Architectures With Modular Construction
Computer science, particularly software engineering, and even the field of
artificial intelligence itself, has fostered a modular approach, both as we
understand the organization of the field and as we attempt to design
intelligent agents. While is it certainly not a requirement for an
intelligent agent to have a modular design, it is a principle which may lend
itself well to agent organization. Certainly, many principles of software
engineering are directly applicable, but it may be that modularity enhances
abilities that are of concern to artificial intelligence in particular. For
example, we hope that an intelligent agent would be able to "maintain" its own
system; modularity would simplify this task. If learning is to be an
important capability of an agent, modularity may serve to create more general
heuristics which the agent could modify more easily. And psychology has shown
us that the brain itself, at least in some manner, is organized in a modular
fashion.
Of course, this sort of horizontal structure is not the only option; several
agents have been constructed on a more hierarchical organizing principle,
wherein the agent is able to act and react on several time scales.
The following is a list of agents built on a modular organizing principle:
Reference: Minton, S., "On Modularity in Integrated Architectures,"
SIGART Bulletin 2, 1991.
Click here to see a discussion on the
dichotomy between modular and hierarchical design.
Back to List of Other Properties