MULTI-AGENT INTERACTION IN SOFTWARE PROJECTS
A SYSTEM DYNAMIC APPROACH
y. K. RAL AND B. MAHANTY
Department of Industria! Engineering and Management
Indian Institute of Technology; Kharagpur- 721302
India.
Abstract: Software development is intellectual skill intensive group activity. Since the software
project environment is characterized by accumulation and distribution of knowledge for decision
making , it is not sufficient to treat the software development personnel merely as an intellectually
skilled person. He/ she is, in fact, an autonomous intelligent agent acting in a multi-agent
environment. In this article the authors outline the system dynamics equivalent of architecture of an
autonomous intelligent and interaction of an agent with another agent. The authors also outline certain
styles of cooperation and non-cooperation emerging out of certain basic preconditions for them.
Key words and phrases: Multi-agent interaction , software development, cooperation, non-
cooperation.
INTRODUCTION: Software development is a process. Personnel, with skill reasoning and decision
making ability work in a group to attain a common goal, thus, constituting a multi-agent
environment. When an autonomous agent is introduced in multi-agent environment he/she brings with
him/her an agent space characterized by perception, skill, reasoning capability , and decision
making — capability
( Demazeau, Y. & Muller, IP. 1990). These agent spaces interact through their; - Perceiving
capabilities (Steels, L.; 1990), Reasoning capabilities (Martial, F.V., 1990) and Decision making
capability ( Castelfranchi, C.; Campbell, J.A. & D’ Inverno, M.P.; 1990).
‘When agents interact with each other through their perceiving-, reasoning- and decision
making- capability, cooperative and non-cooperative aspects come to the fore, for , they are the
property of interaction ( Rai, V.K., Bandyopadhyay, S. & Basu, A.; 1996). Demazeau, Y. & Muller,
IP. (1990), present several kind of agent behavior based on the following criteria,
1) The locality of the task to be performed by the agent: Personal( local) or interpersonal (global).
435
2) The capability of the agent to perform the task alone: abie or unable.
According to these criteria there are naturally four ways to describe how to take into account the
other agents evolving in the same world.
Cohabitation: The agent has to successfully accomplish a task and it is able to execute it alone.
Cooperation: In order to perform a personal task , an agent will have to cooperate with others either
because it is not able to accomplish it alone or because others can accomplish it more efficiently.
Collaboration: Some global goals may concern all agents and may be realized individually by several
agents. The main probiem deals with electing one of the agents to carry out the task.
Distribution: Finally, some global goals can be achieved only by several agents. collectively. The
main problem deals with splitting the global. task and distributing it to the cooperative agents.
In carrying out these modes of behavior following are exchanged among agents.
i) Knowledge it) Possible solution iii) Choice
These criteria and subsequent mode of. behavior thereof has been discussed in the paradigm where
intelligent agent is a machine/ expert system . While dealing with human agents, authors believe,
another set of criteria , involving cognition , be taken. Our presentation of several kinds of agents
behavior is based on the following two criteria.
Capability of the agents: Whether the agent is capable or not capable.
Willingness of the agent: Whether the agent is willing or not willing.
And therefore, cooperative and non-cooperative aspects both come to the fore in multi-agent
interaction given the fact that an agent may or may not be capable and may or may not be willing to
act or cooperate. Some of the most frequently occurting non-cooperative styles are:
i) Conflicts ii), Lack of capability iii) Lack of Interest iv) Deception v) Lack of ;
sommunication and Knowledge & Information sharing.
It should be noted that while no one disputes the desirability of cooperation, non-cooperation
insinuates to the deeper malaise in the system which must be addressed in order to create
preconditions for cooperation. Also, non-cooperation is some times desirable for the robustness of the
system ( Galliers, J.R., 1990).
In the light of the above, we propose the following in the form of figures / illustrations.
D Cooperation / Non-cooperation as a joint function of capability and willingness of an agent.
436
i, +
Agent A ° Action + a Choice Sg pecision .
al sate D capability
wo TA
Action: Possibilities
+
+* Action Reasaning
consequences ability
Communication Perceptive +
vith environment TOMY ———F-—g, Knowledge
i >
Agent 8 Aetion Z—— “hice Lg | pecision
— rote ii capability +
+
Action Possibilities
y
Action Reasoning
consequences ability
Communication ‘ a
with environment 5 Knowledge
Fig. 2
Strong interaction between decision ond reasoning capabilities of
Agent A and Agent B-
#
Rework
Rework rate
+
Errors,
>) Conflict
+
PO Non -cooperitfion
Ambiguities
Perspective
| entargement
fr
+
Shared vision
Cooperation S——
+
Alternatives /
Possibilities
NEGATIVE ROLE QF CONFLICT POSITIVE ROLE GF CONFLICT
Fig. 3 Negative and Positive rote ‘of confticis.
1) The architecture of an intelligent agent as envisaged by a causif loop and its interaction
with other intelligent agent.
TLD Positive and negative role of non-cooperation, specifically, conflicts.
REFERENCES
The following °* marked papers in this list of reference may be found in the same source appearing
at the end of them.
*Campbell, J.A, & D’ Inverno, M.P. (1990), Knowledge Interchange Protocols.
*Castelfranchi, C. (1990). Social Power.
*Demazeau, Y. & Muller, J.P. (1990). Decentralized Artificial Intelligence.
*Galliers, J.R. (1990). The Positive Role of Conflict in Mult-Agent Systems.
+Martial F.V. (1990). Interaction Among Autonomous Agents.
*Steels, Luc, (1990). Cooperation between Distributed Agents through Self-organization. -
*Source: In Decentralized A.I. Demazeau, Y. & Muller, J.P. (Eds), Proceedings of the first
European Workshop on Modelling Autonomous Agents in a Multi-agent World, Cambridge,
England, August 16-18. Elsevier Science Publishers , North-Holland.
Rai, V.K., Bandyopadhyay, S. & Basu, A. (1996). In proceedings of 7th National Conference on
,”Roie of Science and Technology in National Development- A System Dynamics Approach.
Action
Action
being
* carried
out
i
‘
“a
Fig-1 Action as 9 joint function of capebility
and willingness of cn agent.
"| 437
