A Reasonable Apprehension of AI Bias: Lessons from R v. RDS

Abstract

This paper explores discriminatory bias in automated decision-making. It does so by using a series of court decisions in R. v. RDS (culminating in a 1997 decision of the Supreme Court of Canada) to illustrate some of the potential frailties in approaches to this issue. This case involved claims that certain comments made by Canada’s first black woman judge in favouring the testimony of a Black youth over a White police officer raised a reasonable apprehension of bias. The decisions across three levels of court in this case help reveal the challenges of risk-mitigation approaches to AI bias in which we identify risks, develop strategies to mitigate them, and monitor outcomes. Risk-based approaches tend to assume that there is a social consensus about what bias is and how it is manifested. They also tend to lead us towards technological solutions. R v. RDS teaches us that understanding, identifying, and addressing bias may be much messier.

This paper begins with a brief overview of discriminatory bias in AI systems. Part 2 provides a summary of the dispute at the heart of R. v. RDS. Part 3 teases out four themes emanating from R v. RDS that are relevant to the AI context. The first is the tension between facts and opinion. The second is a common theme in the AI context: transparency and explainability. The third is the issue of biased input and biased output. Fourth is the role of the human-in-the-loop. The paper concludes by considering that a statistical and technological approach to identifying and mitigating bias in automated decision-making may unduly narrow the focus and argues for a more robust approach to addressing bias in AI.

Keywords: artificial intelligence, bias, AI bias, AI discrimination, AI, discriminatory bias, automated decision-making, automated decision making