CMSC 477/677 - Spring 2005
Discussion Questions for Class #4, February 8
Reading: Soar website (overview); Lehman et al., "A Gentle Introduction
to Soar."
Reading suggestions: This is a fairly long paper. Some parts
are more important than others. Start by skimming the whole paper (this
should take no more than 2-5 minutes), and deciding which sections seem important.
My comments on importance: Introductions are always important.
For this class, sections 2 and 3, which motivate the "big ideas," are
critical since they put the work in context. Section 4 introduces the
key notion of problem spaces, and is worth spending some time understanding,
if it isn't clear on first reading. Sections 5 through 8 give the details,
and lots of examples. You should dip into these sections, sampling
to get a good sense of what's going on, but without getting hung up on undertanding
any one particular example or idea. Sections 9, 10, and 11 are very
important because they analyze the ideas presented in the paper, and will
help you to understand the other architectures we will
study later in the semester.
Note that I'm not saying "technical details don't matter," but it's certainly
more important to understand why Soar works the way it does, and what
the designers were trying to accomplish, than to understand the arcane details
of how contexts are used to trigger associations in Soar's elaboration phase.
Soar Concepts
- How are the following concepts used in the Soar architecture?
- Principle of rationality
- Goals, problem spaces, states, and operators
- Working memory
- Elaboration, preferences, and decision
- Impasses
- Chunking
- Does chunking seem like a good universal model for learning? Why
or why not? If it is, why are people still doing research in machine
learning?
Soar as an Architecture
- Laird et al. [Soar: An Architecture for General Intelligence]
state that "The adoption of the problem space as the fundamental organization
for all goal-oriented symbolic activity (called the Problem Space
Hypothesis [and due to Newell]) is a principal feature of Soar." How
does Soar's adoption of this hypothesis differ from other AI methods?
- Do you think the Problem Space Hypothesis is a reasonable model of
how humans think?
- Is "generic search" a reasonable summation of Soar's model of reasoning?
This model allows it to be used to easily implement "weak" search methods.
Are there tasks for which this model is inadequate?
- Here are two opposite views of the "generic" nature of Soar as an
architecture for intelligence. Be prepared to argue for or against either
of these viewpoints, and to offer other points of view:
- The "generic" universal subgoaling of Soar makes it an exceptionally
powerful architecture for reasoning.
- The "generic" universal subgoaling of Soar makes it such a general
model that it could cover any sort of behavior -- and therefore, it really
says nothing about reasoning. In fact, all of the so-called "power" of
Soar is embedded in the domain knowledge (production rules and heuristics).