To Err is Human, to Audit Divine: A Critical Assessment of Canada’s AI Directive

Abstract

In recent years, an increasing number of global governments have been quietly incorporating automated decision-making systems into their governance processes. The potential efficiency gains from these systems are easy to see. However, their broader impact on core government functions is much less clear, not least because of a broader lack of transparency in how they are being rolled out, and a dearth of public discussion around how they should be used responsibly. In April 2019, Canada's federal government passed the Directive on Automated Decision-Making, with the goal of supporting transparency and public accountability around the adoption of automated decision-making technologies. This paper analyzes the strengths and weaknesses of the Directive on Automated Decision-Making in the context of broader challenges around the use of automated decision-making systems, and provides substantive recommendations for improvement. Ultimately, while the Directive's transparency provisions are a game-changer in supporting public policy debates in this space, there is still room for improvement, particularly in expanding the scope of application and in supporting a specific and meaningful procedure to facilitate complaints or feedback from individuals or communities impacted by automated systems, including a formal process to raise concerns about biased or otherwise problematic results being returned.