CMSC 491M/691M - Spring 2003
Discussion Questions for Class #4, February 5
Reading: Wooldridge, "Intelligent Agents..."
I promise we will not discuss the question "What is an agent?" Been
there, done that...
Intentionality and belief
What is the intentional stance? What are pro-attitudes?
Why is belief reasoning not truth functional? What is referential opacity?
What are modal operators? What are possible worlds semantics? Does it seem
practical to you to implement an agent using possible-worlds modal logic?
(Reading guideline: You can skim over section 2.3, Alternatives to
the Possible Worlds Model)
Persistent goals and intentions are useful for reasoning about agents'
plans in dynamic environments, in contrast to traditional planning, where
the planning process is essentially static. What's the difference
between a persistent goal in Cohen and Levesque's intention theory
and a plan old goal in planning? What's the difference between a
persistent goal and an intention?
What's speech act theory?
Why do you think agent communication languages might be useful? What do
you think the limitations of an ACL such as KQML might be?
Wooldridge distinguishes between deliberative architectures (based on planning
and theorem proving), reactive architectures (based on "non-symbolic" reasoning
methods), and hybrid architectures (which have a bit of each). The breakdown
I used in the class syllabus includes cognitive architectures, logical
architectures, and reactive architectures. Any guesses about how these
categorizations might relate to each other?
Agent languages (high-level languages and tools for building agent-based
applications) are becoming popular in the agent R&D community.
We're not really covering these languages in this class. Do you wish we
were? Why or why not? (Project hint: If you want to do an implementation-oriented
project, the survey of agent languages here, and the paper I mentioned
on agent toolkits, might be a good place to start.)
Another topic we're not coming is "believable agents" or emotional agents,
which are quite in vogue (especially among those who are into computer
gaming and want realistic artificial agents in their gaming environments).
I might try to squeeze in a paper or two on this topic. What do you think?