Problem Solving in Soar
Problem Solving in Soar
Soar's chunking mechanism is based on
a deductive procedure (the dependency analysis algorithm) but
this does not necessarily imply the ability to reason deductively.
However, Soar is a Turing-equivalent computational device and, as
such, combined with its attribute-value
representation, may be used to reason deductively in a variety of
task domains. However, such deduction must be programmed explicitly;
there is no theorem-proving system built into the architecture.
Additionally, although Soar systems may be designed to explicitly
reason deductively, the uniformity
of the representation is a weakness when compared to knowledge
level logic systems.
For a long time, many researchers contended that Soar could not reason
inductively since its mechanism for
learning, chunking, was experience-based. In general, it is
difficult to use Soar for inductive reasoning. However, Rosenbloom
and Aasman reported using Soar to perform an inductive technique
(version spaces). Thus, although inductive reasoning may be awkward
or difficult in Soar (in comparison to other systems), it is possible.
One issue related to this awkwardness
is the contention that human learning in based on induction.
Return to the top of this architecture.
Go to a discussion of this capability
for multiple architectures.
Current Location: Soar-Capabilities-Problem Solving