From Data Minimization to Data Minimummization - Discrimination and Privacy in the Information Society - page 270

data in a given context and expects reasonably that it will be processed in this same
context, at the risk of it being judged “out of context”.' 37
Finally, contextuality is important in assessing the value of the outcomes of the
data mining process, either in patterns, profiles or concrete decisions. This may be
especially important since, as has been said, automatically processed profiles and
decisions usually do not evaluate the outcome and result of the data mining
process in specific contexts, effecting specific individuals. Again, there is a ten-
dency in knowledge discovery in databases to disregard the context of data.
The tendency in data mining processes to disregard the context of data are ag-
gravated by the use of data minimization techniques 38 and cannot be addressed if
stuck to this principle, since what is needed is gathering a minimum rather than a
minimized amount of data, the data must be updated every now and then, which
requires a continued search for data, and the context in which the patterns, profiles
and rules acquired by data mining are applied must be evaluated after the process
is done. 39 Although the principle of data minimization aims at excluding or at least
minimizing the risk of privacy and discrimination problems, it may sometimes on-
ly aggravate these problems.
For example, if police surveillance mostly takes place in particular neighbour-
hoods with a lot of immigrants or ethnical minorities, then the gathered data about
criminal activities would be heavily tilted towards these groups in society. Incor-
poration of the methodology of the research in the metadata is thus essential to
avoid discrimination and stigmatization towards these minorities. 40 Furthermore,
not keeping data accurate and up to date may lead to privacy and discrimination
problems. If a person has decided to quit smoking, but a cigarette company keeps
on profiling a consumer as a smoker, this might violate his autonomy and privacy.
Subsequently, the data mining and harvesting process must respect the context
of the data. First, disregard of the purpose for which the data were gathered, the
purpose limitation principle, may not only lead to a loss of the contextuality of
data, but may also undermine the autonomy of the individual as his informed con-
sent with regard to data processing for a specific purpose is transgressed. 41
Secondly, data minimization is not always able to exclude privacy violating or
discriminatory results 42 given the redlining effect. 43 Data minimization not only
offers no adequate solution in this respect, it might also make it difficult to assess
whether a rule is indirectly discriminating or privacy violating. 44
Finally, during the stage in which the acquired patterns and profiles are used in
practice it is vital to assess the context in which they are applied. Even although
37 Poullet & Rouvroy (2008), p. 10 & 14.
38 Guzik (2009); Müller (2009).
39 The only principle that safeguards the contextuality in data mining that is not in tension
with data minimization techniques is the purpose limitation principle, which both limits
the use of data and ensures that the context of the data is retained.
40 Custers (2004).
41 Taviani (2004).
42 Calders & Verwer (2010); Ruggieri, Pedreschi & Turini (2010).
43 Calders & Verwer (2010).
44 Pedreschi, Ruggieri & Turini (2008).