==========================================


     University of Hertfordshire

 Computer Science Research Colloquium

         6 February 2002
          Room LC 108
          3 pm - 4 pm+


====

Self-Reference and System Engineering with Wrappings
Christopher Landauer, Kirstie L. Bellman
Aerospace Integration Science Center
cal@aero.org, bellman@aero.org

Outline

System Engineering not Software Engineering
Problem Posing Programming Paradigm
	Reuse without Modification
Wrappings for Integration: Systems, Components, Languages, Logics
	Self-Reference and Computational Reflection
		Generic Programming and Partial Evaluation
		Autonomous Systems
Other Research Directions
	Symbol Systems and Computational Semiotics
	Your Worst Nightmare of Self-Modifying Code

Long Abstract

This talk is about some new methods in Software System Engineering, with an
emphasis on integration infrastructure and what it can do.  We start by
describing the Problem Posing Programming Paradigm, which is a declarative
interpretation that can be applied to any programming language, and, indeed,
to any formal or even merely systematic notation that can be parsed into
discrete syntactic structures, so in particular, the notation need not be
linearly written (though the most common examples are).  The main idea of the
Problem Posing interpretation is to separate information service requests
(called problems) from the information service providers (called resources),
so that they may be connected in much more interesting and flexible ways than
mere identification of equal names.  Its most immediate application is the
notion of Reuse without Modification: if there is enough code in a particular
language and compilation context, it becomes useful to change the compiler
instead of the code.

Then we describe Wrappings, an approach to the integration of Constructed
Complex Systems that is based on the twin notions of meta-knowledge and its
automatic interpretation.  The meta-knowledge is explicit, machine
interpretable information about the uses of computational resources, collected
into Wrapping Knowledge Bases: in which contexts the resources are
appropriate, and whether, when, and how to apply them to problems in context.
This use of Knowledge-Based Polymorphism to connect problems to resources
leads to an extremely flexible system infrastructure.

The processes that use the meta-knowledge are called Problem Managers.  They
perform the Intelligent User Support functions: Select, Assemble, Integrate,
Adapt, and Explain.  The Problem Managers are also computational resources,
and they are also Wrapped as any other resource is.  The system is therefore
computationally reflective: it has a complete model of its own behavior, down
to some level of detail.  Since every part of the system is selectable, there
are no privileged resources in the system at all.

This kind of self-reference is enormously useful in areas as divergent as
partial evaluation, in which a program is used to simplify the code of another
program as it applies to a partially specified context and problem; and
autonomous computing systems, in which it is necessary for the system to be
quite knowledgeable about its own internal and external capabilities, so it
can use that information to reason about its own decisions and actions.

Short Abstract

This talk is about some new methods in Software System Engineering, with an
emphasis on integration infrastructure.  We describe Wrappings, an approach to
the integration of Constructed Complex Systems that is based on the twin
notions of meta-knowledge and its automatic interpretation.  The Wrappings are
the meta-knowledge: explicit, machine interpretable information about the uses
of every computational resource: whether, when, and how it applies to
particular (classes of) problems in particular contexts.

The processes that use the meta-knowledge to connect problems in context to
applicable resources are called Problem Managers.  They are also computational
resources, and they are also Wrapped as any other resource is.  The system is
therefore computationally reflective: it has a complete model of its own
behavior, down to some level of detail.  Since every part of the system is
selectable, there are no privileged resources in the system at all.

====