Autonomous Agents, Dynamic Systems, Reactive Systems, etc. (Re: Formalization of the Artificial Life Systems)

[Bert . Baker]

On 5/19/97 Barry McMullin <address@hidden> wrote:

>I hope that gives you some additional ideas.  The
>problem I'm most concerned with at the moment is
>whether or not there is a fundamental clash between the
>formalisation of autonomous agents and dynamical
>systems theory (in the broadest sense).

On 5/20/97  Vladimir Jojic <address@hidden> wrote:

>Dear Barry,
>
>First of all, thanks a lot for the list of the references.
>
>Interesting thing is that, Swarm agents should be dynamic, am I right?
>(adding new methods (objc feature) , allocating new space for new
>variables and so on ...)


I think the distinction you make between dynamic systems theory and
autonomous agents formalization is interesting.  From a systems theory
perspective, the classic definition of a dynamic system is a system whose
current output depends on its past inputs and maybe its current inputs.
Roughly we call a dynamic system a state determined system (though
non-dynamic systems can also have a degenerate concept of state.)

Non-dynamic systems are also called instantaneous or memoryless systems,
because their output is only dependent on the instantaneous value of the
input.  There is no memory of past inputs.

A common example of a memoryless system is a resistor where the input is a
current and the output is a voltage.  A common example of a dynamic system
is a capacitor which can store charge/energy.  Or maybe a better example is
a mechanical system where the mechanisms are expected to have inertia from
past excitations.  Such inertia causes the system to be dynamic, these past
inputs result in a current output.  Some people talk about a dynamic system
as a system which has inertia (i.e. energy storage).  Dynamic systems often
require the mathematical apparatus of differential or difference equations,
whereas algebra will often do for a non-dynamic or memoryless system.

Some people find it counterintuitive that a resistive network is not
considered a dynamic system from a classic systems theory perspective.  But
it is not.  There is no energy storage.  The system has no inertia.

I think we can now see why there is a conflict here.  What systems
theorists call a dynamic system is different than the vernacular
connotation of the term.  We can also see why the mathematical apparatus
developed in the systems community for dynamic systems may not help us with
autonomous agent systems.  I think Vladimir makes an interesting point that
autonomous agents can have a form of dynamicism which classic dynamic
systems do not have.  That is dynamic evolution of the system.  Systems
theorists only talk about the state of a dynamic system changing over time,
but they do not put much thought into systems where state space itself
(i.e. the set of states which are being considered) may be changing with
time.

As a side note, I think it is also interesting to compare the definition of
a dynamic system to the AI concepts of reactive and proactive systems.  A
reactive system is a system where an input causes an output.   We can see
that an instantaneous system is also a reactive system.  But, so is a
dynamic system where the output is determined not only by past inputs but
by a current input.  So instantaneous systems and a subset of dynamic
systems are reactive.

The concept of proactivity is not adequately covered in the dynamic vs.
non-dynamic distinction.  A proactive system is somewhat related to the
idea of "causality."  Within the systems theory community, a "causal
system" is a system whose output is only caused by current and past inputs
but not buy future inputs.  (This may strike you as a "why would anyone
care about that" kind of concept.  But the distinction is needed for
mathematical reasons.)  Note that I say future inputs, NOT expected future
inputs.  That is, a proactive system may make decisions based on what it
expects to happen in the future, but it does not make these decisions by
knowing what the future holds.  That is though a proactive system would
seem non-causal, it really is a causal system.

Along these lines, FIPA
(http://drogo.cselt.stet.it/fipa/fipa_rationale.htm) defines an agent as:

        "an entity that resides in environments where it interprets 'sensor'
        data that reflects events in the environment and executes 'motor'
        commands that produce effects in the environment."

I find it interesting that their definition of an agent seems very similar
to the definition I gave earlier of a dynamic system.   Does this mean that
any feedback controller is also an agent?

I think we need to better understand how an agent is more than a dynamic
system, both from a classic dynamic systems perspective and from an agent
technology perspective.  There have been exhaustive arguments on defining
what agents are.  I do not want to get involved with any of these.  But, if
we are going to model agents we need to get a better idea of what they are,
and what they are not.


address@hidden
University of Cincinnati






                  ==================================
   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.
                  ==================================