CMSC 477/677 - Spring 2007

Discussion Questions for Class #16, March 29

Reading: Levesque, Cohen, and Nunes, "On Acting Together"; Grosz, Hunsberger, and Kraus, "Planning and Acting Together."

Levesque, Cohen, and Nunes

• The scenario described on page 95 involves agents who initially are in a state of mutual belief, but lose synchrony when one agent comes to believe something that the other agent is not aware of. What would happen if the following assumptions made by Levesque et al. (near the end of page 95) were violated? Are the assumptions reasonable for real-world agent systems?
  • "...each agent has perfect introspection about both his beliefs and goals..."
  • "....beliefs are consistent..."
  • "...goals are consistent with each other and with what is believed."
• What definition does this paper use for commitment to a goal? For intention?
• What's wrong with the first definition of joint commitment (given on page 96)? What's the difference between a weak goal and a joint persistent goal?
• What's the difference between a joint persistent goal and a joint commitment?
• Levesque et al. are particularly interested in linguistic interactions in the context of joint activity, "seen as attempts to establish and maintain... mutual beliefs." In the joint commitment framework, what clues might an agent have that it needed to communicate with another agent? How would the shared knowledge of joint commitments make their communication more efficient?

Grosz, Hunsberger, and Kraus

• Grosz et al. give three key aspects of SharedPlans that make them useful for collaborative agents. Why are each of these aspects important? Give an example of situations where lacking each of these aspects would cause collaborative failures or inefficiencies.
  • "provision for agents to interleave planning and acting"
  • "inclusion of commitments that can lead agents to behave helpfully"
  • "constraints that prohibit adoption of conflicting intentions"
• What is the difference between intention to and intention that in the SharedPlans framework? Why are both of these forms of intention necessary for collaborative planning?
• What are the differences and similarities between the SharedPlans framework and Levesque et al.'s model of joint intentions? Do you think either framework is:
  • ...a more accurate model of how people act jointly? Why?
  • ...a better normative model for how agents should act jointly?
  • ...a more useful model for building practical systems?