International Journal on Criminology Volume 4, Number 2, Winter 2016 | Page 70

International Journal on Criminology - Winter 2016, Volume 4, Number 2
Telling Tales with Inspector PredPol
Xavier Raufer
1. Predictive Policing: Myth and Reality
It is always the same story: a journalist, who is either gullible or has taken a
backhander, is in seventh heaven as he or she reports the news we have all been
waiting for. In the fall of 2015 it was the turn of a certain Kevin, author of a piece on
the French website Science Post boasting: “Crime-Predicting Artificial Intelligence is
in the Pipeline.” Kevin’s article featured the predictable reference to the film Minority
Report, followed by a spate of technical terms intended to mesmerize and daze the
reader. We also learn that we are soon going to be “arresting criminals … before they
commit a crime.” Is this the stuff of science fiction? No, it is tomorrow’s world.
Except for one thing: it is no easier to model uncertainty today than it was when
Aristotle was alive (and our conception of time is still underpinned by ancient Greece).
Let us put it another way: we would not have been able to predict September 11 even
if we had had a document database containing everything we know on terrorism from
the year dot to September 10, 2001. It is a bit embarrassing to have to keep on saying
it, but if yesterday’s known automatically solved tomorrow’s unknown, we would all
be lottery winners…
The gullible journalists who peddle these tall tales on behalf of software dealers
seem to be ignorant about how predictive policing systems work. So, let us give them
a hand: algorithms sift through data on past crimes with the aim of anticipating future
offenses. But—and this is something we really need to stress—this is absolutely
impossible! So, what does predictive software actually do? It works on the following
assumption: since a crime was committed in such-and-such a place yesterday, another
felony might be carried out in the same spot tomorrow. But that is not predictive
reasoning; it is wishful thinking.
And just how credulous are the individuals who spread these fairy tales? In
August 2015, the French newspaper Le Parisien made the bold claim that predictive
policing software in Munich had “cut burglaries by 30%.” But what do we find
when we check the official statistics supplied by the Bundeskriminalamt (the Federal
Criminal Police Office)? Burglaries increased by 16.9% in Germany between
2003 and 2012. How about in 2014, when Munich's famous miracle software was
introduced? The burglary rate did not even level out, rising once more, this time by
1.8%. The “predictive” magic wand simply triggered a displacement effect: once
burglars felt they were being watched, they began plying their trade elsewhere, going
A
CNAM Paris
69