Contestability in Quantitative Argumentation

TL;DR

This paper introduces a novel method using gradient-based explanations to modify argumentation frameworks for better alignment with human preferences, enhancing contestability in AI decisions.

Contribution

It presents the contestability problem for EW-QBAFs and proposes an iterative algorithm with G-RAEs for effective weight adjustments.

Findings

The approach effectively adjusts argument strengths in synthetic frameworks.

G-RAEs provide interpretable guidance for weight modifications.

The method demonstrates success in simulated recommender systems and neural models.

Abstract

Contestable AI requires that AI-driven decisions align with human preferences. While various forms of argumentation have been shown to support contestability, Edge-Weighted Quantitative Bipolar Argumentation Frameworks (EW-QBAFs) have received little attention. In this work, we show how EW-QBAFs can be deployed for this purpose. Specifically, we introduce the contestability problem for EW-QBAFs, which asks how to modify edge weights (e.g., preferences) to achieve a desired strength for a specific argument of interest (i.e., a topic argument). To address this problem, we propose gradient-based relation attribution explanations (G-RAEs), which quantify the sensitivity of the topic argument's strength to changes in individual edge weights, thus providing interpretable guidance for weight adjustments towards contestability. Building on G-RAEs, we develop an iterative algorithm that progressively adjusts the edge weights to attain the desired strength. We evaluate our approach experimentally on synthetic EW-QBAFs that simulate the structural characteristics of personalised recommender systems and multi-layer perceptrons, and demonstrate that it can solve the problem effectively.