Intelligent Icons use visualization and mining techniques to convey information about the content of files by means of small informative icons. Icons are built in a way that files with similar content share similar visual appearance and positioning, thus enabling comparison and grouping.

There are several interesting features in intelligent icons:

• Different color scales or layouts can be used to accommodate and distinguish between different file types, so that they don't get mixed up.
  
• The icons can be used at different resolutions so that they can scale to any default icon size provided by the operating system.
  
• A generic model is devised to permit the comparison of new file formats (through the development of novel plug-ins). The examples in the paper provide mechanisms for: dna data, time-series, pdf files, games.
  

The thing I like the most of this work, however, is the original approach and the paradigm shift from typical infovis tasks and scenarios. And this seems to be a specific purpose of the paper! In the authors' words:

The vast majority of visualization tools introduced so far are specialized pieces of software that are explicitly run on a particular dataset at a particular time for a particular purpose. In this work we introduce a novel framework for allowing visualization to take place in the background of normal day to day operation of any GUI based operation system such as MS Windows, OS X or Linux. By allowing visualization to occur in the background of quotidian computer activity (i.e. finding, moving, deleting, copying files etc) we allow a greater possibility of unexpected and serendipitous discoveries.

I like the idea of escaping the traditional infovis scenario: "get some data + give it a shape + provide interaction + let's go hunting for new knowledge!". I am sure the type of tasks that can be enabled by infovis and the different scenarios that can be covered go far beyond traditional data exploration (even if I think there is yet a lot to do in the traditional case), and this is a very good example.

For more information see the paper:

Intelligent Icons: Integrating Lite-Weight Data Mining and Visualization into GUI Operating Systems. E Keogh, L Wei, X Xi, S Lonardi, J Shieh, S Sirowy - ICDM 2006.

For some background info see also:

VisualIDs: automatic distinctive icons for desktop interfaces. JP Lewis, R Rosenholtz, N Fong, U Neumann - ACM Transactions on Graphics (TOG), 2004.