If ubicomp is to come into being, user issues in the dense ecology of
computational services will become paramount. In this talk, I argue
that there is a new kind of interactional work coming to the
foreground, configuration work - the work that is involved in keeping
systems running, running together, and running over time. This is
work we do now but is largely invisible and unremarkable; in the
future, it may be overwhelming.
One way to deal with this problem is have proprietary stovepipes.
Another is to automate the process. The first does not provide an open
systems solution, and the second is likely to remain brittle. As
another line of attack on the problem, I present Escalier, a system we
are developing that uses social computing techniques to help people
with their configuration work. Using system configurations gathered
from the crowd, the Escalier system can help users know what
configurations are stable and reasonable. Escalier creates a Bayesian
map of the crowd's settings, but also contains an expertise layer to
know which crowd members to trust. In this talk, I will argue for
configuration work, and then provide an overview of our work on
Escalier and its applications.
