Transparency in Data Mining: From Theory to Practice - Discrimination and Privacy in the Information Society - page 307

disclosure of the information within them (with the exception of the relevant data
subjects). The difficult questions pertain to the nature of transparency regarding
the technical measures for collating these datasets. These are probably best kept
out of the reach of the public eye. Auditing of this process would be carried out
internally, with the help of selected experts.
These conclusions follow from accounting for the elements discussed above.
With the exception of the actual datasets used, I doubt whether any disclosures
made regarding this segment will generate sufficient public interest to “shame”
lower level officials (who will be making most of the technical decisions) into
changing their practices. On the other hand, the “crowdsourcing” objective carries
merit at this juncture, especially regarding technical decisions. However, they
probably could be acheived by engaging in selective disclosure to experts.
The “autonomy” based arguments do not provide substantial insights. Those
premised upon the rights of data subjects (addressed in section 4.2.3) might justify
additional disclosure of these factors - especially regarding the personal
information used in this process. Yet I am skeptical whether this theory (which, as
mentioned suffers from several analytical flaws) can justify the disclosure of
ancillary information regarding the collation and matching process of the analysis.
On the other hand, it is quite a long shot to connect disclosure requirements at this
segment to the autonomy rights of those affected by the data mining analysis -
concerns arising on the opposite side of the information flow (and addressed in
section 4.2.4).
Segment (B) presents more of an analytic challenge. Currently, the public is left
almost entirely in the dark at this stage, in which the data is analyzed and patterns
formulated. This must change. Additional layers of disclosure should be applied to
both technological elements and human and policy decisions.
These arguments can be justified under several theories. Let us separately
approach the technology and policy aspects of this segment. In terms of the
technology used at this juncture, transparency will only serve as a minimal
“check” on governmental actions. The public would have limited interest in these
technical details. Thus, there is only limited potential for effective shaming.
“Crowdsourcing” is the argument which seems to have the greatest force. Both
autonomy based arguments are quite a stretch, as “data subjects” and “affected
individuals” will have a difficult case linking these rights to the actual computer
analysis. In view of the obvious detriments of sharing the technology with external
sources, a possible compromise calls for releasing the software to a selected group
of experts throughout the industry. These experts will be barred from sharing such
code commercially. They, however, would be able to inform the public if hidden
agendas are imbedded within the code. This seems to be a reasonable policy
strategy for disclosure of technology-related information at this juncture.
Moving to the realm of policy decisions, I find that information concerning the
decisions regarding the acceptable level of errors within the process (sometimes
referred to in the technical jargon as “support” and “confidence” (Zarsky 2002-3))
requires greater transparency. The internal balances between accuracy and security
will generate public interest that will prove to be an effective check on
government. These decisions also impact the personal autonomy of those affected