CMSC 491M/691M - Spring 2003
Discussion Questions for Class #10, February 26
Reading: Weiss Chap. 8, Levesque et al., Levesque and Reiter
Formal Methods
-
The chapter makes a distinction between "formal methods" and "ad hoc,
quick-and-dirty approaches to system construction." Do you think
there is a middle ground? What are the pluses and minuses of
using formal methods in agent design?
-
FYI, from dictionary.com: "Deontic: Of, relating to, or concerning
duties or obligations. (From the Greek deon, deont-, obligation,
necessity).
-
Devil's advocate: It's hard enough
to build a single intelligent agent; how can we possibly hope to
coordinate them??
-
The section on communication talks about languages for inter-agent
communication, but what about communication breakdowns? How would
they influence, for example, the scheme for coordination given in 8.5?
-
What's an ontology?? (This is one of those "What's an agent?"
questions, by the way...)
GOLOG
-
p. 2: "The presentation throughout is informal in nature; in a
companion paper [14], we explore the more formal aspects of the work."
Aren't you glad I didn't assign [14]? :-)
-
Situation Calculus 101:
- 471/671 pop quiz: What are the limitations of the situation
calculus? (Hint: In section 5.3, the discussion of progressing
the databases addresses part of the key problem.)
- What's a fluent?
- How are action preconditions and effects modeled? What's the
meaning of the Poss operator?
- What's the frame problem? Are frame axioms an adequate
solution? Is the "simple solution" they propose a good one? Why
or why not?
-
Complex Actions 101: In class, we'll walk through the specifications
for items 1-6 in section 3.
-
Semi-rhetorical question: Gee, it seems like GOLOG is just another
programming language. So why not just write a program?
-
p. 10: "By opting to define programs as macros, we obtain a much
simpler theory than if we were to reify these actions." Bonus credit
to anybody who can clearly articulate what this means and why it's
true.
-
p. 20, bullet 2: "Nondeterminism can be used to infer the missing
details [in] sketchy plans." \ This is great! The planning
problem is solved!\ True or false?
-
What are GOLOG's most significant strengths and weaknesses? (Hint:
see p. 21 for some of the limitations.)
- Can you give some examples of domains or environments that
would be particularly well or ill suited for a GOLOG application?
- What aspects of the elevator controller domain make it a good GOLOG
domain? Do you think it would be possible
to write a more sophisticated/"smart" controller (e.g., for a large
building with many elevators, where utilization must be maximized and
energy usage and wait time must be minimized)?
- How does GOLOG compare to Soar, ACT-R, Prodigy, and APEX in
terms of its cognitive plausibility, usability for application
design, verifiability, and expressiveness?