DAI-List Digest         Friday, 14 June 1991            Issue Number 38

Topics:
  Research on Multiagent Systems at Hagen

Please send submissions to DAI-List@mcc.com.  Send other requests,
such as changes in your e-mail address, to DAI-List-Request@mcc.com.
------------------------------------------------------------------------

Date: Fri, 14 Jun 91 12:48:13 +0200
From: kirn@fuhasch.uucp (Stefan Kirn)
Subject: Task decomposition in multiagent systems

We are a german DAI research group interested in all aspects concerning
multiagent systems. Main direction of our work are so-called federative
(knowledge based) systems (FKBS).  The agents in an FKBS are autonomous.
They decide on their own, wether to take part in cooperation processes
or not. They are able to solve problems stand alone, too.  The agents
are rather complex systems, e.g. expert systems.  Cooperation processes
are supported by a so-called "cooperation competence," which is very
different from local (domain-based) problem solving competence and which
is not completely identical with the social competence of agents.  The
actual research approach is directed to the question, how to make
existing (stand alone) agents (expert systems) cooperative.  Two
essential question arise:
1.   What are the features to make KBS cooperative?
2.   Development of tools to support this task.
In our work there are three main issues:
1.   Competence assessment: How can intelligent systems assess their own
     competence to solve a given task, how can they communicate their
     competence knowledge to support the global process of problem
     solving and what representation formalisms / evaluation strategies
     are needed to represent, evaluate and compare competence knowledge?
2.   Cooperation behavior: Are there different classes of cooperation
     strategies? What coop. strategy would be appropriate for which
     types of tasks / problems? In which cases / under what conditions
     cooperative problem solving would increase / decrease the usability
     of results. In which cases cooperation will be better than
     stand-alone solutions? ...
3.   Common internode language: Building up multiagent systems from
     pre-# existing stand-alone agents leads to the problem of common
     internode languages.  Existing approaches (not only in DAI) are
     very poor.  Until today we are building up our agents with PROLOG
     and, therefore, we are using PROLOG as common internode language,
     too. 

The purpose of the above description is as follows:

1.   We are interested to discuss problems concerning the above themes
     with other researchers interested in those fields.
2.   During these months we are implementing an FKBS called FRESCO that
     solves complex tasks in banking.  FRESCO consists of a group of
     expert systems (credit decisions, analysis of balancing sheets,
     economic growth, ...) coupled via an extension of the classical
     contract net system. Here we are interested to hear about other
     implementations of contract net systems.
3.   One special issue in cooperative problem solving is task
     decomposition.  In FRESCO tasks are defined as (complex) Prolog
     statements.  In this area we are searching for intelligent
     approaches to task decomposition (for example with respect to the
     distribution of competences over the overall system, ...).  Here we
     would like to hear from other researchers and their approaches,
     especially approaches to intellgent decomposition of tasks defined
     by Prolog.

First results of our work has been published recently (1. CKBS,
Keele(GB, Oct. 1990, 2nd genereation expert systems, Avignon/France,
May 1991 etc.).  Requests for copies of these papers should be sent to
me via email (kirn@fuhasch.fernuni-hagen.de).

Looking forward to your responses
Stefan Kirn

FernUniversitaet Hagen
Praktische Informatik I
- Stefan Kirn -
Postfach 940
D-5800 Hagen 1 Germany

Tel.: 02331 / 804-2726
FAX:  02331 / 804-314