Previous: Facilitators and Mediators Up: KQML and Intelligent Information Integration

The role of KQML

As a communication language for intelligent information agents, KQML draws on work in both distributed systems and distributed AI and offers a level of abstraction that should be useful to both.

With respect to distributed software systems in general, KQML provides an abstraction of a process as an information agent as a knowledge-based system (KBS). The KBS model easily subsumes a broad range of commonly used information agent models in use today, including database management systems, hypertext systems, server-oriented software (e.g. finger demons, mail servers, HTML servers, etc), simulations, etc. Such systems can usually be modeled as having two virtual knowledge bases -- one representing the agent's information store (i.e., beliefs) and the other representing its intentions (i.e., goals). We hope that future standards for interchange and interoperability languages and protocols will be based on this very powerful and rich model. This will avoid the built-in limitations of more constrained models (e.g., that of a simple remote procedure call or relational database query) and also make it easier to integrate truly intelligent agents with simpler and more mundane information clients and servers. Current KQML implementations have used standard communication and messaging protocols as a transport layer, including TCP/IP, email, Linda, HTTP, and CORBA. As standards in this area evolve and new standards are introduced, we expect that KQML implementations will use them.

The contribution that KQML makes to Distributed AI research is to offer a standard language and protocol that intelligent agents can use to communicate among themselves as well as with other information servers and clients. The independence of the communication and content languages affords a flexibility which is quite useful. In designing KQML, our goal is to build in the primitives necessary to support all of the interesting agent architectures currently in use. If we have been successful, then KQML should serve to be a good tool for DAI research, and, if used widely, should enable greater research collaboration among DAI researchers.