[absintelagent] Re: Defining "agent" -- the teleological view
- In the context of
At 20.53 28/09/99 -0400, Luke Kaven wrote:
>About Function, Goal and Agent
>Gary Noel Boone writes:
>>It was fashionable in the 60s to try to explain curious features of
>>animals by figuring out how these features were adaptive. What was the
>>purpose of wings? Why, to fly, of co ...
In everyday practice we use the concept function in two notions.
It is either mathematical function or a goal-oriented property of a thing.
The last one depends on our practical application/use of the thing, it
means it is not a physical property.
From the engineering perspective, engineering system functions are
internal and external.
External results from the user requirements and are relations between the
system and its environment, internals are consequence of technologies
Using engineering point of view on human organizations we may say that they
have functions because they have always a foundation goal.
Pseudo-functions are properties of not engineering systems (living systems)
which satisfy some pseudo-goals, for instance the integrity, survival, ...
of the system.
In practice, this operational assumption (metaphor) is so useful for the
system diagnosis, modifications and so on, that if something continuously
offers some "service" to the others, it is considered as the "real"
definition of function, by many researchers.
Goal, as a concept, exists only for intelligent agent. It is a state
localized out site of the system in discourse.
In general, goal and functions can be considered relative. If we assume
that X is a goal for SS in the distinguished another system SE, then a
property Y of SS necessary to achieve X, is a function. If we assume that
Z is a goal for the integrated system SS+ SE, and X is
the necessary property for this then X can be considered as a requested
function of SS+ SE.
In the above context we also have functional definitions.
For instance, quasi all agent definitions are functional. They specify what
agent does, but
the goals are usually implicit.
Usually, the function specifies only what system does/may do, but not how
it is/will be
'Agent' term is used to call certain software systems, cognitive systems,
Of course in all these domains, every agent should have the same set of
This aggregate of abstract common properties can be called 'abstract agent'.
Not only in my opinion, if we intend to exclude all physical interactions
between physical bodies, an agent must be goal-oriented, and as a
consequence, it also has functions.
Here, we need to distinguish design-goal and intervention-goal.
A design-goals have every mechanical tool which we exclude from the
denotation field of agent concept. Therefore, an agent should have
intervention-goals and, in the behavioral perspective, goals should be
dependent on circumstances, in the contrary, all software programs will be
The problem is how in a simplest and sufficiently general manner to define
a goal choice.
As an intervention-goal is defined as a hypothetical state of the agent's
domain of activity,
every agent should have some more or less ordered "base" of the possible
states of this domain, according to an utility/importance/... scale.
I think that we may accept this, as an assumption for us and as a condition
for an agent definition.
A minimal ordering is done by the ordered couples, it could be the
mathematical preference relation.
From the preferences base results such concepts as desires, if we add
concrete goals we may obtain intentions.
- More information about the above conceptualization, and how it is going
on, you can find on my home-pages.
The next will be added in the near future, I hope.
Adam Maria Gadomski Sr. research scientist
Italian National Research Agency ENEA
C.R.Casaccia,s.p.111 tel.+39-06-3048-3404 (-3504)
00060 Rome fax.+39-06-3048 6511
Home page: http://wwwerg.casaccia.enea.it/ing/tispi/gadomski/
ENEA sites: http://www.enea.it, http://tisgi.casaccia.enea.it/