I am representing a group of scientists who will be making the case, in our SAIC/OntologyStream submission in response to the

Research Note: 16.1

Question to DARPA REAL BAA management

August 8, 2003

Paper on the need for human distance learning to support machine aided temporal reasoning

I am representing a group of scientists who will be making the case, in an upcoming SAIC/OntologyStream submission in response to the REAL BAA, that temporal reasoning is beyond the formalisms that are currently applied to knowledge representation and machine inferencing problems.

The argument comes from Hameroff, Pribram and Penrose; and others, who suggest that the standard approaches, using mathematics and/or algorithms, is in fact attributing the description "complex" to something thing that an algorithm is able to do BY ITSELF. We feel that this attribution is a terminological error that has lead to continuing funding interests in the extreme forms of strong AI. Whereas some AI type research may still be justified, our group has established a well-defined alternative. Last year the SAIC/OntologyStream proposal met NIMA’s criteria for funding, but was not funded.

http://www.ontologystream.com/private/NIMAtechnical.htm

We define "complex" as something that precludes an algorithmic decision. Natural complexity allows multiple next states to be co-existing in a fashion that cannot be un-entangled by precise crisp logic - such as first order logics. The argument is made that natural complexity must always present a halting condition to a serial computer. Providing some types of stochastic estimates of the likely next steps involves a measurement problem that is not addressed properly by the standard approaches.

We suggest making an accommodation by human in the loop reification adjustments at exactly those places where sense-making limitations on algorithms become critical. This human accommodation of the limitation of algorithmic processes can only be done if the domain expert understands the algorithms behaviorally and technically. Therefore, an educational process is essential to the development of real world reasoning by humans that involve the use of complicated computer processes, such as Latent Semantic Indexing, reinforcement based categorizers and deductive inferencing.

In principle, established by logical argument and natural science fundamentals, there can be no all-purpose temporal reasoning algorithm about natural processes. To assume otherwise is to make the mistake that category theoretician Robert Rosen referred to as the category error - mistaking the complexity of any natural system with the non-complexity of any algorithm. One should be able to recognize the mistake based on failures of AI projects to meet reasonable expectations

This does not mean, to our group of scientists, that a new form of effective, practical, automated assistance to human reasoners is not possible. But it does mean that human cognitive acuity must be better understood by computer science and that ontology services and algorithmic processes such as Latent Semantic Indexing or NLP taggers must conform to the limitations that formal system always must, in principle, have. This requirement for natural science means that social science and cognitive science must have a more dominate role in the design and development of information systems.

Our approach is grounded in an extensive literature from perceptional physics, ecological psychology, mathematical biology, mathematics, quantum neuroscience, topological logics and both the cognitive and social sciences. An action perception cycle is established that must find one half of the cycle within the experience of a human, and the other half of the cycle involved in differential and formative ontologies. The differential and formative ontologies relay on a localization of (type:value) pairs where the theory of type and the possible instantiations of values is developed as human work product. These localizations are then organized situationally

Is it possible to have some discussions with the REAL program management on issues relative to our approach?

***********************************

Dr. Paul S. Prueitt

Research Professor

The George Washington University

Founder, (1992) BCNGroup.org

Founder, (2000) OntologyStream Inc

Knowledge Scientist

Cell: 703-981-2676

paul@ontologystream.com

***********************************