What are the exact and agreed-upon characteristics that define a "software 
agent system" and can thus be used to distinguish between agent and non-agent 
software systems? All sources I have found so far indicate that there is no 
such definition. Please help!

---

Clarification of Question by safari-ga on 22 Apr 2002 08:22 PDT

I am in fact trying to differentiate between "software agent" technology 
and "personalization" or "user adaptive technology" - in that regard, Mrs. Maes 
is the problem, not the solution.

---

Clarification of Question by safari-ga on 22 Apr 2002 08:25 PDT

I am in fact trying to differentiate between "software agent" technology 
and "personalization" or "user adaptive technology" - in that regard, Mrs. Maes 
is the problem, not the solution.

---

Clarification of Question by safari-ga on 22 Apr 2002 08:26 PDT

hm, sh**ty reload error, sorry for reposting.

http://216.239.51.100/search?
q=cache:1RB8RLyxzG8C:agents.www.media.mit.edu/people/pattie/CHI97/+software+agen
t+system&hl=en

Above is the searchterm I used in Google. This is the final link:

http://agents.www.media.mit.edu/people/pattie/CHI97/

Software Agent tutorial

http://www.acm.org/sigchi/chi97/proceedings/tutorial/mm.htm

Thanks for your hints, but using the Maes definition I could e.g. argue that a 
mail client (say Outlook Express) is an agent system, as it is autonomous (a 
stand-alone program), has sensors (looks for mail), effectors (fetches mail, 
notifies me)and does all of that autonomously (I don't need to push a "look for 
mail" button). While this could even be an agent system, the criteria would 
also apply to my fridge, that (thank god) autonomously delivers cool drinks and 
would thus from now on be called my "cooling agent".

I think that software agents must evolve in a some autonomuous mannor, not only 
by being customized with a script.

Further software agents are the top of current programs and here is why I 
postulate the border is moving:

calculator -> spreadsheet -> scenario / report / chart generator

or

text editor -> word processor -> content processor with your style of writing

or

basic chess program -> with strong moves library -> with knowledge about your 
playing style and teacher capabilities

or

web directory / index -> human researcher -> search agent

etc.

From the technical point of view, the term "software agent" seems to be a kind
of buzzword
that can be used on a lot of different algorithmic concepts. That's probably the
reason why
you did not find "the answer" yet.

To get an idea about the meaning take a look at the word "agent": This is latin
and means
"a person who is acting". So the direct translation of "software agent" is
"software that somehow
acts". Not really a clear answer. :)

If you take a look at existing technologies that are called "software agents",
you'll find basically
two concepts:

1. A software that only simulates individual action, but internally works
without setting up
an individual process to solve a certain problem. Examplare: bidding on auctions
at Ebay.
On the Ebay servers, there are obviously no millions of user-specific bidding
processes.
They have a huge database instead and simulate "bidding agents" with that.

Another "passive" agent could be an "information agent" in some search engine
that sends
you mail whenever new pages/articles on a pre-specified topic or search pattern
have been
found.  Such a search engine will not gather pages exclusively for *you*, but
will expand
the knowledge base accessible for all users and then will use user-specific list
to make the
agent thing work. Again we do not have millions of user processes, but the
individual
"agents" are simulated.

2. The second type of software is really active for an individual. In this case,
starting an
"agent" means that somewhere some new processes (running programs) appear. In
this
case, the "agent" idea is obvious: The agent is started to solve a specific
problem, and
when this is done (or the agent failed), somehow a feedback is returned and the
(specific)
agent ceases to exist.


I think the differenciation in these two classes is in effect very similar to
your distinction
as added in the comments to your original query. The "real" software agents
might be seen
in category 2, and statistically most of the internet services titulated as
"software agents"
actually belong to category 1.


Personally, I wouldn't take that "software agent" term too seriously. In the New
Economy boom
this buzzword has been used often to hide real technical features, which were
oftenly not
really innovative that times.

Well, I think such agent system are called "intelligent" or "learning" agents. 
The question here is: Is the ability to learn or reason necessarily a part of 
the definition of an agent (web crawlers, which are often used as examples of 
agent systems, do not exhibit intelligent or learning bevavior)? And - if the 
characteristics given in my previous reply posting do actually describe an 
agent system - which degrees of perception, activities, autonomy are necessary 
to determine an agent system? Or is an agent system simply another term for an 
encapsulated object oriented system design?

Again, thanks for your comments. First, to clarify "the posting order 
situation", my previous posting refers not to daniel's posting, but to 
cybergenius64's.

Back to daniel's posting, I agree with what you wrote, and it is actually a 
point that I am trying to find arguments for - or be proven wrong in.