Database Reference
In-Depth Information
9.4.5 Dealing with Discrimination
Discrimination has a strong association with generalizations related to identity-
related characteristics, such as race, ethnicity, religion, gender, social class, politi-
cal affiliation, and so on, upon which action is illegal. Discrimination, however,
can also be approached from a less-loaded mathematical angle, i.e. being able to
differentiate.
Because in our case the synthesized part of augmented reality produces the
leads to direct police attention, the impact of personal bias has to be eliminated
during model and profile construction. Discriminating tendencies that are nonethe-
less encoded in the algorithms can further be neutralized by using corrective tech-
niques to create discrimination-free classifiers (chapter 14). Most contributing to
discrimination-free selection, however, is the fact that profiles are geared to de-
tecting (time-spatial) behavior, rather than personal or social-economical charac-
teristics (Alpert et al. 2005). Our rational is that being a drug-trafficker is not an
offense: only the act of drug trafficking is. In the endeavor of identifying criminals
red-handedly, reference to a single hotspot, hot moment, or hotshot observation
may be used to strengthen a profile. The profile grows stronger, however, both in
terms of effectiveness and in reduced bias, if multiple observations are used to de-
termine behavioral pattern. Notwithstanding these efforts, behavioral profiling
cannot completely prevent discrimination. For example, drug-trafficking related
behavior, such as cruising a particular route, may also be characteristic to other
(non-criminal) groups (Warren et al. 2006). In such cases profiles may produce
too many false positives. This renders the profile less economical and, thus,
mounts pressure to adjust the profile (which is true, of course, for all profiles).
9.4.6 Dealing with Group-Think
Group-think is a single-minded self-confirming pattern of thinking, not receptive
for conflicting signals of the outside world (Cannon-Bowers et al. 1993; Janis
1972). The risk of group-think when working with augmented realities increases
when feedback on profiles is not organized and subsequently used to update the
profiles, or when a few dominant participants in model construction leave little
room for others to discuss alternative explanations. Group-think shields police of-
ficers from identifying criminal behavior that does not fit their pattern of thinking.
It reduces their creativity, their adaptivity, and, thus, their effectiveness. Measures
to avoid group-think include diversifying the team (also in time), prevent domi-
nant leadership, and building in randomness in the selection process (Cannon-
Bowers et al. 1993). Moreover, police officers involved in creating layers of aug-
mented reality as well as police officers making use of it during their operations
need to nurture a critical attitude towards illegal discriminating bias that may have
crept into their augmented reality. Just like they do in their not-augmented reality.
As described by Thatcher (2005), failing to do so is bound to lead to 'trust-decay'
in police operations.
Search WWH ::
Custom Search