AI and policing: the governance and use of artificial intelligence by police in Canada
About the project
Police services across Canada are using, or exploring, a broad range of artificial intelligence (AI) technologies, including facial recognition, location-based predictive policing, automated licence plate readers, AI-assisted police reports, social media monitoring, gunshot detection, probabilistic genotyping, data mining and video analysis tools used to identify child sexual abuse and exploitation materials.
While promising productivity and public safety benefits, these tools also raise various concerns and risks. Canada currently lacks AI-specific legislation and regulations governing the use of these technologies, or even a statement of principles on the governance and use of AI by police.
This project examines the growing use of AI by police, the benefits and risks these technologies present, and opportunities for stronger governance across the following areas:
- Police uses of AI: What types of AI technologies are currently in use, or have been used, by Canadian police forces, and for what specific purposes? What evidence exists of their benefits and risks?
- Police self-regulation: How effectively are police agencies in Canada governing their own use of AI technologies?
- Governance approaches: How can international and Canadian AI governance initiatives help guide responsible and trustworthy AI adoption in Canadian policing?
Key findings
- Limited independent peer-reviewed empirical validation of police AI tools: There is strikingly little independent scientific evidence demonstrating that many police AI technologies are accurate, reliable, effective or necessary. Despite widespread claims of efficiency and objectivity, documented error rates, bias amplification and false positives remain significant. However, there is evidence that some AI tools have led to public safety and efficiency gains, and have demonstrated improvements under controlled conditions as technology improves. As a result, a high degree of skepticism and caution is needed in approaching any AI policing technology.
- Widely documented risks with police AI tools: The most prevalent and persistent risks posed by many operational applications are bias amplification and discriminatory outcomes, privacy infringements, mass surveillance, lack of transparency and accountability, accuracy and reliability concerns and infringements of constitutional rights.
- Police self-governance of AI use varies widely: A wide range of approaches has been adopted by Canadian police services with respect to their use of AI technologies, ranging from formal policies with requirements varying based on risk level (Toronto Police Service) to public transparency and internal oversight mechanisms (RCMP) to directives on specific AI technologies (Peel Regional Police) to commitments to adopt policies before the deployment of new AI technologies (Vancouver Police Department). However, concerns have arisen about reactive policymaking, disputed risk classification, a lack of independent third-party oversight, the extent of public consultation and transparency gaps.
- Important—but limited—role of impact assessment tools and existing oversight bodies: Privacy and human rights commissions have stepped in to address some of the challenges posed by police uses of AI. This is especially important given the absence of specific legislative and regulatory measures and the few judicial decisions to date in this area. Privacy Impact Assessments, Algorithmic Impact Assessments and Human Rights AI Impact Assessments are key operational instruments for risk management of police uses of AI. However, these tools are general in nature and not designed specifically for the coercive, surveillance and evidentiary dimensions of policing. They also do not fully address compliance with the Canadian Charter of Rights and Freedoms in all aspects relevant to criminal justice.
- Consensus on core AI governance principles: Across AI governance initiatives relevant to policing, consistent themes emerge, including necessity and proportionality, lawful authority, human rights and privacy protection, accuracy standards, bias mitigation, explainability, human oversight, transparency and public notice, public engagement, accountability, and access to recourse.
Policy implications
- Existing laws of general application should be rigorously applied to police AI tools: The Charter, criminal procedure, privacy law, evidence law and human rights law already apply to any use of AI by police in Canada. While these frameworks are insufficient on their own, they provide a crucial set of standards for immediate application. Criminal justice practitioners, judges and legal scholars play an essential role in safeguarding the integrity of the criminal justice system, making AI literacy and legal competency in this area vital.
- The need for principles on the governance and use of AI by police in Canada: Consistent, national standards could be informed by a wealth of Canadian and international AI governance initiatives, including those specific to policing. Notably, a significant blind spot in these otherwise insightful instruments concerns the impacts of police use of AI on Indigenous Peoples.
- Public engagement and multi-stakeholder discussions are needed regarding the development, procurement and deployment of AI by Canadian police agencies.
- The importance of AI-specific policing legislation and regulations: There is a lack of federal or provincial legislation and regulations that directly address the use of AI by police. This persists despite repeated calls from law reform commissions, privacy commissioners, human rights bodies and civil society organizations for clear statutory frameworks tailored to address the known risks of the use of AI by police.
Contact the researchers
Benjamin Perrin, Professor, Peter A. Allard School of Law, The University of British Columbia: perrin@law.ubc.ca
Further information
Read full report (Coming soon)