Documente Academic
Documente Profesional
Documente Cultură
R
iding high in their squad car, offi- of aging, pock-marked Victorians on the But starting next month, Pascucci and
cers Jamie Pascucci and Joe Kania east side. Young, white officers from outside Kania may get a new type of guidance.
are cruising the neighborhood of the neighborhood, Pascucci and Kania pa- Homewood is set to become the initial
Homewood, scanning the streets trol using a mixture of police radio, calls to pilot zone for Pittsburgh’s “predictive po-
for trouble. Pittsburgh, Pennsylva- their department’s communications center, licing” program. Police car laptops will dis-
Caption goes here. Nat ad quae
nia, has one of the highest murder and instinct. They get occasional help from play maps showing locations where crime
rem quidest, qui odicatiorem quias
rates among large U.S. cities, and ShotSpotter, a network of sensors that de- is likely to occur, based on data-crunching
sitia iliquiant rerum sequam et.
violent crime is particularly severe tects gunshots and relays the information to algorithms developed by scientists at Carn-
in Homewood, a 98% black pocket a laptop mounted between the front seats. egie Mellon University here. In theory, the
Published by AAAS
NEWS
Officer Shane Kovach patrols Homewood, making crime forecasting “a real possibility
a Pittsburgh, Pennsylvania, neighborhood where rather than just a theoretical novelty,” says
predictive policing will soon be introduced. UCLA anthropologist Jeffrey Brantingham.
LAPD was using hot spot maps of past
focused strategies—as part of a palliative for crimes to determine where to send patrols—
policing’s ills. a strategy the department called “cops on
But civil liberties groups and racial justice the dot.” Brantingham’s team believed they
organizations are wary. They argue that pre- could make the maps predictive rather than
dictive policing perpetuates racial prejudice merely descriptive.
in a dangerous new way, by shrouding it in Postdoctoral scholar George Mohler, now
the legitimacy accorded by science. Crime a mathematician at Indiana University-
prediction models rely on flawed statistics Purdue University, Indianapolis, suggested
that reflect the inher- that borrowing models
ent bias in the crimi- from seismology might
nal justice system, they be useful. Earthquakes
contend—the same type “They’re not predicting take place at a relatively
of bias that makes black the future. What they’re fixed rate along existing
men more likely to get fault lines, but quakes
shot dead by the police actually predicting is can also occur in clus-
than white men. Privacy where the next recorded ters, when an initial
SCIENCE sciencemag.org
Corrected 6 October 2016. See full text. 30 SEP TEMBER 2016 • VOL 353 ISSUE 6307 1485
Published by AAAS
NEWS | F E AT U R E S
Data sets
Predictive policing starts
Response
with large amounts of data
on past crimes and factors
that may infuence crime.
Software analyzes the social An algorithm identifes People are told they're
10
0
11
01001000 01100001 01110110 0110010100100000 01000110 01110101 01101110 00101100 00100000 010011000
Predicting ofenders or victims contacts of those that have probable perpetrators considered at risk, or social
policing would lead to better outcomes. it’s called, was inspired in part by stud- at people and how risky a person is,” he says.
Like PredPol, Pittsburgh’s CrimeScan pro- ies done by Andrew Papachristos, a socio- The list has invited allegations that CPD
gram has a geographic focus, but it draws logist at Yale University. Papachristos grew is veering dangerously close to the flawed
on a broader variety of indicators. Gorr and up in Chicago’s Rogers Park neighborhood “precrime” unit in the sci-fi film Minority
Neill took their inspiration from crimino- in the 1980s and ’90s, at the height of the Report, which taps the premonitions of a
logy research showing that criminals tend to crack era. Being white insulated him from trio of mutated humans to stop potential
be generalists, and they tend to progress from some of the violence, he says: “The color murderers before they act. And in bringing
minor to more serious crimes. As a result, the of my skin meant I never had to join a bad press, the program has contributed to
duo hypothesized, reports of minor crimes gang.” But one night, Papachristos watched the problems of the beleaguered CPD, which
could help predict potential flare-ups of as a gang burned his parents’ diner to a mayoral task force described last April as
violent crime. In a gang confrontation, Neill the ground because they refused to pay having “systemic institutional failures going
says, “maybe it starts out extortion money. back decades that can no longer be ignored.”
with harsh words and Decades later, when he Papachristos—who is not involved with
offensive graffiti, and started studying crime, the Strategic Subject List himself—cautions
turns into fist fights, Papachristos wanted to “There are some cities that the program overemphasizes both
which turn into shoot- understand the networks where they have done an individual’s potential to offend and the
ings, which turn into behind it. For a 2014 pa- use of policing, rather than other services,
lots of shootings.” Along per, he and Christopher a great job on hot spot to fight crime. That “reinforces the way in
with observations from Wildeman of Cornell policing, and they have which America devalues the lives of young
the recent past, Crime- University studied a high- people of color,” he wrote in the Chicago Tri-
Scan incorporates scores crime neighborhood on terrible relationships bune on 1 August.
