Terminology: ABM, Swarm, MAS, etc. (was: Swarm futures...)

[Gulyas Laszlo]

Hi,

It seems we are on _really_ different grounds as far as terminology
is concerned. Not intending to start an end- and meaninless flame war,
but wanting to expose the terminology I 'grew into', I'll put forward
my case. Those of you not interested in this kind of bubble may
stop reading now...


I am originally coming from the AI field. There people talk about
Agent-Based Programming and/or Multi-Agent Systems (MAS) and implement,
for example, BDI agents with agent-toolkits (in the AI sense), like
Zeus, Jack, JAFMAS and the like.
They may also do swarm-intelligence stuff where their agents' internals
are less detailed or complex, but the population as a whole shows 'intelligence'
(whatever that may mean). For this latter, they may or may not use
AI/agent-tools. Same is true for SDG's Swarm package. I've seen both
kinds of tools used for this.

Then, there is Agent-Based Simulation/Modeling (ABS/ABM), mostly used
in the complexity and/or social sciences. There the agents' cognitive
processesare usually vastly simplified (at least, in the sense of
neglecting most of the philosophical and theoretical work that has been
done in AI, cognitive sciences, or in agent-litterature, for that
matter...).  The focus is more on creating simple, generative models of
system-level phenomena.

And finally, there are the concrete versions of ABM tools: Swarm, Repast,
Ascape, StarLogo and the like. (Of course, there are MAS tools, too, like
Aglets, AUML, JAFMAS, Jack, Zeus, etc., but those should not be relevant
in this discussion.)

When I talk to 'agent people' (in the AI sense) I usually have to explain
that Swarm and the like would really have to be called 'object-based
simulation', at least in the strict sense of their definitions. [So much
so, that once, after a presentation to a general complexity audience on
Repast, I received serious critique from an AI researcher, because I
'failed to show' how its agents are proactive, reactive, temporally
continuous and all the rest...]


Most AI people usually agree that Swarm-like agents may be called agents,
in a very broad sense of the word, but they usually miss most of the
properiues they require for agents. So, to them, and to me, ABM is a
very specific kind of MAS that only matches some really flexible
definitions for MAS. Of course, as you point out, there is no technical
limit that would prevent someone from implementing AI-type agents in Swarm
or in Repast.  So in that sense MAS is a subset of ABM. But if we assess
what is actually being done with those tools and what Swarm/Repast
is most suitable for, we have to conclude the other way around. In order
to actually implement real AI-type agents in Swarm, one would have to take
on a rather serious project, not much less than trying to do the same
thing in pure Java or C.

I completely agree with you that we shouldn't be sloppy about terminology
and I don't want to be splitting hairs either. But during the last few
years I spent in these fields, I've never come across the kind of
terminology/classification you put forward. Of course, this is partly
because I couldn't participate in the last few SwarmFests... ;-)

My 2 cents,

Gulya




                  ==================================
   Swarm-Modelling is for discussion of Simulation and Modelling techniques
   esp. using Swarm.  For list administration needs (esp. [un]subscribing),
   please send a message to <address@hidden> with "help" in the
   body of the message.
                  ==================================