date: 9 October
author: Joost

Architecture of Intelligent Web Applications

Traditionally end-user applications (word processors spreadsheets etc.) 
are designed to perform a general, well defined task.  Intelligent
applications (mm presentation generation, natural language text
generation, education systems, diagrams design, 'wizards'
architecture... ), in contrast, differ from these in the sense that
they perform specialistic tasks which require different sorts of
expert knowledge. This makes intelligent applications scarce since the
high(er) costs often overweight the relative limited number of users.

The Internet and WWW have shown the value of uniformity which resulted
in several document standards (data formats) (XML, HTML, SVG),
designed for interoperability. Current activities however focus on
uniform knowledge descriptions (semantic web) and sharing
computational resources (grid).

The INTERNET, and subsequently the interoperability developments change
the way of software development. Programs are designed to work as
components (or services) extending, or in cooperation with, other
programs. (open source, plug-in, jdbc, corba, dot com) The reuse of
existing components allow the development of intelligent applications
at relatively low cost.

Cuypers can be considered such an intelligent application since it uses
different components. The software architecture nevertheless is
rather ad hoc chosen, and needed to be extended several times during
development (which resulted mostly in a complete rewrite). Due to the
number of dependencies, which an intelligent system typically has, such
a change is rather influential. 

A system architectures for intelligent applications is different from
existing software architectures because it models interaction between
systems rather then data structures. The granularity therefore is on
the system level. This makes combining systems not a straight forward
task since they have to 'fit', which means the systems should be able
to interact. Most modern systems are 'open' in the sense that they
provide a mechanism to allow system interaction but this often
requires transformation of data format and a mapping of interfaces
('plumbing'). Especially when combining a number of systems this
becomes a problem. In addition, when one of the components changes the
whole system breaks down. For the integrity of intelligent applications 
this is a problem.

The purpose of the semantic web is to make information explicit and
therefore machine processable. The same analogue might be true for
systems/components; Make the interfaces explicit by using a common
vocabulary (using ontologies). Instead of a human programmer
interpreting an interface and constructing data transformations might
be done by a system itself. Besides that changes in interface don't
break down the whole system which is in an intelligent system an
important feature.