Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reassessing Evaluation Functions in Algorithmic Recourse: An Empirical Study from a Human-Centered Perspective

Tomu Tominaga, Naomi Yamashita, Takeshi Kurashima

TL;DR

This study empirically tests the core assumption of algorithmic recourse that minimizing a recourse distance leads to higher acceptance and action, using a car-loan scenario with 362 participants. It formalizes recourse construction around an objective distance d and analyzes two norms, L0 sparsity and L1 proximity, via generalized additive mixed models to capture nonlinear effects and individual differences. The results show that acceptance is not tied to recourse distance, while willingness to act is highest at the smallest distances but plateaus beyond a threshold, revealing a nuanced interaction between distance and human behavior. The findings challenge the standard evaluation functions in recourse research and argue for adaptive, human-centered evaluation functions that account for individual variation to improve the practicality and uptake of recourse plans.

Abstract

In this study, we critically examine the foundational premise of algorithmic recourse - a process of generating counterfactual action plans (i.e., recourses) assisting individuals to reverse adverse decisions made by AI systems. The assumption underlying algorithmic recourse is that individuals accept and act on recourses that minimize the gap between their current and desired states. This assumption, however, remains empirically unverified. To address this issue, we conducted a user study with 362 participants and assessed whether minimizing the distance function, a metric of the gap between the current and desired states, indeed prompts them to accept and act upon suggested recourses. Our findings reveal a nuanced landscape: participants' acceptance of recourses did not correlate with the recourse distance. Moreover, participants' willingness to act upon recourses peaked at the minimal recourse distance but was otherwise constant. These findings cast doubt on the prevailing assumption of algorithmic recourse research and signal the need to rethink the evaluation functions to pave the way for human-centered recourse generation.

Reassessing Evaluation Functions in Algorithmic Recourse: An Empirical Study from a Human-Centered Perspective

TL;DR

This study empirically tests the core assumption of algorithmic recourse that minimizing a recourse distance leads to higher acceptance and action, using a car-loan scenario with 362 participants. It formalizes recourse construction around an objective distance d and analyzes two norms, L0 sparsity and L1 proximity, via generalized additive mixed models to capture nonlinear effects and individual differences. The results show that acceptance is not tied to recourse distance, while willingness to act is highest at the smallest distances but plateaus beyond a threshold, revealing a nuanced interaction between distance and human behavior. The findings challenge the standard evaluation functions in recourse research and argue for adaptive, human-centered evaluation functions that account for individual variation to improve the practicality and uptake of recourse plans.

Abstract

In this study, we critically examine the foundational premise of algorithmic recourse - a process of generating counterfactual action plans (i.e., recourses) assisting individuals to reverse adverse decisions made by AI systems. The assumption underlying algorithmic recourse is that individuals accept and act on recourses that minimize the gap between their current and desired states. This assumption, however, remains empirically unverified. To address this issue, we conducted a user study with 362 participants and assessed whether minimizing the distance function, a metric of the gap between the current and desired states, indeed prompts them to accept and act upon suggested recourses. Our findings reveal a nuanced landscape: participants' acceptance of recourses did not correlate with the recourse distance. Moreover, participants' willingness to act upon recourses peaked at the minimal recourse distance but was otherwise constant. These findings cast doubt on the prevailing assumption of algorithmic recourse research and signal the need to rethink the evaluation functions to pave the way for human-centered recourse generation.
Paper Structure (41 sections, 4 equations, 3 figures, 4 tables)

This paper contains 41 sections, 4 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Contributions.
  3. Related Work
  4. Algorithmic Recourse Generation
  5. Human Evaluation of Algorithmic Recourse
  6. Experiment
  7. Experimental Setup
  8. Hypothetical Scenario for Recourse Provision
  9. Scenario Design
  10. Recourse Construction
  11. Conditions
  12. Distance Metrics
  13. Selection Process
  14. Experimental Procedure
  15. Participant Recruitment and Screening
  16. ...and 26 more sections

Figures (3)

  • Figure 1: Distributions of distance metrics of selected recourses.
  • Figure 2: Overview of the experiment process. Black circles indicate steps taken by participants, white boxes represent interventions from us to participants, and thin dashed arrows depict data flow.
  • Figure 3: GAMM fits of the distance metrics to the propensity for acceptance (a) and action (b). The GAMMs include individual participant-specific effects as the nested random intercept and slopes. The error bars are 95% confidence intervals.