The Technical Reference Model
Intelligent Agents Presentation
Click here to go to the
Presentation that Mark Bergers gave in Manchester in June
1998,
it illustrates the basic idea of the Intelligent Agents
technology.
What we consider to be an intelligent Software agent
Every day, in every job, there are recurring jobs to be
done. E.g. the VCR needs to be programmed, We need to drive
to work, We need to fill in our time sheets, etc. All these
things are necessary, but usually tedious tasks. It would be
handy if we could delegate these tasks to someone or
something else.
This is where Software agents appear. One of their tasks is
to automate the recurring and not so interesting things we
have to do. An agent could, for example, take note of the
things you do all week, and fill in your time sheet.
A Software agent does not need to wait for a message or a
command from the user to start doing a task. He should be
intelligent enough to make this decision autonomously. In
this way, agents differ a lot from traditional software,
because they are active all the time on the background,
minding their own business, looking at what the user is
doing, and appearing only when they decide they can help the
user. Because the agent is always there, he can do much more
than just sit there and wait to help us. By observing user
behaviour, the agent can find out what the repetitive tasks
are, which pages you browse to, etc. By using this
information in an intelligent way, he can try to predict the
goal of a current task and eventually help the user get to
that goal faster, or more efficient.
Another way Agents can be used is to filter the information
we get. This way we don't need to waist time scrolling
through Junk mails and Advertisements or browse to sites
that don't have the information we ask for. As this may be a
very helpful application of Agent technology, it is also one
of the most difficult to program. To accomplish a good
information filter, the agent needs to know what the user
sees as relevant documents. This can be done by simply
giving a number of keywords, or can be gathered by observing
which documents are read.
In a Web environment, the concept of software agents has to
be rethought a little bit. Previous examples both got their
information by doing some kind of user profiling. So the
agent has to know who the user is in order to help him/ her
in a personalised way. One of the main characteristics of
the web is the anonymity of the users. A lot of people only
visit a site once, in search of information concerning a
specific topic, or they stumble on the site by accident. So
there can be no (or at most a very limited) user-profile, or
user behaviour analysis. The intelligence of the agent
should therefore all be programmed, and not dependent of
what the user does. This way, an agent could e.g. provide
help to guide the user through a site.
There are numerous other possible applications in which
Software agents can be useful. There can be reminder agents,
telling you to go to a meeting, or to go to a garage on
time. Agents could observe whom you usually send mail to and
compose mailing lists, etc, etc.
In the end, everyone could have his own virtual self
represented by a personal agent which knows everything you
like or do, and interaction with the computer wouldn't go
via the keyboard, but by talking to ones virtual
representation.
|