Predictive Policing and AI in Canada

Featured in Robson Crim
4 hours ago
8 min read

By AK

Introduction

This blog explores how law enforcement agencies across Canada are increasingly relying on Artificial Intelligence (AI) to both prevent and target crime across the country. Specifically, this blog focuses on two emerging technologies within AI–predictive analytics and facial recognition and it explores how law enforcement agencies have utilized these innovative forms of policing to both predict crimes before they happen and solve ongoing criminal cases. At the same time, an important policy question arises: can AI-enabled policing tools improve public safety without reproducing bias or eroding privacy rights? Thus, this blog considers the significant ethical and privacy concerns relating to these technologies, all of which demand careful oversight from police forces, policy makers and government.

AI-Driven Predictive Analytics

Predictive policing is a form of AI which uses mathematics and algorithms to predict where future crimes may occur.[i] In simpler terms, predictive systems analyze historical crime data and identify patterns that may indicate where similar crimes could occur again. Furthermore, it is revolutionizing how law enforcement targets crime, especially in urban areas. In Vancouver, for instance, the Vancouver Police Department (VPD) introduced a new predictive technology, called GeoDASH, designed primarily to target property crime.[ii] This technology, the first of its kind in Canada, allows the VPD to “forecast the location of property crime and take proactive measures to prevent it.”[iii] Moreover, it proactively predicts when and where break-ins may occur, which then allows the VPD to send patrols to those specific areas.[iv] These areas become known as hot spots.[v] This technology, which is now available to all officers in their cruisers, was first introduced after the city engaged in a pilot project.[vi] Early results from the pilot suggested measurable effects. The pilot proved to be largely successful, as it resulted in a “substantial decrease in residential break-and-enters.”[vii] In particular, it decreased property crime by 27 percent in the areas where it was tested.[viii] One of the key elements of this AI software is that it uses past crime trends, and gathers records of break-ins around the city, in order to generate predictions on where they will likely occur in the future.[ix] This has proven to be significant in a city like Vancouver, where property crime is fairly common.[x] Yet, the Police Chief of the VPD, Adam Palmer, noted that the goal of this technology is not to arrest a bunch of people, but rather to deter crime before it happens.[xi] As he put it, “We’re not targeting people, we’re targeting locations.”[xii]

Similarly, in Saskatoon, the Saskatoon Police Service (SPS), working in partnership with the Government of Saskatchewan and the University of Saskatchewan, recently launched a Predictive Analytics Lab (the “Lab”).[xiii] This Lab, the first of its kind in Canada, uses AI technology similar to GeoDASH, as it takes complex data on crime and generates algorithms to predict crime.[xiv] However, it is not limited to property crime. Instead, it focuses on identifying patterns across a wide range of incidents, such as assaults and suspicious activity.[xv] Moreover, a vital component of the algorithmic technology is its ability to generate common factors which can contribute to someone going missing (particularly youth).[xvi] This predictive technology allows law enforcement to intervene before crime may happen and allocate resources more efficiently to ensure the most vulnerable are better protected.[xvii] By utilizing innovative forms of AI in the police force, both GeoDASH in Vancouver and the Lab in Saskatoon are focusing on a proactive, predictive and deterrence-based approach to public safety, rather than the traditional reactive framework.

Predictive Policing: Ethical and Legal Concerns

The ethics behind AI-driven predictive analytics is controversial. Stephen Wormith, director of the Centre for Forensic Behavioral Science and Justice Studies at the University of Saskatchewan, and a manager at the Lab, has said that “at the end of the day we’re talking about prediction, per se about criminal behavior, the benefits far outweigh the costs and risks in our view.”[xviii] However, several concerns remain regarding how predictive systems operate in practice. First, predictive systems rely heavily on historical policing data. If that data reflects past enforcement patterns or systemic bias, those patterns may be reproduced by the algorithm itself. Research has shown that AI-driven predictive analytics may be more harmful to society than originally thought. Eurac Research, a non-profit private research centre based in Europe, recently concluded that predictive policing tools “may disproportionately target ethnic minorities based on past crime data, which itself may be the product of systemic discrimination in law enforcement.”[xix] Indeed, if the data which is inserted into the AI algorithms is flawed due to corrupt and discriminatory policing practices, then this may amplify the systemic issues that are already present in society. In Canada, where the Federal Government has acknowledged that “racism in policing is real,”[xx] politicians (who introduce laws governing AI) and law enforcement (who enforce them) must ensure that the use of AI in policing is fair, transparent and nonbiased. Moreover, as technology such as GeoDASH and the algorithmic AI from the Lab become more prominent across Canada, and as additional technologies are created and introduced, society as a whole must be cognizant of the need for ethically sound and principled oversight. In particular, there should be a mindfulness of potential Charter[xxi] infringements and implications, such as the right to life, liberty and the security of the person under s. 7,[xxii] and the right to equality under s. 15.[xxiii]

