Autonomous Negotiating Teams

1
Autonomous Negotiating Teams
Jon Doyle Robert Laddaga Howard E. Shrobe
Massachusetts Institute of Technology http//www
.cdm.lcs.mit.edu/projects/ants
2
Research Foci

• Agent system architecture
• Time-bounded negotiation mechanisms
• Negotiation information dynamics
• Preference management and evolution

3
Agent System Architecture

• Add intentional negotiation mechanisms to
  distributed object substrate
• Representing negotiation goals, plans, and
  strategies
• Communicating goals, constraints, and preferences
• Supporting cooperative and competitive behaviors
• Robust distributed objects
• Dynamic reconnection of remote method invocations
• Self-monitoring and repair strategies
• Meta-glue, Corba, etc.
• Extend and interoperate with Java and Lisp

4
Time-Bounded Negotiation Mechanisms

• Negotiating intentional constraints on services,
  solutions, and the negotiation process itself
• Assumptions and stipulations, goals and
  preferences
• Policies and principles, processes and mechanisms
• Technologies Contract nets, KQML, AOP, Corba,
  etc.
• Negotiating and optimizing agreement specifics
• Simple agreement mechanisms (e.g., splitting
  differences)
• Computational market mechanisms
• Dynamically construct markets tailored to
  specific tasks
• Libraries of market mechanisms, task-oriented
  market structures, and market agent types
• Market templates for standard negotiation task
• Standard types of consumers, producers,
  contractors, agents
• Cooperative and competitive bidders

5
Negotiation Information Dynamics

• Negotiation produces new information that can
  change everything
• Moderate goals if found to be infeasible
• Temper goals to further cooperation
• Modify preferences if counterproductive
• Reconcile conflicting goals and preferences
• Negotiation exposes unforeseen uncertainties that
  require amelioration
• Decompose goals that prove nontrivial
• Invent reasonable preferences among novel options
• Specialize generic preferences to situation at
  hand

6
Preference Management and Evolution

• Representing generic preferences
• Preferences among classes, not among individuals
• Timeliness outweighs cost, prefer domestic
  suppliers
• Robust guidelines insensitive to tradeoff
  uncertainties
• Let informants state simple changes simply
• Constructing negotiation preference models
• Qualitative bases for reasoned decision-model
  construction
• Specifying constraints on structure of
  quantitative models
• Expressing desirability of negotiation options
• Constructing market agent utility measures
• Sanity checking market decisions
• Qualitative checks on quantitative outcomes
• Reversing preferences based on decision outcomes