Critiquing Predictive Policing - Mikal Sokolowski
Computer algorithms are extremely pervasive in everyday life. People rely on and use computer algorithms to make their lives more convenient. As an example, when you are waiting for a bus and check the arrival time, using map software to find the quickest way to your destination, or even when you google something, computer algorithms are all being utilized. A recent phenomenon is seeing algorithms be imported in the criminal justice system through the introduction of predictive policing. Although there is no clear definition on what predictive policing is, it is best described by Albert Meijer & Martijn Wessels in their 2019 article, Predictive Policing: Review of Benefits and Drawbacks, where on page 1033 they state that predictive policing is, “the collection and analysis of data about previous crimes for identification and statistical prediction of individuals or geospatial areas with an increased probability of criminal activity to help developing policing intervention and prevention strategies and tactics.” In other words, predictive policing is a computer algorithm that takes previous crime data and processes it to predict where crimes will take place, when crimes will take place, and who will be involved in these crimes.
This is reminiscent of the 2002 film starring Tom Cruise, “Minority Report” (that is based on the Philip K. Dick book of the same name) where a special police unit would arrest would be murderers before they commit murder. Predictive policing is used by law enforcement as a way to utilize their staff more efficiently and deploy police officers where they believe a crime will take place.
It is clear to see the benefits of predictive policing as it can lead to a decrease in the staff needed by police forces because they would be more efficient with their resources. Additionally, predictive policing can improve safety by deploying officers more precisely to areas where a crime is likely to occur. However, there are some potential drawbacks to predictive policing such as a lack of transparency, the potential further utilization of biases in policing, and discrimination issues. This BLAWG will view predictive policing through a critical lens by considering advantages and disadvantages of predictive policing and then consider the legal implications of predictive policing. To begin, predictive policing adoption in North America will be reviewed.
Predictive Policing Adoption
Predictive policing has predominantly been used in the United States and was first adopted by the Santa Cruz Police Department in 2011. The Santa Cruz Police Department used years of home burglary data to predict the areas that future burglars will target for their next crime. This program primarily focused on previous offenders of burglaries and the predictive policing efforts focused on the top 300 burglars in the geographical area. Since then, there has been widespread adoption of predictive policing throughout the United States as major cities in California, South Carolina, Washington, Tennessee, Florida, Pennsylvania, and New York have purchased predictive policing software and police forces throughout the United States are rapidly adopting predictive policing programs. Predictive policing is also becoming more sophisticated with its increased adoption. As an example, larger cities like Los Angeles and Chicago have adopted more advanced predictive policing measures that include complex social network analysis with social maps that link people together, such as friends and gangs, to better catch future perpetrators.
In comparison to the United States, Canadian police forces have adopted predictive policing sparingly. Only three police forces in Canada are known to have utilized predictive policing software and those are the Vancouver Police Department, the Calgary Police Services, and the Saskatoon Police Services. The Saskatoon Police Services’ predictive policing focus has primarily been on identifying children and youth who are at risk of going missing.
However, predictive policing is being considered by several police departments throughout Canada to varying degrees, including in Toronto. The list of police forces that have considered adopting these measures is likely much larger as Canadian police forces generally do not disclose whether they are looking at implementing predictive policing or when they do implement predictive policing. Moreover, police departments often launch these predictive policing measures without public notice, making it difficult to know which police forces have started the process of adopting predictive policing measures. Overall, it seems that Canadian police forces have been more reluctant in adopting predictive policing in comparison to their United States counterparts. However, there clearly is still an interest in adopting predictive policing by Canadian police forces and the trend seems to be going towards more police forces adopting these measures in the near future.
It is important to note that there are different vendors that offer predictive policing software’, meaning that there are differences in the way in which they operate. Additionally, since predictive policing is a relatively new concept, the research on predictive policing is not extensive. This difficulty is further expanded by the consideration that many of these predictive policing software’s are purchased from private organizations, meaning that the intellectual property is owned by a third party and that the information is not generally released to the public.
Advantages of Predictive Policing
As mentioned previously, the main benefit of predictive policing is that it increases the efficiency of the police force, which can in turn lead to more arrests, improved safety, and savings on staffing costs. The two ways that predictive policing attempts to do this is by forecasting (1) where and when criminal activities are likely to occur and (2) who will be involved in these crimes, including both the perpetrator and the victim.
To predict the where, predictive policing analyzes copious amounts of historic crime data that then identifies hot spot areas and performs risk terrain analysis to predict where criminal activity is likely to occur. The historical data is also utilized to predict when criminal activity is going to occur through spatiotemporal analysis where the predictive policing software forecasts at what specific time and days that a crime is projected to occur. Moreover, a variety of other inputs are considered by the predictive policing software to better predict crimes. As an example, a study found that weather conditions could lead to an increase in criminal activity and the software can use this as an additional factor to consider when making predictions. These predictions assume that future crimes are likely to occur in the vicinity and during the time that crimes have taken place in the past.
