Published in Proceedings of the 2002 ONR Decision-Support Workshop Series: Transformation..., September 18, 2002, pages 209-221.
Over the past several years there has been an increasing recognition of the shortcomings of message-passing data-processing systems that compute data without understanding, and the vastly superior potential capabilities of information-centric systems that incorporate an internal information model with sufficient context to support a useful level of automatic reasoning.
The key difference between a data-processing and an information-centric environment is the ability to embed in the information-centric software some understanding of the information being processed. The term information-centric refers to the representation of information in the computer, not to the way it is actually stored in a digital machine. This notion of understanding can be achieved in software through the representational medium of an ontological framework of objects with characteristics and interrelationships (i.e., an internal information model). How these objects, characteristics and relationships are actually stored at the lowest level of bits in the computer is immaterial to the ability of the computer to undertake reasoning tasks. The conversion of these bits into data and the transformation of data into information, knowledge and context takes place at higher levels, and is ultimately made possible by the skillful construction of a network of richly described objects and their relationships that represent those physical and conceptual aspects of the real world that the computer is required to reason about.
In a distributed environment such information-centric systems interoperate by exchanging ontology-based information instead of data expressed in standardized formats. The use of ontologies is designed to provide a context that enhances the ability of the software to reason about information received from outside sources. In the past, approaches to inter-system communication have relied on agreements to use pre-defined formats for data representation. Each participant in the communication then implemented translation from the communication format to its own internal data or information model. While relatively simple to construct, this approach led to distributed systems that are brittle, static, and resistant to change.
It is the premise of the TEGRID (Taming the Electric Grid) proof-of-concept demonstration that, for large scale ontology-based systems to be practical, we must allow for dynamic ontology definitions instead of static, pre-defined standards. The need for ontology models that can change after deployment can be most clearly seen when we consider providing information on the World Wide Web as a set of web services augmented with ontologies. In that case, we need to allow client programs to discover the ontologies of services at run-time, enabling opportunistic access to remote information. As clients incorporate new ontologies into their own internal information models, the clients build context that enables them to reason on the information they receive from other systems. The flexible information model of such systems allows them to evolve over time as new information needs and new information sources are found.