Transparency in Data Mining: From Theory to Practice - Discrimination and Privacy in the Information Society - page 294

information and disseminating it. In others, it calls for the creation of guidelines
and protocols. In the most extreme cases, transparency might call for proactive
research on behalf of the government, which will provide additional insights as to
the processes it carries out and their outcomes. To enable full transparency, the
conclusions drawn out in these studies must be shared with the public.
Therefore, to properly understand the meaning of transparency in this unique
context, the predictive process must be broken down into several segments. To
effectively illustrate this point, this part identifies four distinct segments of the
prediction process. Each such segment generates different transparency
requirements and needs on both the technological and administrative level.
Current scholarship has failed to properly distinguish among these segments. Yet
understanding the different challenges of every segment are the key to resolving
the apparent tension between transparency and the will to implement successful
and acceptable prediction schemes. The next few paragraphs map out these
segments. In addition, they briefly demonstrate the very different meaning of
transparency in every context, and how it might be achieved. In doing so the
analysis emphasizes the three foundations of the process articulated above:
technology, human decisions and overall policy.
Transparency concerns already arise at the first steps of the predictive modeling
process - (a) the collection of data and aggregation of datasets . At this stage,
transparency refers to providing information regarding the kinds and forms of data
and datasets used in the analysis. On its face, such disclosures generate limited
social risks. When these exist, specific secretive governmental datasets could be
excluded. An additional layer of transparency pertains to the human decisions
made during the aggregation and collation stage. Human discretion plays out in a
broad array of crucial stages. For instance, in the way similar records in different
datasets are matched into one source. 8 Transparency at this juncture could be
achieved by providing the working protocols analysts use for these tasks. This
latter task is easier said than done. Clear protocols on the human role in data
aggregation might not exist. Therefore, transparency will call for their creation,
updating and enforcement.
Finally, transparency in this early stage has an additional, more extensive
meaning. It might call for providing access to the data used in the analysis process.
In some contexts, such a right of access already exists, yet to only a limited
segment of the population.
Transparency considerations play a role in the next segment of the analysis
process as well - (b) data analysis . This stage includes both technical and human-
related aspects. The “technical” aspect relates to the technology used in this
process. It could be rendered transparent by disclosing the names of the software
applications used (if they are in commercial use). If these are custom-made,
transparency could be achieved by releasing the source code of these programs.
In the realm of human decision and public policy, transparency requirements
pertain to a variety of elements. It can relate to the acceptable rates of errors in the
8 Studies indicate that this stage is a “major contributor to inaccuracies in data mining.”
Cate, 2008, at 470.