---- Proximity application idea: personal information manager ----

I recently tried a piece of commercial software called "eGems" whose
utility I believe could be radically improved if it were implemented
using a Proximity-based personal information system. (My working title
for such a system is "POW - Personal Object Web".) I'll provide an
introduction to the "big idea", describe the specific program (eGems),
then indicate how I think we could do it much better. I welcome
discussion of this topic, which I find singularly diverting.

Introduction/History: Over the last five years I've reviewed and tried
hundreds of applications called Personal Information Managers (or
PIMs). I believe they have all "missed the boat" in that almost all of
them assume that a hierarchical-based representation for managing
personal information is ideal. I'd argue that a semi-structured,
self-structuring database provides many benefits. The few PIMs that
are organized using networks don't have our other ideas (discussed
below).

The Idea: What would a Proximity-based PIM be like? First, it would be
an application that primarily supports research into areas whose
nature is unknown until discovered. I'm sure we can come up with a
better definition, but David calls these "forensic" activities, and
would include representing personal information, doing historical
research, analyzing the spread of infection, and performing CS
research. These tasks are ones which require a semi-structured
database because the structure of the information being captured and
developed during the research *evolves* during the course of the
work. That's the semi-structured part, but I'd argue that to make
effective *use* of this possibly vast and confusing network users will
need advanced tools to help manage it, which we've been grouping under
the term "self-structuring." These tools would be the work products of
research we're doing in KDL's, including classification, existential
prediction, clustering, etc.

A system with these two ingredients (i.e., a semi-structured,
self-structuring database) that could effectively handle large amounts
of data would have many revolutionary applications. (Sorry if that
sounds bombastic, but I'm excited.) We can talk later about "subsuming
the file system," "Smart Paper," and other interesting topics, but a
simple one I want to describe is the "web research tool," which I
describe next and compare to eGems.

Application: Web research tool: eGems (http://www.egems.com/) is a
tool for collecting diverse information obtained while doing research
using the Web. It uses drag-and-drop
(http://www.egems.com/products/features.html) to integrate with
Windows applications, and you use it by selecting content and dropping
it on the window's icon. The program then asks you to describe the
information (via notes, source, and bibliographic information) and
asks you where to place it in the hierarchy. (They use the gem
metaphor to add excitement - each piece of information is a "gem", and
they are stored in "trays" within "chests," i.e., folders.) You use
the information via hierarchy navigation and textual search, and via
some reports (bibliographies). Value is added by providing for sharing
of databases, and the company hires the creation of useful ones
(Shakespeare's Sonnets, Mark Twain's Maxims, the U.S. Constitution).

Analysis: People like this tool because it provides a single
easy-to-use and non-intrusive tool for collecting research
information. Although the hierarchy it is based on is easy to
understand and display, it is just one possible organizing approach,
has important limitations, the main ones being 1) it doesn't capture
the information's interconnectedness, and 2) it provides inferior
navigation and use of the captured information. The search tool is
text-based, which requires users to consistently choose and apply the
right keywords in the notes section. In this way, text keywords
provide an orthogonal organizing method, but one that's completely
unstructured (also called "free-form" in the PIM literature).

Proximity solution: So how would a Proximity-based tool work, and how
would it be better?  First, bits of information ("gems") would be
organized into an interconnected network, instead of a
hierarchy. Because all analysis tools require that the network be as
rich as possible, the program must automate creating good connections
(part of "self-structuring"). It would do so by analyzing each dropped
gem and creating links based on the dropped item's content and the
network patterns of similar items. I envision a "mixed initiative"
approach to managing the network - links would be created 1)
interactively by the user, and 2) automatically by the program.

An example creation scenario will help to clarify: Suppose you're
researching the origin of the moon (I just finished "From the earth to
the moon" - great series). You've started by creating objects for the
Moon, Earth, and Sun, and connecting them (Moon -SatelliteOf-> Earth,
Earth -PlanetOf-> Sun). Now you hit britannica.com and drop a grab a
few sentences on origins. The program creates an object for the
dropped text, including its source web page, and connects the words
"moon", "earth", and "sun" to your existing objects. (Yes I know: how
this works is a major research topic.) You edit the moon link to go
through a new intermediate object called "origin theories", and you
link to it an image you drop in. You find two more web sites that have
other theories (including some world creation myths) and link them in,
but you don't add images. The program tells you these are similar to
the first origin, but are missing images. You find some and add them
in. As you've been working you've thought of additional information to
add to origins, including when it was first hypothesized, and who
contributed them. You update the first origin by creating the
appropriate links and objects, then you tell the program to create a
similar configuration for the other two, which it does, indicating
where data is missing. And so on.

Once the information starts building up, how would you use it? First,
you could browse it in various ways, depending on your needs. Graph
views would emphasize connectedness by showing objects and links in 2D
or 3D as nodes and edges. Tree views would emphasize containment by
treating each link as a "contains" relation, thus providing
hierarchies when needed. One hierarchy might be:

  Solar System
    Earth
    Moon
      Origin Hypotheses
	Origin 1
	Origin 2
	...
    Sun

Notice that the network can provide a meaningful hierarchical view for
free. Similarly, a timeline view would be able to show events in
chronological order for any object or link that has contains date. You
could view objects with mixed text and link data as "sticky notes"
that would show them like tiny web browsers. An example in patented
"AsciiVision", where links are indicated in square brackets:

  +----------------Origin 1---------------+
  |  At the end of the 19th century the   |
  |  English astronomer [Sir George H.    |
  |  Darwin] advanced a [hypothesis] that |
  |  attracted considerable attention     |
  |  for decades. On the basis of the     |
  |  [mathematical theory of solar        |
  |  tides], Darwin suggested that the    |
  |  [Moon] had been originally part of   |
  |  the [Earth] but was broken away by   |
  |  tidal action and receded from the    |
  |  planet.                              |
  +---------------------------------------+

A second way to use the network is via search, either traditional text
search (e.g., All Objects Containing "lava"), or relational search
(e.g., All Hypothesis Before Darwin's Birth, or All Hypothesis Linked
Somehow To Disproving Arguments).


Conclusion: I hope this note gives you an idea of an exciting
Proximity possibility, and opens the door to considering other cool
applications. I'd love to discuss this further, and see what else it
might bring up for cool application ideas.

---- EOF ----