Transparency in Data Mining: From Theory to Practice - Discrimination and Privacy in the Information Society - page 309

However, causation has its downsides. It will slow down the overall process
and render it cumbersome and possibly inefficient. Furthermore, disclosure of
such theories might open the door to serious privacy and stigma concerns.
Causation (in addition to mere correlation) will strengthen the negative stigma
attached to those indicated by the prediction model. This might even follow from
developing such theories and examining them internally, in view of potential
leaking of such information to the general public. Balancing these concerns leads
to recommending that a mandatory causation requirement is unnecessary. In
addition, human imagination could probably find causation at almost every point.
Thus, its effectiveness as a “check” on governmental actions could be seriously
doubted.
Segment (D) is a crucial (yet often overlooked) element in the policy discussion
of achieving transparency in a data mining process. Disclosure at this stage must
be enhanced by proactive governmental research. Almost all of the theories
mentioned above indicate such an outcome. Incentives in accordance to the first
theory (“effective policy” or “shaming”) are especially strong. The issues
addressed in this segment are those that are most likely to gain public and political
traction - false negatives in the project, the actual success of the program and
studies regarding its inner dynamics and its effects on minorities and weaker
segments of the population. Autonomy would also be enhanced if individuals will
know the process which impacted them or used their personal data is overall
successful, and thus worth their personal sacrifice. Thus, the government must
initiate studies examining the impact on minorities. These should join studies as to
the level of false negatives, and overall whether the data used is helpful in
predicting human behavior.
17.6 Coda: The Limits of Transparency
Transparency is hailed as an important policy tool which could enhance autonomy
and forward democracy. Its role in the age of information technology has yet to be
firmly established. This chapter takes initial steps in setting forth a comprehensive
mapping for meeting the transparency challenge in a specific context - that of
predictive data mining of personal information.
While acknowledging the important strengths of transparency, it is crucial to
recognize that there is much harm that governmental prediction models could
generate, and transparency alone cannot cure. For instance, one must question
whether allowing the government to obtain a powerful tool, which can generate
such insights, is wise. Additional powerful arguments set forth are that the process
is ineffective, ridden with errors, generates chilling effects, leads to unfair
discrimination and is prone to facilitate function creep. Transparency provides a
partial response to these problems. For instance, enhanced disclosure might chill
the government from expanding data mining initiatives into unacceptable realms.
Additional work must establish how effective a cure transparency is to the various
ills mentioned, and what other steps must be taken. For this reason, the analysis
here presented is an essential, yet certainly not a final step. I hope, however, that