From Explanation to Recommendation: Ethical Standards for Algorithmic Recourse

Algorithmic recourse requires ethical frameworks beyond explanation: The paper argues that when people are denied by algorithmic systems (like loan rejections), simply explaining why they were rejected is insufficient. Instead, systems should provide actionable recourse—specific steps people can take to change the outcome—which requires different ethical considerations than traditional explainability.

The capability approach offers a practical ethical standard: Rather than focusing solely on technical metrics, the authors propose using the "capability approach" (a framework from philosophy and economics) to evaluate recourse systems. This means assessing whether algorithmic recourse actually expands people's real-world capabilities and freedoms to improve their circumstances, not just whether recommendations are technically sound.

Diversity in recourse options matters ethically: The paper highlights that algorithmic recourse systems should offer diverse pathways to better outcomes. For example, a rejected loan applicant shouldn't receive only one recommendation (like "increase your income"); systems should present multiple realistic options that respect different people's values, constraints, and life circumstances.

Current recourse systems may overlook fairness concerns: By treating recourse as a recommendation problem rather than purely an explanation problem, the paper suggests existing approaches may miss important ethical issues—such as whether the suggested actions are actually achievable for different groups or whether they perpetuate existing inequalities.

Recourse design has real consequences for human autonomy: The framework emphasizes that how companies design algorithmic recourse directly affects people's ability to contest unfavorable decisions and improve their lives, making ethical standards for recourse design a matter of fundamental fairness and human dignity.