Predictive policing also aims to identify the individuals that will be involved in crimes, including both the perpetrators and victims. This can be done by predictive policing software utilizing inductive profiling to identify who will be involved in a crime by considering violent group affiliation, re-offenders, behaviours of individuals, and connections to people who perpetrated crimes previously. Like the where and when, predicting the who can input special considerations that can assist in predicting future perpetrators of crimes. As an example, some predictive policing software can include opened-source social media posts and consider criminal communication that is posted on social media websites.
Proponents of predictive policing also assert that since predictive policing is based on historical empirical data, it is free from human biases. This is of course an extremely attractive aspect of predictive policing to police forces, who are often accused of being biased towards marginalized groups. This is exacerbated by recent movements such as “Black Lives Matter” that called for police budgets to be reduced through the “defund the police” campaign. It is clear why reducing police biases would be something that police forces all over North America would like to achieve; predictive policing offers a way in which these biases can potentially be reduced.
The above “selling points” of predictive policing seem positive and it is evident why police forces would consider adopting predictive policing. However, it begs the question of whether predictive policing software reduces crime? Due to many factors that are at play in criminal justice, it is difficult to isolate predictive policing’s effect on crime rates. Nevertheless, police forces that have adopted predictive policing are asserting that it is effective at reducing crime. For instance, the New York City Police Department that implemented an advanced predictive policing system found that the overall crime index of New York decreased by 6% after implementation; they qualified the system as a success. Closer to home in Canada, officials of the Vancouver Police Department stated that in the first six months that the predictive policing measures were implemented there was a 20% decline in burglaries. Similar statistics have been reported by police forces that have adopted predictive policing measures throughout North America, echoing that these measures are effective.
Disadvantages of Predictive Policing
The advantages make it understandable why there has been a rapid adoption of predictive policing in the United States. However, there are several disadvantages that have led police forces in places such as Canada to be more cautious in their adoption. These disadvantages stem from:
predictive policing software being offered by third party organizations;
problematic data that is inputted into the software; and
police departments claiming that they are using an objective method.
Predictive policing software is generally sold and operated by third party organizations that analyze the crime data and then distribute the results back to police departments. This can create several problems on its face, such as:
law enforcement not getting to the root of the problem;
the public not being informed of the model used; and
law enforcement not interpreting the information properly.
Firstly, the information that is presented to the police department is data driven and does not get at the underlying theory of: what actually causes the crime, choice of area, or the perpetrators of the crimes. This will further focus the justice system on simply catching criminals without getting to the deeper meaning of why these individuals commit crimes or why criminal activity is more prevalent in certain areas. This can further exacerbate biases in the justice system.
Secondly, because the software is generally owned by third parties, the algorithms are proprietary; that they would not be subject to public scrutiny, undermining transparency in the justice system. This can potentially lead to greater distrust of police departments, especially among people that the predictive policing software targets.
Lastly, since the results of the data are simply shared with police forces, these police forces can potentially make their own inferences about the data or might not in fact comprehend the data that is presented to them. These inferences can potentially lead police forces to make decisions not based on the algorithms but based on their intuition in which they can implement their own biases and beliefs. This will also undermine the effectiveness of the software.
Another potential disadvantage of predictive policing is the issue regarding the data that has been input into these algorithms. For starters, crime statistics that are compiled by police forces have many concerns about accuracy and completeness. This incomplete data will further erode the usefulness of predictive policing. There are also concerns with the data itself reflecting the biases that some police officers may have; inputting this data will thus only allow the police to further entrench these biases. In Ontario, police stop data has been reported to disproportionately affect Black, Indigenous, and other racialized individuals and this data would be included in predictive policing data, meaning that it itself would potentially target these races or neighborhoods that predominantly house these races. It is also noteworthy that arrest data is frequently included in predictive policing, which does not actually prove that a crime was committed by an individual and which disproportionately affects racialized individuals. Other concerns regarding police data include:
predictive policing not including under-reported crimes and communities that do not report crimes for fear of law enforcement;
a lack of reporting of undocumented crimes like sexual assault;
‘gang database data’ being problematic and found to target minority groups;
reliance on public reports that obtain questionable data; and
data being rarely corrected and verified for things such as wrongful convictions.
All of these problems are exacerbated by the fact that police departments can point to the predictions and say that they are using objective data by interpreting them and therefore are free from any biases. Several points discussed above make this problematic, such as that the data can itself be biased and that the police departments may not know how to interpret the data properly. In other words, police departments can virtue signal that they are using objective information, when in fact that information still has biases and can be interpreted by the police forces in a biased way. Moreover, predictive policing can serve to confirm the biases that police departments have and the work they do.
Overall, these disadvantages must be considered for any police forces that are looking to implement predictive policing. It is also important to consider these disadvantages with the significant costs associated with implementing predictive policing software including monetary costs of an initial licensing fee of $1.5 million and a continued payment to use this software.
Constitutional issues arising from predictive policing
With any major changes in policing, there will be legal considerations to contemplate and that is no different in this case. Predictive policing touches many areas of the legal environment, but we will consider sections 2, 7, 9, and 15 of Canadian Charter of Rights and Freedoms (“the Charter”) specifically.
