Transparency in Data Mining: From Theory to Practice - Discrimination and Privacy in the Information Society - page 300

regarding segment (d) - the feedback process following the automated/predictive
steps. The public would be interested in the overall success of the project, as well
as in systematic errors and failures. Yet it might ignore the more technical aspects
of the dynamics. Therefore, for every one of these segments, this transparency
justification carries a varying level of persuasive force.
Second, for “shaming” to have an effect, the “shamed” must respond to it (or be
deterred by the prospect of such disclosure). This dynamic will fail, for instance, if
public opinion does not associate the specific decision maker with the relevant
action. In such a case, transparency will not necessarily promote fairness and
efficiency. The nature of automated prediction leads to the fact that many
important decisions will be made by lower level analysts and IT experts -
especially in stages (a), (b) and even (c). Shaming might not have the needed
effect on these officials. They might already be at another position at the time of
revelation, or not clearly and directly indicated. Again, shaming seems to have a
limited effect on the more technical elements of this project, questioning the
wisdom of transparency regarding these factors.
Finally, a third underlying assumption which flows from the two already
mentioned states that shaming will work well when transparency reveals official
conduct that conflicts with well established norms or existing laws. For instance,
sloppily constructing the data mining process and operating it, will easily generate
backlash. This will counter the accepted norm that governmental work must be
carried out with precision and accuracy It can also prove effective if transparency
revealed that rather than relying on neutral factors, officials reverted to relying upon
“sensitive factors” (either knowingly or unknowingly) such as race and religion -
practices which are socially unacceptable. Yet shame might not prove helpful in
other important instances, where social norms have yet to formulate. In such cases,
disclosure will not lead to “chilling” unwanted governmental conduct. For instance,
there has probably yet to emerge a social norm regarding accepted and non-accepted
measures of data collation and levels of acceptable in this process.
Returning to the segments of the data mining task drawn out above again shows
different outcomes for different segments. Much of the information available in
stages (c) and (d) falls within this category. The risks of false positives and the
forms of correlations used are currently within “gray areas” of social norms.
Transparency could be an important measure to promote a discussion on these
issues. However, it is questionable whether this context will generate shaming,
which will act as an effective “check” on governmental actions.
To conclude, shaming acts as an effective “check” in instances where decisions
are made by high ranking officials and clearly counter social norms held by a
broad segment of the population. It is also helpful if the practices at hand are
understandable, or at least easily built into a convincing narrative. In all other
contexts, a shaming-based transparency theory might be unable to justify the costs
and detriments it generates. This distinction will prove helpful in formulating a
general blueprint of transparency policy for the data mining context.
Transparency and Crowdsourcing
Transparency might enhance the accuracy and fairness of predictive models in
a very different way. Rather than incentivize effective governmental actions,