Since its founding in 2011, politicians and police departments have united in lauding PredPol (predictive policing) as the next stage of the “War on Crime.” PredPol is a private company that sells software to corporations and police departments that maps probable crime hotspots; officers can then include those locations during routine patrols. The FBI Law Enforcement Bulletin claims that “each [Los Angeles Police Department division] that implemented the predictive policing software achieved crime reduction […] the computer eliminates the bias that people have.” On its website, PredPol emphasizes that it only uses three data points—crime type, crime location, and crime date and time—to generate 500 square-foot perimeters in which crime is statistically likely to happen. The lack of demographic data, PredPol says, eliminates the possibility for civil rights violations—i.e., discriminatory policing.
PredPol’s cousin, algorithmic risk assessment, has also received substantial attention in recent years. Algorithmic risk assessment uses various data to determine whether individuals are likely to commit another crime after being released from jail or prison. The First Step Act relies on algorithmic risk assessment to determine which federal prisoners can “cash in” on earned-time credits—and with a national recidivism rate of 76.6% within five years, this is an important thing to consider. The plans to end cash bail in California and New York also involve using algorithmic risk assessment to determine whether a person should be held in pretrial detention. Like PredPol, algorithmic risk assessment is promoted as replacing potentially discriminatory human decisions with the disinterested opinion of a computer.
But many articles have pointed out the racist implications of PredPol and algorithmic risk assessment. PredPol’s claims of “place-based policing” are meaningless when physical spaces are still heavily segregated by race and class, and are surveilled accordingly. Algorithmic risk assessments likewise generate false positives that disproportionately rest on people of color. Data collected under racist circumstances will produce more racist circumstances.
However, this might not matter to policymakers who believe crime should be punished; some stakeholders maintain that, no matter the circumstances, crime comes down to personal choice rather than systemic inequities. If PredPol and algorithmic risk assessment enact justice on people who commit crimes, then what does it matter?
The question now comes to: do PredPol and risk assessment algorithms actually advance justice? And how do stakeholders determine that?
It is hard to say, and that is part of the problem. PredPol’s proponents cite both decreases in crime and increases in arrests as indicators of the software’s success. Jackie Wang, author of Carceral Capitalism, rightly points out this circular logic:
When police officers are dispatched to the 500-by-500 feet square boxes marked in red on city maps, are they expected to catch criminals in the act of committing crimes, or are they supposed to deter crime with their presence? The former implies that an increase in arrests in designated areas would be a benchmark of success, while the latter implies that a decrease in crime is proof of the software’s efficacy. However, both outcomes have been used to validate the success of PredPol (Wang 243).
There are no reliable criteria on which to judge PredPol’s efficacy. Furthermore, no independently-produced studies exist that evaluate PredPol’s strategies or validate the data it uses. The public is expected to trust that PredPol’s placement of police in high-crime areas will lead to a decrease in crime—until, of course, real-life evidence shows the opposite.
Algorithmic risk assessment presents a similar issue. People labeled “high-risk” for recidivism are much more likely to be detained pretrial, and consequently convicted and sentenced. If a person labeled “high-risk” recidivates, the risk assessment algorithm would be proven correct; but, if that person does not recidivate, the algorithm was effective in preventing recidivism by suggesting pretrial detention. And as with PredPol, assessments like COMPAS do not have any data to support their claims outside of their own internally-produced metrics.
The lack of independent verification of PredPol and algorithmic risk assessments’ methods and outcomes is startling. And, ultimately, both rely on the assumption that crime is inevitable, and all that can be done to stop it is through more police, more jails, and more data. But data do not capture societal factors like poverty or unemployment, which determine recidivism as much as, if not more than, individual risk factors—if they did, perhaps policymakers would move beyond moderate prison reforms, and instead turn toward holistic approaches to crime and justice.
One need not look far to find successful examples of prevention-based approaches to crime that do not rely on increased policing, as PredPol does. One example can be found in Canada, which, despite having its own struggles with racist police violence and disproportionate incarceration of Black and Indigenous people, has made strides toward community-based crime prevention. Canada’s National Crime Prevention Strategy (NCPS), established in 1993, has delivered consistently lower crime rates while reducing incarceration and maintaining a relatively constant police population. The NCPS is based on two premises:
The first is that well-designed interventions can positively influence behaviours that lead to crime, especially among youth. The second is that crimes can be reduced or prevented by addressing risk factors that can lead to offences (“2017-2018 Evaluation of the National Crime Prevention Strategy”, 1).
The NCPS provides grants to community-based crime prevention programs that work to alleviate poverty, homelessness, and other risk factors. From 2014 to 2015, 83% of NCPS-funded projects “reported a decrease in charges among targeted populations as a result of program participation,” (“2017-2018 Evaluation”, 17). While statistics regarding long-term impact are unavailable due to the 5-year time limit of NCPS program grants, Canada’s approach to crime is as sustainable as it is effective—and, importantly, is based on independently-produced data.
There are also alternatives to imprisonment that go beyond current calls to end cash bail and release nonviolent drug offenders. In her brilliant opinion piece for The New York Times, Michelle Alexander proposes restorative justice as “a meaningful pathway to accountability without perpetrating the harms endemic to mass incarceration.” Restorative justice takes on a survivor-centered approach to violence, emphasizing accountability over punishment. Common Justice, a Brooklyn-based nonprofit organization, has achieved a 93% success rate in preventing recidivism through facilitating reconciliation between harmed and responsible parties. Common Justice has also promoted healthy responses to trauma that secure long-term emotional wellbeing: survivors who enter restorative justice processes are half as likely as those who go through the criminal justice system to display signs of post-traumatic stress. Incarceration, on the other hand, neither satisfies survivors of violence nor prevents those responsible from committing violence again.
With all this in mind, stakeholders should seriously consider the risks of implementing policies that rely on ineffective approaches to crime and justice. PredPol markets itself as an innovation, when it simply entrenches harmful policing practices in a computer program. Algorithmic risk assessment falls into the same trap: while purported to reduce economic strife and recidivism, it still relies on detention and surveillance as a means of mitigating harm. As our criminal justice expenditures continue to balloon, reaching over $182 billion in 2017, with little impact on crime and recidivism, it is time for policymakers to pursue evidence-based, sustainable alternatives to policing and imprisonment.