Title : Using reciprocity to adapt to others
Authors : Sandip Sen & Mahendra Sekaran
	  Department of Mathematical & Computer Sciences
          University of Tulsa
          sandip@kolkata.mcs.utulsa.edu
          mahend@euler.mcs.utulsa.edu

Abstract:

A key concern about the design of multiagent systems is the
tradeoff between local autonomy and global performance.  If
agents can be assumed to be working towards a common goal or to
be cooperative in nature, significant overall performance
improvements can be obtained over selfish and uncooperative
agents.  In a distributed and open system, assumptions about a
system-wide common goal or about the philanthropic nature of
agents are unrealistic.  We provide a simple behavioral rule
based on reciprocity that allows autonomous agents to effectively
share workload with others.  We use a package delivery domain to
demonstrate how reciprocal behavior can allow agents to produce
desirable global performance without sacrificing individual
autonomy or efficiency.  More interestingly, we show that agents
who do not help others are less efficient with this mechanism
than agents who share workloads with others.  As a result, the
best self-motivated actions lead to system-wide cooperation.