Motivation-Based Selection of Negotiation Opponents
Steve Munroe and Michael LuckIAM Group
University of Southampton
Presentation Outline
The problem: dynamic domains and negotiation
Motivated agents for dynamic domains
Negotiation goals Selecting opponents
Autonomous Negotiation Three phases of negotiation
Dealing with dynamics and uncertainty
Strategies / Tactics / Protocols
Agent driven
Pre-negotiation:
Where do the issues come from?
How are reservation values determined?
Who best to negotiate with?
A Motivated Agent Architecture Motivational Cues
Intensity
Mitigation
The Warehouse Domain Controller agent must negotiate with
delivery agent about moving boxes around the warehouse
Negotiation Goals Dynamic reconfiguration of issues to meet
current demands Fixed attributes Potential attribute
Negotiation Goals - 2 Two types of potential attributes
Non resource-based Resource-based
Non Resource-Based Attributes: constructing preferences
What is the structure of the preference? Determined by assessing each possibility in
terms of motivational worth
Non Resource-Based Attributes: attribute classification rules Fixed if the preferences of the agent contains
at most one value that has positive motivational worth and all the rest have negative motivational worth.
Negotiable if the preferences of the agent contains more than one value that has positive motivational worth.
Slack if all the values contained in the agent’s preferences have the same motivational worth (both positive or negative).
Resource-Based Attributes: dynamic constraints Dynamic evaluation of resource use Never slack! Preference structure is monotonic Problem is to determine reservation on the use of a
resource
Negotiation Goal Structure A negotiation goal
contains sets of Fixed attributes Negotiable
attributes Slack attributes Reservation values
for resource-dependent attributes
Opponent Selection
Selecting to minimise conflict
Selecting to optimise resource use
Selecting to Minimise Conflict Each agent selects
its own negotiable attributes
Intersection of choices defines issues
Smaller intersections means less conflict (easier negotiation)
The Conflict Minimisation Selection Mechanism Issue analyser
calculates the expected intersection size of this agent’s issues and opponent’s issues
Uses attribute selection frequency information about the opponent
Selecting To Optimise Resource Use Resource manager
determines the expected deal price of an opponent
Checks to see if the expected deal price is below reservation
Price profiles Concessionary
flexibility
Combining the Mechanisms Selection based on
both conflict minimisation and resource optimisation
Motivationally weighted by the worth of the goal and the worth of the resource
Preliminary Evaluation Tested only on
price minimisation Compare opponent
selection against optimal selection
Agent learns to select the optimal
Conclusions Little autonomy apparent in pre-negotiation
stage Motivation enables autonomous decision-
making in dynamic negotiation settings New model of negotiation goals gives scope
for motivation-based dynamic decision-making
Characteristics of negotiation goal guides opponent selection.
End of Presentation…
Questions?