As machine learning systems increasingly shape decisions in hiring, lending, and criminal justice, researchers have proposed a range of ethical guardrails—from formal mathematical definitions of fairness to procedural checklists. A new paper argues that these solutions often miss the messy, power-laden realities of how algorithms actually operate in the world.

The authors turn to urban planning theory, which has long wrestled with similar tensions between technical expertise and community values. They propose a "critical pragmatism" approach: one that treats ethical questions as open-ended, context-dependent problems rather than puzzles with a single correct answer. This means engaging stakeholders, acknowledging uncertainty, and revisiting decisions as circumstances change.

The paper does not offer a specific technical tool or metric. Instead, it suggests a mindset—borrowed from planners—that treats AI ethics as an ongoing, deliberative practice. For data scientists facing real-world trade-offs, the lesson is that ethical ML design may require less formula and more dialogue.