Facial Recognition and Identity Technologies

One of the most profound ways in which AI has been used by law enforcement agencies across Canada is via facial recognition technology. Two different models of facial recognition have emerged in policing: systems that rely on large internet-scraped image databases, and systems that rely on closed law-enforcement databases such as mugshot collections. In October of 2019, the Royal Canadian Mounted Police (RCMP), in its commitment to utilizing AI to combat crime and enhance criminal investigations, collaborated with a tech-start-up company known as Clearview AI.[xxiv] Utilizing Clearview AI’s Facial Recognition Software (FRS), the RCMP relied on Clearview AI to consolidate a database of more than three billion images of the faces of people across the globe, including Canadians.[xxv] These images, which were selected from the internet without the users’ consent, allowed the RCMP to take photographs of people from criminal investigations and match them against the photographs in the AI databank.[xxvi] Specifically, the technology utilized the database to select online photos of faces and–through analyzing patterns, shapes, and proportions of facial features and contours–compare them to photos and videos captured by the RCMP, with the intention of identifying suspects.[xxvii] Such an example of its use would be the RCMP uploading CCTV footage into the FRS and then generating potential identity matches based off the online database.[xxviii] One of the main goals underpinning the RCMP’s utilization of FRS was to decrease the amount of time it takes for investigators to review potential matches for perpetrators of crimes.[xxix]

In May of 2024, in Ontario, the York Regional Police and Peel Regional Police similarly implemented the use of AI facial recognition technology to assist in the investigations of crimes.[xxx] Specifically, the police agencies now use a digital software provided by “industry leader” and global tech company IDEMIA, which allows the agencies to compare images of suspects or persons of interest at crime scenes to criminal booking images (also known as mugshots).[xxxi] It is stated by the two police agencies that these mugshots photos, which are collected in the database and thus fed into the AI software, are collected lawfully under the Identification of Criminals Act.[xxxii] This technology assists both of the police agencies in solving crimes relating to homicides, sexual assaults, and robberies.[xxxiii] The two police agencies have stated that such use of AI not only protects the integrity of the investigative process and assists officers in identifying suspects, but also decreases the purchasing, maintenance and operating costs of investigations.[xxxiv] One of the main and most profound differences, however, between the technology used by the RCMP and the technology used by the York and Peel Regional Police Forces, is the fact that the latter does not utilize images from the internet.[xxxv] Rather, the only images involved in the software are the images obtained through lawful mugshots, and lawful crime scene photos.[xxxvi] Another key difference is that the latter does not utilize Clearview AI. Rather, the technology is provided by IDEMIA, whose facial recognition technology “ranked most accurate on the false match rate fairness test.”[xxxvii]

A Slippery Slope: Facial Recognition

It is clear that the innovation of biometric and AI facial recognition technology is promising in its ability to assist law enforcement. Yet, similar to AI-driven predictive analytics in policing, the use of facial recognition technology raises serious ethical concerns. One of the most troubling issues is that facial recognition systems have been shown to perform less accurately when identifying people of color.[xxxviii] Such an example was seen recently in Detroit, where a Black man named Robert Williams was arrested in front of his children and held in detention for a night as a result of being falsely identified as a suspect by an AI facial recognition system.[xxxix] As a result of the prejudice Williams experienced, he sued both the local police and government.[xl] While this case occurred in the United States, it illustrates risks that could arise in Canada absent strong regulatory safeguards. Such risks include facial recognition disproportionately targeting minority populations and communities across the country.

