reasoning processes. However, we do need to put these into concrete form. I believe Rob, Sylvia and perhaps Amir can be counted on to develop these materials.

We also need to develop introspection to a greater degree than has been previously attempted to not only monitor the sytem's performance and capabilities, but to also select heuristics to anticipate problems such as load balancing when asked to perform multiple tasks which are also composed of multiple parts, phases, components, etc.

Communications and Interaction Technology: This is our meat and potatoes, and we already have a lot of substantial work proceeding, especially in adding the emotional components that have thus far gone absent in human and machine communication. Even though this work is barely in its infancy, that is a cut above the largely nonexistence that has preceeded us in this work. Modeling responsiveness to emotional concerns, especially in cultural terms will undoubtedly yield better results than have previously been possible.

Cognitive Architectures and Integrated Cognitive Agents: This is where we can truly excel. However, we do need someone who can translate our foundational work into a functional system architecture that supports the semiotic model of communication and cognition--accommodating the influences of group membership and shared context quickly and accurately. (This is something I doubt DARPA appreciates much if at all within the computing environment, and if we can persuade them that this is the key to building more accurate, and predictive models of perception, hence behavior, we might be able to ensure that we win this award.)

Robust Software and Hardware: I think what they want here is mobile platform integration of robust, secure cognitive systems that can be used by teams in the field (This is DARPA after all).

Cognitive Teams: see above.

Underlying Foundations: Let's use Wolfram's seminal ideas on pattern recognition to facilitate cognitive systems as the scientific and mathematical foundation for our work. There is a problem with this. We need someone who has the background to understand this work and the time and energy to follow through on it if we get the award or a parallel award.

Among the many other direct interests cited which we fill the need for better than most could are:

* innovative ideas on effective (our word is accurate) machine information systems and network architectures (our concept is distributed, object-oriented, flexibly robust semiotic client and flexibly robust semiotic server architecturess--adaptable, self-load-balancing systems);

* cognitive computing based on biological systems; and

* scientific and technical advances in the understanding and implementation of trust relationships among autonomous cognitive systems (I can perhaps snare IBM into this one on the basis of their efforts moving toward 'autonomic omputing.')

We definitely want to recruit larger, more robust corporate partners, for which I will suggest IBM for its PartnerWorld for Developers and PartnerWorld Solution Centers (San Mateo with the Western US Grid) in the Autonomic Computing initiative (and for the WebSphere/DB2 frrebies and perhaps Rational suite, too) and Sun, through Monica Martin if we can recruit her, for the hardware advances even if we have to live with Solaris for that. I might suggest Oracle, which could even work with IBM despite the competing databases, which would also facilitate using Linux, which would make Sylvia happy. Obviously, we would have to proceed cautiously with these recruitments, and I would also suggest we contact Phil Rossomando again, since Unisys might be more approachable than IBM, but, this would be of more interest to his supervisors than HumanML was previously.

huml message