The first legal consideration is that predictive policing may undermine the freedoms of expression, association, and peaceful assembly as protected by section 2 of the Charter. Recent empirical evidence has revealed that government online surveillance creates a chilling effect on citizens, making them likely to engage in certain legal activities and exercise greater caution. Recent developments such as predictive policing also undermine the right of people to freely assemble for events such as protests where people can be identified as attending a protest or for wearing a mask during a protest – which is a criminal offence in sections 65 and 66 of the Criminal Code (“the Code”) and would in turn play a factor in the predictive policing algorithm. Similarly, predictive policing could create a chilling effect on peaceful protests that are an important component of a free and democratic society.
Additionally, because predictive policing software incorporates a person’s online presence, use of predictive policing by police forces can increase the risk that individuals will self-censor themselves to mediate the risk that they are being monitored. This will particularly affect marginalized communities in self-monitoring what they say online in fear of having government surveillance watching their communication.
Another legal consideration concerns section 15 of the Charter that offers freedom from discrimination and the right to equality. As discussed in the disadvantages of predictive policing above, predictive policing can perpetuate biases and stereotypes that police forces have, leading to discrimination against minority groups through already biased data. This is further complicated by the fact that police forces can claim that reliance upon such data is an objective measure that is free from bias. A 2019 study in New Orleans and Chicago found that the data that is used in predictive policing software was used at a time when government documentation included racial biases, corruption, and illegal practices, leading to the data being considered “dirty”. It was then theorized that this creates a feedback loop in which such dirty data continues to perpetuate the biases that the data was initially based on. Additionally, when police officers are deployed in these “high risk” areas, it may make them mentally inclined to perform stops or use excessive force in order to find the crimes in these areas.
Predictive policing data is also perceived to disproportionately target areas that have a higher proportionality of marginalized groups and more socioeconomically disadvantaged communities. Similar to what was discussed above in analyzing predictive policing’s potential violation of section 2 of the Charter, it can be theorized that it can cause a chilling effect among these communities and it can potentially further erode police trust in these communities. In sum, it is likely that, if predicted policing were to be adopted in Canada it would likely receive legal scrutiny based on the right to equality and discrimination outlined in section 15 of the Charter.
The last Charter sections that will be considered are sections 7, that protects life, liberty, and security of the person, as well as section 9, that protects the right to arbitrary detention. This ties into the previous paragraph that mentions the fact that police officers deployed in areas highlighted by predictive policing will be looking for an arrest. This can especially be the case for re-offenders, as mentioned in the advantage section, since predictive policing attempts to predict who will commit the crimes and flags past criminals in the area. This can potentially undermine the justice system by putting formal criminals under the microscope and discounts former criminals being reintroduced into society while going through the rehabilitation process.
Other legal considerations that predictive policing also touches on are the right to privacy, right to due process, human rights violations, the intersection of private business and the judicial system, and many more. These legal issues should all be considered in great deal prior to any police force adopting predictive policing, otherwise police forces may be on the hook for damages or redress from utilizing predictive policing.
Steps in the right direction with much work to be done
From the considerations above, predictive policing is considered a divisive controversial policing method, even considering the purported benefits that it offers. Although predictive policing has not been fully adopted in Canada, it has seen rapid adoption in the United States over the last ten years, which is understandable considering the benefits that predictive policing potentially offers. These benefits include predicting when, where, and who will be involved in crimes based on empirical data. Moreover, the reduction in crime rates once predictive policing has been implemented are staggering. However, there are several concerns with the adoption of predictive policing, including that it is generally operated by third parties. This reality means the police may not understand how to properly interpret the data and the data may not be subject to public scrutiny. Moreover the data utilized by police forces may itself be incomplete, inaccurate, or biased.
Considering these disadvantages, there are several legal concerns that arise from its use, such as predictive policing potentially violating sections 2, 7, 9, and 15 of the Charter, and other legal considerations that should be analyzed prior to the adoption of predictive policing software. Overall, there is a room for predictive policing software in the criminal justice system, but it should be adopted in a way that minimizes the disadvantages and legal considerations.Canada is still early in its adoption of predictive policing; it is possible that the slow-moving legal system can preemptively prepare for predictive policing to be adopted in Canada. However, this is increasingly unlikely, and the legal system will likely be reactive regarding predictive policing. Police forces in Canada that have adopted predictive policing measures have observed and learned from the mistakes that police forces in the United States have made. As an example, the Saskatoon Police Services has partnered with the University of Saskatoon to create an in-house algorithm to address concerns of transparency and to have the ability to review, adopt, and defend the software. Similarly, the Vancouver Police Department has attempted to mitigate biases by removing reports that are done by police officers from the data and only using verified data for its breaking-and-entering model. These efforts are steps in the right direction in the adoption of predictive policing in Canada. However, much more should be done, such as independent audits, verification of data, greater transparency, and education on utilizing the software. It will certainly be interesting to see how predictive policing develops in Canada and it should be on the radar of current and future criminal lawyers.