Likewise, facial recognition also raises privacy concerns in the context of policing. Such an example was seen in June of 2021, for instance, when an investigation by the Office of the Privacy Commissioner of Canada (OPC) concluded that the RCMP’s use of Clearview AI’s facial recognition technology violates the Privacy Act.[xli] The OPC concluded that since the images were gathered without the user’s consent, the images were thus taken illegally.[xlii] As soon as the RCMP became aware of the OPC’s investigation, it ceased using the technology.[xliii] Additionally, as a result of OPC’s findings, it was announced that there would be public consultations across the country to “help establish clearer rules and consider whether new laws are desirable.”[xliv] Such recommendations (which the RCMP agreed to implement) includes conducting fulsome assessments of third party data collection practices, and the creation of an oversight function intended to ensure that any new technology which the RCMP implements is used in accordance with Canadian law.[xlv]

Conclusion

Law enforcement agencies across Canada have been steadfast in developing and implementing AI in policing. Two specific uses of AI, predictive analytics and facial recognition, have emerged as prominent. Yet, with the risk of systemically compromised data and racial profiling, there are significant ethical and privacy concerns. If AI-based policing tools are to be used responsibly in Canada, several safeguards should accompany their deployment. These could include public impact assessments before new systems are implemented, independent accuracy and bias audits, and clear rules governing data retention and permissible uses. Whether AI represents a natural step forward in policing ultimately depends on whether governments and police agencies can ensure that these systems operate fairly, transparently, and without discriminatory bias. If Canadian governments, law enforcement agencies and policy makers are unable to guarantee that both predictive analytics and facial recognition software are indiscriminatory and unbiased, then these issues will likely not go away. Rather, with the ever-evolving digitization of society, these issues will only grow more prevalent. In the Canadian context, Charter rights must be rigorously protected to deter events such as wrongful detention and discriminatory misidentification.

Endnotes

[i] Vancouver Police Department, “Vancouver Police Adopt New Technology to Predict Property Crime” (21 July 2017) online: < https://vpd.ca/news/2017/07/21/vancouver-police-adopt-new-technology-to-predict-property-crime/> [Vancouver].

[ii] Ibid.

[iii] Ibid.

[iv] Ibid.

[v] Ibid.

[vi] Ibid.

[vii] Ibid.

[viii] Matt Meuse, “Vancouver Police Now Using Machine Learning to Prevent Property Crime” (22 July 2017) online: https://www.cbc.ca/news/canada/british-columbia/vancouver-predictive-policing-1.4217111 [CBC].

[ix] https://www.cbc.ca/news/canada/british-columbia/vancouver-predictive-policing-1.4217111

[x] Ibid.

[xi] Ibid.

[xii] Ibid.

[xiii]Meaghan Craig, “Saskatoon Police Lead the Country with Predictive Analytics Lab” (15 January 2016) online: < https://globalnews.ca/news/2455063/saskatoon-police-lead-the-country-with-predictive-analytics-lab/> [Global].

[xiv] Ibid.

[xv] Ibid.

[xvi] Ibid.

[xvii] Ibid.

[xviii] Ibid.

[xix] Roberta Medda-Windischer and Katharina Crepaz, “Reframing Minority Rights Amid Global Challenges: The Role of AI and Algorithmic Fairness in Promoting Diversity and Inclusion” (12 May 2025) online: https://www.eurac.edu/en/blogs/midas/reframing-minority-rights-amid-global-challenges-the-role-of-ai-and-algorithmic-f.

[xx] Government of Canada, “Policing” (27 June 2024) online: < https://www.justice.gc.ca/eng/cj-jp/cbjs-scjn/transformative-transformateur/p9.html>.

[xxi] Canadian Charter of Rights and Freedoms, Part 1 of the Constitution Act, 1982, being Schedule B to the Canada Act 1982 (UK), 1982, c 11.

[xxii] Ibid at s 7.

[xxiii] Ibid at s 15.

[xxiv] Office of the Privacy Commissioner of Canada, “RCMP’s Use of Clearview AI’s Facial Recognition Technology Violated Privacy Act, investigation concludes” (10 June 2021) online: < https://www.priv.gc.ca/en/opc-news/news-and-announcements/2021/nr-c_210610/> [Privacy].

[xxv] Ibid.

[xxvi] Ibid.

[xxvii] Public Safety Canada, “Facial Recognition” (12 June 2020) online: <https://www.passengerprotect-protectiondespassagers.gc.ca/cnt/trnsprnc/brfng-mtrls/prlmntry-bndrs/20200930/027/index-en.aspx?wbdisable=true>.

[xxviii] Ibid.

[xxix] Ibid.

[xxx] York Regional Police, “Facial Recognition Technology” online: <https://www.yrp.ca/en/crime-prevention/facial-recognition-technology.asp > [York].

[xxxi] Ibid.

[xxxii] Ibid.

[xxxiii] Ibid.

[xxxiv] Ibid.

[xxxv] Ibid.

[xxxvi] Ibid.