Personalized Causal Recourse A Human-In-The-Loop Approach

Jul 16, 2026

Upload by Eliana

Algorithmic recourse provides tailored recommendations for users affected by unfavorable machine learning decisions in high-stakes settings, where people must understand and contest outcomes. Conventional causal recourse often depends on closest counterfactuals or assumes the user’s causal structure is known, leading to interventions that ignore individual context and feature interactions. This work presents a human-in-the-loop Bayesian framework that iteratively queries users to estimate a surrogate structural causal model, enabling personalized recourse that is plausible, cost-effective, and causally aligned. Simulations on linear and non-linear models show promising gains, while highlighting the need for accurate, robust approximations and noise modeling.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages15

Reading Time38 min

Introduction

Problem setting and motivation

Proposed approach: Personalized Causal Recourse

Contributions

arXiv :2607 .03425v 1 [ cs .AI] 3 Jul 2026 Personalized Causal Recourse: A Human-In-The-Loop Approach Denise Tampieri 1⋆[0009−0002−2520−8716], Giovanni De Toni2[0000−0002−8387−9983], and Paolo Giudici3[0000−0000−0000−0000] 1 The University of Edinburgh, Edinburgh, UK [D.Tampieri@sms.ed.ac.uk](D.Tampieri@sms.ed.ac.uk) 2 Fondazione Bruno Kessler, Trento, Italy 3 University of Pavia, Pavia, Italy Abstract. Algorithmic recourse addresses the challenge of providing tailored recommendations to users affected by unfavorable machine learning decisions, in potentially high-stakes scenarios. Traditional approaches torecourse often rely on the closest counterfactual explanations or assume a priori knowledge of a user’s causal structure, resulting in interventions that overlook individual contexts and specific feature interactions. To overcome these limitations, we study a human-in-the-loop framework that iteratively approximates the user’s structural causal model through interactive queries via Bayesian inference before producing recourse recommendations. This framework exploits humans’ feedback to improve the identification of causal effects, allowing personalized recourse that is plausible, cost-effective, and aligned with the actual causal dependencies of each user. As a proof of concept, we evaluate this framework through simulated human responses. Our simulations across linear and non-linear causal models show promising results, though challenges remain in capturing complex, non-linear structures, emphasizing the importance of accurate approximations and robust noise distribution modeling. Keywords: Algorithmic Recourse · Explainable AI · Human-in-the-Loop · Causality · Counterfactual Explanations 1 Introduction Machine learning models are increasingly used in high-stakes decision-making tasks such as healthcare, finance, criminal justice, defense, and autonomous systems [7, 12 ,22] . In these settings, trust in automated systems requires more than predictive accuracy: affected individuals must be able to understand and contest decisions that significantly impact their lives. Counterfactual explanations are a prominent form of local explanation that describe how an individual’s input features would need to be changed to reverse an unfavorable decision. Based on this idea, algorithmic recourse aims to provide actionable recommendations that allow people to improve their results [21] . ⋆ Work conducted during the author’s MSc at the University of Trento, Trento, Italy. 2 Tampieri et al. Fig. 1: Personalized Causal Recourse. Overview of the role of the Structural Causal Model (SCM) estimation in the algorithmic recourse. Gray annotations illustrate an example in a loan application setting. Recent work frames recourse in causal terms, modeling recommendations as interventions on an individual’s features [15,6] rather than as independent feature changes [21] . Causal recourse provides a principled way to reason about how actions propagate through interdependent features and holds the promise to yield more realistic and lower-effort interventions. However, existing causal recourse methods are based on a strong assumption: that the true causal model governing an individual’s features, including both the causal graph and structural equations, is known or can be reliably approximated from observational data [16 ,6] . In practice, this assumption rarely holds [11] . Causal mechanisms may differ between individuals, observational data may be unavailable, and causal knowledge maybe incomplete or subjective [16] . As a result, recourse recommendations derived from an incorrect causal model may be costly or ineffective when implemented in the real world. This limitation applies broadly to current approaches that propose real-world interventions in decision-making systems. In this work, we address this challenge by proposing an alternative human-in-theloop (HITL) framework to infer user-specific causal models and ge

FAQs

What problem does personalized causal recourse address in high-stakes ML decisions?

It tackles the need for actionable, personalized recommendations that can reverse unfavorable outcomes while users must be able to understand and contest decisions that affect their lives.

Why do traditional causal recourse methods often fail in practice?

They typically assume the true user causal model is known or can be reliably approximated from observational data, which is often unrealistic because causal mechanisms differ across individuals and causal knowledge may be incomplete.

How does the proposed human-in-the-loop framework estimate a user’s causal model?

It queries the user via interactive intervention–response pairs and uses Bayesian inference with a surrogate structural causal model estimated through MCMC before generating recourse grounded in the user’s own causal structure.

Jul 16, 2026

Upload by Eliana