Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Contestable AI needs Computational Argumentation

Francesco Leofante, Hamed Ayoobi, Adam Dejl, Gabriel Freedman, Deniz Gorur, Junqi Jiang, Guilherme Paulino-Passos, Antonio Rago, Anna Rapberger, Fabrizio Russo, Xiang Yin, Dekai Zhang, Francesca Toni

TL;DR

The paper tackles the lack of contestability in current AI systems and argues that computational argumentation (CA) is well suited to support contestation. It introduces an abstract framework wherein a contested ADS is described by a model $M:I \rightarrow O$, with explanation $E$, redress $R$, and a contester equipped with ground generator $G$, interacting via $I$. The authors outline how contestation can occur across outputs, reasoning, and the full model (settings A–C) and how explanations, robustness, and faithfulness affect redress. They discuss CA-based explanations, redress, and interaction as the basis for end-to-end contestable AI and point to future work on multi-party scenarios, engineering challenges, and integration with formal verification and logic-based explainability.

Abstract

AI has become pervasive in recent years, but state-of-the-art approaches predominantly neglect the need for AI systems to be contestable. Instead, contestability is advocated by AI guidelines (e.g. by the OECD) and regulation of automated decision-making (e.g. GDPR). In this position paper we explore how contestability can be achieved computationally in and for AI. We argue that contestable AI requires dynamic (human-machine and/or machine-machine) explainability and decision-making processes, whereby machines can (i) interact with humans and/or other machines to progressively explain their outputs and/or their reasoning as well as assess grounds for contestation provided by these humans and/or other machines, and (ii) revise their decision-making processes to redress any issues successfully raised during contestation. Given that much of the current AI landscape is tailored to static AIs, the need to accommodate contestability will require a radical rethinking, that, we argue, computational argumentation is ideally suited to support.

Contestable AI needs Computational Argumentation

TL;DR

The paper tackles the lack of contestability in current AI systems and argues that computational argumentation (CA) is well suited to support contestation. It introduces an abstract framework wherein a contested ADS is described by a model , with explanation , redress , and a contester equipped with ground generator , interacting via . The authors outline how contestation can occur across outputs, reasoning, and the full model (settings A–C) and how explanations, robustness, and faithfulness affect redress. They discuss CA-based explanations, redress, and interaction as the basis for end-to-end contestable AI and point to future work on multi-party scenarios, engineering challenges, and integration with formal verification and logic-based explainability.

Abstract

AI has become pervasive in recent years, but state-of-the-art approaches predominantly neglect the need for AI systems to be contestable. Instead, contestability is advocated by AI guidelines (e.g. by the OECD) and regulation of automated decision-making (e.g. GDPR). In this position paper we explore how contestability can be achieved computationally in and for AI. We argue that contestable AI requires dynamic (human-machine and/or machine-machine) explainability and decision-making processes, whereby machines can (i) interact with humans and/or other machines to progressively explain their outputs and/or their reasoning as well as assess grounds for contestation provided by these humans and/or other machines, and (ii) revise their decision-making processes to redress any issues successfully raised during contestation. Given that much of the current AI landscape is tailored to static AIs, the need to accommodate contestability will require a radical rethinking, that, we argue, computational argumentation is ideally suited to support.
Paper Structure (5 sections, 1 equation, 1 figure)

This paper contains 5 sections, 1 equation, 1 figure.

Table of Contents

  1. Introduction
  2. AI Contestability in the literature
  3. An abstract view of AI contestability
  4. The Role of Computational Argumentation
  5. Discussion and Future Work

Figures (1)

  • Figure 1: An abstract view of AI contestability: the contested ADS (left) is equipped with a model ($M$), an explanation method ($E$), and a redress method ($R$); the contester (right) is a human or an ADS equipped with a ground generator for contestations ($G$); both contested ADS and contester are able to interact ($I$) .

Theorems & Definitions (1)

  • Example 1