[Modeling] Modeling an Agent Class- register your opinion

f.tolman f.tolman@chello.nl
Fri, 13 Jun 2003 14:10:40 +0200

Hi Jim,

Thanks for the welcome. And "yes" we are looking for a modelling language
that helps us to communicate about the designs of our agents.

As to the idea to use the concept of roles as behaviours I wonder why roles
do not fit with the method concept in UML class diagrams. Is it because you
want to explicitly detail each role in a separate class diagram?

An other question is why you do not introduce a two or three level
hierarchy. Meta-meta, meta, ...and apply the same UML notation on each
level? In one model role#1 would appear as a method, and in another model
(an extension of the first model) role#1 would be a class.

Thanks again


----- Original Message ----- 
From: "James Odell" <email@jamesodell.com>
To: "f.tolman" <f.tolman@chello.nl>; "ModelingTC" <modeling@fipa.org>
Sent: Friday, June 13, 2003 1:27 AM
Subject: Re: [Modeling] Modeling an Agent Class- register your opinion

> Hi Frits,
> Welcome to the list.
> On 6/12/03 4:25 PM, "f.tolman" indited:
> > In my field of research we are trying to use agent technology to build
> > virtual actors in the building and construction industry. For example a
> > Virtual Architect, Virtual Adviser, Virtual Cost Engineer, etc. Such
> > actors can be consulted by humans and other virtual actors. When I think
> > such a concept it does not seem to match to the simple UML class. At the
> > other hand it also does not seem to fit too well with the agent concept,
> > in this discussion agents seem to be small (and beautiful?). A Most
> > important parts of a Virtual Architect can be described using models of
> > classes, relations, constraints, properties and methods. Some parts,
> > on a meta level, can be desrcibed by agent concepts.
> >So the question is of course: is there agreement over the question which
> > types of agents are in scope, and which not?
> Currently, we do not discriminate against (of for) any particular kind of
> agent -- within reason.  In fact, we are trying to get our arms around the
> question: what are agents and what do they entail.  If we are going to
> develop a modeling language, we need to understand that kinds of agents
> need to be modeled.  So, if you have a kind of agent that you need a
> modeling language to express and communicate about, this might be your
> group.  You task would be to determine what kind of graphical modeling
> notation would best express and communicate "virtual actors."  And,
> you can find others in the Modeling TC that might share you quest.
> >And next: an agent that performes its roles is an
> > occurrence, but does that mean that its behaviours and such can not
> > described by a type-like something (class, template,..)?
> Currently, there is a small subgroup of people working on "role."  For us,
> role is esentially a class (or AgentClass).  When an agent is an instance
> a role, it acquires the properties of that role -- similar to an object
> acquiring the structural and behavioral features of its class.  For
> when you say that particular Broker agent "performs its role", this
> that the agent is executing the Broker-based behavior.
> Does this make sense?
> Cheers,
> Jim