of minor crime offenses
and 911 calls—about
Chicago’s West Side. They
found that 41% of all gun
with their communities What’s more, the police data that this
and other predictive policing programs rely
things like disorderly homicide victims in the of color.” on are skewed toward crimes committed
conduct, narcotics, and Cameron McLay, Pittsburgh community of 82,000 by people of color, says William Isaac, an
loitering—to spit out Bureau of Police belonged to a network of analyst with the Human Rights Data Analy-
predictions about city people who had been ar- sis Group and a Ph.D. candidate at Michi-
blocks likely to see upsurges in violent rested together, and who comprised a mere gan State University in East Lansing. That
crime in the next few days or weeks. 4% of the population—suggesting, with other renders any predictions suspect, he says:
studies, that much can be learned about crime “They’re not predicting the future. What
THE CHICAGO POLICE DEPARTMENT (CPD), by examining the company people keep, they’re actually predicting is where the next
meanwhile, has taken predictive policing Papachristos says. recorded police observations are going to
one step further—and made it personal. Intrigued by these ideas, the Chicago po- occur.” Predictions, indeed, can become self-
The department is using network analysis lice teamed up with Miles Wernick, a medical fulfilling prophecies, says Jennifer Lynch of
to generate a highly controversial Strate- imaging researcher at the Illinois Institute the Electronic Frontier Foundation in San
DIAGRAM: G. GRULLÓN/SCIENCE
gic Subject List of people deemed at risk of Technology in Chicago, to develop the Francisco, California. “We know from past
of becoming either victims or perpetra- Custom Notification program. Because gang examples that when police are expecting
tors of violent crimes. Officers and com- violence was distributed across the city, hot violence, they often respond with violence.”
munity members then pay visits to people spot policing wasn’t as effective in Chicago, Brantingham, the architect of Pred-
on the list to inform them that they are says Commander Jonathan Lewin, head of Pol, agrees that civil liberties concerns
considered high-risk. technology for the department. “The geo- “are really important questions.” But he
The Custom Notification program, as graphy of the map isn’t as helpful as looking says that predictive policing can be more
Published by AAAS
fair than the status quo: “What’s often periment was conducted by the Shreveport, PERCHED AT A CONFERENCE TABLE over-
forgotten is that any time you put an of- Louisiana, police department in 2012 with looking the blighted Allegheny-West
ficer in the field there’s a risk of civil NIJ funding. The study found that the dif- neighborhood, Chief McLay says he is
liberties violations.” ference in crime reduction between the keenly aware that rolling out CrimeScan will
Other critics, meanwhile, raise a more control and experimental districts was sta- not solve all the Pittsburgh department’s
fundamental question about predictive tistically insignificant. But the experiment, problems. “There are some cities where they
policing: Does it even work? which focused on property crimes, also re- have done a great job on hot spot policing,
vealed the challenges of such studies. Take- and they have terrible relationships with
IN A 2012 IBM SMARTER PLANET commer- up of predictive hot spot policing among their communities of color,” he says.
cial, a police officer glances at the screen the three experimental districts was high at The key, some experts say, is not to rely
of his squad car, then speeds to a conve- first, but dropped off after 4 months as en- only on statistical methods, but to com-
nience store. He arrives as a clerk is count- thusiasm waned, likely skewing the results. bine them with other approaches. For ex-
ing money, and moments before a would-be Commanders in one of the control districts, ample, Papachristos is now working with
robber shows up. That’s science fiction, says meanwhile, grew excited by the experimen- the Chicago Violence Reduction Strategy,
RAND’s Hollywood—and likely to stay that
way. To predict specific crimes, he says, “we
would need to improve the precision of our
predictions by a factor of 1000.”
As to whether existing methods of pre-
dictive policing work as advertised, by fore-
and for policymakers, we need to have complex nonlinear relationships between used to solve problems rather than simply
some insight into what’s going on so that input data and output data.” (One exam- focus police work. “Therein lies the key:
it can be validated by outside groups,” Isaac ple is optical character recognition, which Who finds that sweet spot?” he says. “Who
says. But Brantingham says researchers can is used for digitizing printed texts.) With uses just enough data to be really good, and
evaluate the outcome without knowing all crime, he adds, “It’s much more simple—the has the relationships that are just robust
the underlying research. more risk, the more crime. There aren’t re- enough? That’s the challenge that policing
One notable randomized, controlled ex- ally complicated relationships going on.” in this country is facing right now.” j
SCIENCE sciencemag.org
Corrected 6 October 2016. See full text. 30 SEP TEMBER 2016 • VOL 353 ISSUE 6307 1487
Published by AAAS
Crime forecasters
Mara Hvistendahl
PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions
Science (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement of
Science, 1200 New York Avenue NW, Washington, DC 20005. 2017 © The Authors, some rights reserved; exclusive
licensee American Association for the Advancement of Science. No claim to original U.S. Government Works. The title
Science is a registered trademark of AAAS.