International Journal on Criminology Volume 4, Number 2, Winter 2016 | Page 77
International Journal on Criminology
On the Up and Up: Predictive Justice
Predictive technology truly is the tool for all occasions, even spreading to
the justice system. The judiciary quite rightly asks the following questions: Will this
inmate, who is awaiting release, commit a serious crime in the years ahead? Will
this first-time offender turn out to be a repeat offender? Under normal circumstances,
judges only have their experience or intuition to fall back on. But what if predictive
software could help them make the right choices about detainees? And guide their
decisions about who should be given a suspended sentence or parole without any risk
to society?
Is it possible, in short, to devise an objective system that would increase the
number of detainees given parole and reduce the number of repeat offenders? To
select which inmates should be sent on reintegration programs during or after their
sentences? To decide (in a common law system) on who should remain in prison, and
for how long? And to identify easily-influenced young prisoners who should under no
circumstances be detained alongside hardened criminals?
In the US, four-fifths of the bodies that review parole releases now use predictive
software. What data do they use for assessing inmates? Age, sex, background, and
education, together with the date of the first arrest, prior offenses, and the inmate’s
behavior in prison. The criminal record of the detainee’s closest friends and the results
of psychological tests are also included, and even whether his or her mother drank to
excess while pregnant. All this information is then compared with similar profiles.
The fact that decisions are taken on the basis of this type of calculation may seem
worrying. However, as a criminologist is duty bound to point out, it is a little less
disturbing if you have read “Extraneous Factoring in Judicial Decisions” (Princeton,
February 2011). This frightening report, which compares hundreds of criminal cases,
demonstrated the following: judges in the US tend to hand down more severe sentences
before lunchtime, when they are hungry, than a couple of hours later, when they have
been fed and watered.
Whether it is the vagaries of computer software or the ups and downs of a
judge’s digestive system, it is always a lottery. What is constant is this: everything to
do with so-called policing or justice is based solely on retrospective information. We
are told that “predictive models analyze historical data”, and that “we collect all this
data on past events”—and it is here that the conceptual problems begin.
5. How does Predictive Software Actually Work?
Considering the price—$73,000 to buy the predictive policing software plus
an annual subscription of over $45,000—it is important to understand how everything
works. But once again, the articles published on the subject all give the same
explanations, starting with entering “all the data on crimes committed from the year
X together with their location.” Or: “We analyze the records of around 1.5 million
crimes committed from 2003 to 2012.” Why? Because “future offenses will often be
76