Expert System Reasoning in Soar
Expert System Reasoning in Soar
Soar has been applied to a computer configuration task that was also a
problem for the expert system R1. R1-Soar does a large part of the
configuration task, but in a different way from R1. R1 has what is
called a shallow structure. It is given rules about what to do
in a particular situation but little ability to reason about the task.
R1-Soar, on the other hand, has deep structure. It has a set
of basic problem spaces in which
to carry out basic tasks and knowledge about the underlying
constraints -- this formed the deep structure. Since the system
included no heuristic knowledge, R1-Soar was designed to search in the
problem spaces for a particular solution rather than move directly to
a solution as in R1. However, as Soar learned in this domain (with within-trial and across-trial transfer), it
exhibited a significant increase in performance, approximating that of
R1 for the part of R1 that R1-Soar was designed to cover. This
application of Soar to an expert-systems task demonstrated three
things about Soar empirically that were predicted by the structure of
its architecture:
Return to the top of this architecture.
Go to a discussion of this capability
for multiple architectures.
Current Location: Soar-Capabilities-Expert System Reasoning