More details about FdL vs Remembrance Agent

[Wolfgang Müller <muellerw free fr>]

Re - hi,

I've been skimming the thesis of Bradley Rhodes. Effectively, many 
things I want to do in FdL are thought about there, and lots more
(like using a RA in wearable computers) that are out of scope for 
FdL. The thesis is really great and an interesting read.

After looking at Rhodes' thesis, I find his work and related work
done in his group very encouraging for FdL:

According to the thesis, the RA does not do any learning. Imagine
something that learns your preferences, or the preferences of your
group. Imagine learning between users of a package etc. . Plenty
of issues to adress. Fun stuff, and useful. The 
"hand-added semantic links"-idea in the last mail 
would fall into this category.

Rhodes demonstrates that people find the RA useful and entertaining.

Things like the RA call for a common API for programs that might 
make use services of an RA. Without this, you will have to write
an RA-->package interface for any program. They also call for
common ways to deal with the user's needs etc. 

This, in fact are FdL goals:
A framework for using retrieval services integrated into a 
desktop. Interactive and possibly JIT retrieval everywhere,
for every type of file. Learning from user interaction all 
the time.

There is some scalability challenge here: there are lots of 
learning methods out there which show their utility in a limited
context, on a small demo system etc.  There are things like
"direct hit" which do the same on the web. Now I think, it would
be time to see how far you can go if you have learning methods
like that follow each step of your work, see how you work, what's
interesting for you, your associations etc. . FdL is a project
to make the free desktop a unique research tool in this direction.

On the way, we would give users a unique experience.

This is not possible without complete integration into the desktop. 
But somewhere we will have to start, and this would be by showing
that retrieval services are useful, and that they get more useful
if they can be used with more types of files than just text.

With this I hope we can attract the critical mass.

Cheers,
Wolfgang