I spent part of my lazy Sunday on reading a few articles in Beyond the Desktop Metaphor: Designing Integrated Digital Work Environments, a book that Kris dropped on my desk a few weeks ago. It gives an overview of the state-of-the-art in integrated digital work environments and is edited by Victor Kaptelinin and Mary Czerwinski.
Lifestreams was an alternative to the desktop metaphor that was developed starting in 1994 and aimed to be a better way to organize your personal electronic information. One of the primary motivations for this work are the limitations of a static (hierarchical) filesystem. The problem with organizing our documents in the filesystem hierarchy is that information generally falls into fuzzy categories and that it is impossible for users to generate categories which remain unambiguous over time. Furthermore, users are forced to name their files, which often results in meaningless file names such as “draft1.doc” and “draft2.doc”. Names are an ineffective way of categorizing information, since their value decays over time. Traditionally, people do not name their documents as pointed out by Thomas Malone in his paper How do people organize their desks? Implications for the design of office information systems. He noticed that people often just create nameless stacks of related documents on their desk. Freeman and Gelernter discuss a few other problems with the desktop metaphor, such as no support for archiving, reminding and summarizing. The desktop metaphor does not make it easy to archive information, to put information somewhere we can later retrieve it but also remove it from our periphery. Users often place information on their desktop to remind them of tasks to do or leave an email in their inbox to remind them that they still need to reply to it. As the desktop has no semantic notion of reminding, users are just working around the system. Finally, summaries are needed in order to cope with all our electronic information. The authors state that summaries are often application-centric (e.g. an overview of your photo albums, an summary of your music, etc.), instead of system-wide.
I found it interesting that the authors do not see their architecture as another metaphor, but as a unified idea or system. They refer to Nelson’s concept of virtuality as opposed to metaphorics. Nelson (who also coined the term hypertext) argues that adherence to a metaphor prevents the emergence of things that are genuinely new. Trying to adhere to a metaphor may lead to strange results when new functions are added, for example having the drag a CD icon to the trash to eject it on Mac OS X.
A lifestream is a time-ordered stream of documents that functions as a diary of a user’s electronic life. Every document he or she creates is stored in the lifestream. Moving forward from the tail to the present, the stream contains more recent documents. Moving beyond the present into the future, the stream contains documents that the user will need (e.g. reminders, calendar items, etc.). The system has a few primitive operations that together support transparent storage, organization through directories on demand, archiving, reminding and summaries: new, copy, find and summarize. New and copy are used to create or copy documents in the lifestream or between lifestreams. Documents do not have to be named. The find operation allows users to search their documents. It creates a substream with the results of the query. These substreams are not static, but are updated on the fly whenever new documents that are relevant to their query appear. Users can allow substreams to persist, in order to quickly find information they need regulary (e.g. “emails from Joe”). Finally, summarize compresses a substream into an overview document. The method of summarizing varies according to the content of the substream (e.g. a music playlist, a prioritized to-do list, etc.). The figure below shows the Lifestreams user interface:
It’s interesting to see that many of the ideas first explored in Lifestreams are currently supported by several applications. Archiving was one of Gmail‘s defining characteristics (“never lose a message again!”) when it was first released. Apple’s iApps such as iTunes offer summarization, dynamic substreams (“smart playlists”) and time-based visualizations. Desktop search tools such as Google Desktop, Apple Spotlight and Beagle offer a way to quickly find items on your computer. Some of them also offer saved searches (which is again similar to “dynamic substreams”). The authors also discuss this evolution. However, they feel that desktop search, while definitely a step in the right direction, is not sufficient. It only works if you know what to look for. People really need good browse engines instead of search engines. This statement is also made in the next chapter on Haystack where it is called orienteering.
Haystack can be seen as a generalization of Lifestreams. Haystack is a way to visualize and organize a user’s information, but does not restrict the visualization and categorization to be time-based. The authors try to find a solution for the fact that current applications force users to manage information in the way that the application designer envisioned it. This might not be the most natural way for the users, so Haystack gives the users more control over what kinds of information they store and how to visualize and manage it. In traditional email applications for example we can only categorize by the labels that are predefined (e.g. sender, subject. etc.), but not by our own features such as “needs to be handled by such-and-such a date”. The information may even be in the application, but no appropriate interface is offered to use it. Furthermore, every application manages its own data independently while we might want to relate data from different applications together (e.g. emails, articles, blog posts, pictures, songs, people, etc.). A user might also want to add a new data type. Consider the location field in a calendar event: this is just a string, while the user might want a richer presentation (Google Calendar can do this by linking to Google Maps by the way). Existing applications are very bad at extending existing types, since they offer no way of displaying the type, no operations for acting on it and no way of connecting them to other information objects in the application.
Haystack has a generic user interface architecture that supports impressive personalization. Users can for instance create a new “Send to Joe” operation by filling in part of the “Send to” operation, and saving it. Objects can be dragged upon each other to connect them: dragging an object onto a collection adds it to the collection, while dragging an object onto a dialog box arguments binds that argument to the dragged item.
Custom workspaces can be constructed by drag and drop. The figure below shows a workspace specialized for writing a particular research paper, presenting amongst others relevant references, coauthors and outstanding to-do’s.
The system uses Semantic Web technology (more specifically RDF and URIs) to represent information objects, their attributes and relationships to other information objects. However, they do not enforce schema such as RDFS or OWL) in order to allow users to organize information the way they want. It is after all difficult to create an ontology that serves everyone’s needs. Consider for example the composer attribute of a symphony concept. A reasonable constraint is to restrict composers to be people. But this will prevent a user that is interested in computer music from entering a particular computer program as the composer. The authors state that schemata may be of great advisory value, but they argue against enforcing them. Apparently this is also known as a semi-structured data model.
I think this is the most impressive Semantic Web application I have seen, although I am also looking forward to test Twine and Powerset. I have barely touched upon everything that Haystack can do in this blog post so if you are not yet convinced, have a look at a paper that is pretty similar to the book chapter. The level of customization supported by Haystack reminded me of the Meta-UI concept (which I see as a user interface to manipulate an interactive system or its user interface) as discussed by Coutaz at Tamodia’06.
Although Lifestreams and Haystack would certainly improve the way we manage our data, I feel they both ignore an important type of information: information in the physical world. After all, a substantial amount of the information we process is non-digital. Last year, I had a project proposal for the course Actuele Trends in HCI (translated: “Current trends in HCI“) on improving the way we work with digital and physical information. Given that the students had little time for this project, the result was pretty nice.