IEEE Technology and Society Magazine - September 2021 - 80

Feature
Notes on Problem
Formulation in Machine
Learning
Razvan Amironesei
Google, Mountain View, CA 94043 USA
Emily Denton
Google, New York, NY 10011 USA
 Our wOrk starts by offering two statements,
given in different historical times and in different
epistemic contexts, which reveal two radically different
visions of what a problem is and how we should
engage with it.
" [...] people who inquire without first stating the
difficulties are like those who do not know where
they have to go. " 1
" And a lot of times people are too careful, too.
I think it's more useful to make things happen and
then apologize later than it is to make sure that you
dot all your eyes now and then like just not get stuff
done. " 2
At a very basic level, our provisional hypothesis is
that these two statements reveal the role of problems
within different logics of inquiry. Both statements
reveal two distinct
ethical and epistemological
visions, two attitudes that are in many respects unacknowledged
for very different reasons. On the one
hand, slow and considerate inquiry is " too careful "
because it forces and puts us in a position to " dot "
our eyes and " not get stuff done. " On the other
hand, reflective action is disqualified in a highly
ambiguous epistemic environment of contemporary
socio-technical systems. Building on our hypothesis,
we can state that Aristotle formulates the initial historical
conditions for an ethos of problematization
that was subsequently transformed and eventually
lost. We define problematization as a historically,
1 Aristotle, 1984: Book III (B), 995a24-995b4.
2 CS50 Lecture by Mark Zuckerberg - December 7, 2005, [22:30 - 22.48].
Digital Object Identifier 10.1109/MTS.2021.3104380
Date of current version: 2 September 2021.
80
0278-0097/21©2021 IEEE
Alex Hanna
Google, Mountain View, CA 94043 USA
technically, and politically constituted process
which makes or elaborates objects of thought, that
is, problems. Problematization, put simply, refers to
the activity of problem-making, that is, the activity
of constituting problems as objects of thought and
practice. In the absence of an ethos of problematization,
a different ethos is now dominant as an infrastructural
condition of contemporary technological
innovation: an ethos of speed. As a challenge to
" move fast and break things, " Benjamin [3] proposes
instead to " move slower and empower people. " Our
task today is not simply to intellectualize problems
as Aristotle would suggest or to disqualify carefulness
in the name of groundbreaking actions as Mark
Zuckerberg indicates.
This article discusses the relevance of problem
formulation as an interpretive practice of inquiry for
machine learning (ML). First, we analyze one example
developed by LinkedIn practitioners. We show
that problem formulation works as an interpretive
activity to ethically shape the articulation and outcome
of current work practices of ML practitioners
by putting forward strategies to develop meaningfulness
and practices of social inclusion. Second, we
identify paths toward responsible problem formulation
as a way to contribute to the reflection on the
development of standards of analysis and practice
within ML communities at the level of metrics optimization
in ML design.
LinkedIn example
Problematization refers to the general activity
of constituting problems within a given domain
of thought-in our context, an activity of problem
IEEE Technology and Society Magazine
IEEE Technology and Society Magazine - September 2021

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - September 2021
