International Journal on Criminology Volume 4, Number 2, Winter 2016 | Page 78

Telling Tales with Inspector PredPol
committed at the scene of an earlier crime.” As a criminologist, this makes me sit up
and take notice.
And so to the algorithms. The docile journalists or ad execs who write
articles on predictive policing parrot the same story: “The algorithms understand (i.e.
‘assimilate’) criminal patterns and generate a prediction.” The criminologist in me sits
up again.
All this proves (claim the predictive devotees) that prediction is a serious
business, since these famous “algorithms are based on earthquake prediction models”
or (an alternative version) “on models that predict aftershocks.” This so-called “proof”
of the software’s effectiveness is repeated time and again, article after article. And now,
as a criminologist, I am falling off my chair because I know (and will demonstrate
below) that at present it is absolutely impossible to predict earthquakes, something
that none of the aforementioned fairy-tale writers took the trouble to check.
Of course, none of this has stopped some South American countries from
embarking on the predictive odyssey. Engineers in Chile, we learn, have been
“combining criminology and mathematical modeling”, and have begun predicting the
hot spots along the country’s land border (which, at 6,170 km, is enormous) where
crime and illegal migration will occur. In Brazil, the city of São Paulo acquired
Microsoft’s Detecta system in spring 2014, which helps fight crime by “aggregating
data” and “creating automatic associations.” The police are kitted out with laptops,
tablets, and smartphones so they can access the system and organize preventive-based
patrols.
But why should preventive tactics be confined to the internet? Patrolling
the social networks may also be a way of predicting crime. How? The Predictive
Technology Lab at the University of Virginia claims that Twitter can “predict” certain
types of offense. In the March 2014 edition of the science journal Decision Support
Systems, the laboratories’ research team explains that geo-located Tweets (whose
location can be clearly identified) are capable of predicting between 19 and 25 types
of offense, including harassment, robberies and assaults. How can that be possible?
“If enough Twitter users says they want to get drunk in the same neighborhood, we
can ‘predict’ the alcohol-related offenses”, say the authors. These Twitter analyses are
then compared with “historically high concentrations of criminal acts.” Does this type
of operation have anything whatsoever to do with prediction? Not at all, as we will see
below.
In summary, predictive analysis operates as follows: data mining (looking for
data on the internet that is often hidden) plus statistics plus sophisticated algorithms
and special software (mining tools) results in modeling and (it is claimed) predictions.
• Is the Software for Real?
Answering this question properly means providing the reader with some relevant
background information:
77