Algorithmic recourse faces challenges when many compete for limited resources
Algorithmic Recourse Under Competition
Machine LearningArtificial Intelligence
Summary
Sometimes computers decide who gets something, like a loan, and people want advice on how to improve their chances if denied. Usually, this advice assumes the decision rules don’t change, but when many people try to improve at once, the rules shift, making the advice less useful. The authors studied this problem and created a method that gives advice tailored to both the person and how competition changes the rules. They tested this approach on fake and real data, finding personalized goals can increase success but cost more effort.
What this means in practice
- •For credit risk teams: Improve loan approval advice by accounting for shifting acceptance criteria due to many applicants adjusting behavior simultaneously.
- •For healthcare access managers: Design better patient support recommendations when resource limits cause changing criteria for treatment eligibility under competing demand.
Authors
Shahin Jabbari
Abstract
Algorithmic recourse provides individuals who have received undesirable outcomes from machine learning models with suggestions for minimum-cost improvements to achieve the desired outcome. A central assumption when computing recourse is that the decision rule remains fixed throughout the recourse implementation phase. We challenge this assumption in settings where individuals compete for limited resources. In such settings, widespread recourse implementation can change the acceptance threshold even when the scoring model that is used to evaluate individuals remains the same. This change in acceptance threshold can, in turn, invalidate the original recourse recommendations (i.e., following the recourse may not lead to the desired outcome). To address this problem, we introduce a framework called recourse under competition that jointly optimizes for recommendation recipients and the recommended score target they need to satisfy to balance the recourse cost and post-shift validity among initially rejected individuals. We develop an algorithm based on the Implicit Function Theorem and empirically analyze its performance. Experiments on synthetic and real datasets show that personalized score targets can achieve higher validity, albeit at a higher cost. In contrast, common score targets generally offer favorable cost-validity trade-offs for lower to medium validity values.