Algorithmic injustice: a relational ethics approach

The Problem with Rationalist AI

Algorithmic systems are increasingly used to automate complex social decisions, such as medical care allocation, policing, and hiring. The author argues that these systems are built on a "rationalist" foundation—a Western philosophical tradition that prioritizes abstract logic, objectivity, and binary categorization. By treating social challenges as purely mathematical problems, data science often strips away the context, history, and human nuance necessary to understand them. This "view from nowhere" assumes that algorithms can be neutral, leading to the dangerous belief that social harms can be fixed simply by "debiasing" datasets or fine-tuning models.

The Relational Alternative

In contrast to the rationalist approach, the author proposes a "relational ethics" framework. Drawing on diverse traditions—including Afro-feminist epistemology, ubuntu, and enactive cognitive science—this perspective posits that human beings and knowledge are fundamentally co-generated through relationships. Instead of viewing individuals as isolated data points, relational ethics emphasizes interdependence and the importance of lived experience. Under this framework, ethics is not a methodology or a checklist, but a continuous, habit-based practice that requires acknowledging power asymmetries and historical injustices.

This paper challenges the data science community to move beyond the "problem/solution" mindset that dominates AI ethics. By highlighting that technical fixes often fail to address the root causes of systemic harm, the author calls for a fundamental shift in how researchers and practitioners approach their work. It argues that true justice in AI requires centering the voices and welfare of the most vulnerable, rather than prioritizing the efficiency and stability of the systems themselves. The paper serves as a call for critical reflection, suggesting that "critical work" should be recognized as an essential component of AI development rather than an outlier.