Epistemic trustworthiness of AI: (the necessity of) explainable AI in legal decision-making

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216224%3A14220%2F25%3A00142925" target="_blank" >RIV/00216224:14220/25:00142925 - isvavai.cz</a>

Result on the web

<a href="https://www.e-elgar.com/shop/gbp/research-handbook-on-epistemologies-of-law-9781035347995.html?srsltid=AfmBOoqAp-QxpWn5Lw8PNSIaGTsN_ZOKgPCv_24rDUPY70_wF2IK8eKA" target="_blank" >https://www.e-elgar.com/shop/gbp/research-handbook-on-epistemologies-of-law-9781035347995.html?srsltid=AfmBOoqAp-QxpWn5Lw8PNSIaGTsN_ZOKgPCv_24rDUPY70_wF2IK8eKA</a>

DOI - Digital Object Identifier

—

Result language

angličtina

Original language name

Epistemic trustworthiness of AI: (the necessity of) explainable AI in legal decision-making

Original language description

The integration of Artificial Intelligence (AI) in legal decision-making poses profound challenges to the epistemic trustworthiness and legitimacy of judicial processes. At the core of these challenges is the necessity for decisions to be transparent, comprehensible, and justified—standards fundamental to the right to a fair trial and the rule of law. This paper explores the tension between the opaque nature of advanced AI models and the established legal requirement for reasoned judgments. By examining case law under the European Convention on Human Rights (ECHR), it identifies the essential elements of judicial reasoning necessary to uphold the legitimacy of automated legal decisions. Explainable AI (xAI) emerges as a potential solution to the black-box problem inherent in many machine learning models. However, current xAI methodologies, such as model-specific or model-agnostic approaches, often fall short of providing explanations that are accessible and epistemically satisfying to diverse stakeholders. The paper highlights the critical need to adapt xAI outputs to meet the cognitive capacities of affected parties while preserving the efficiency and reliability of AI systems. Moreover, the study advocates for a hybrid approach, combining xAI techniques with domain-specific applications of Large Language Models (LLMs). This integration can enhance the interpretability and personalization of AI-driven legal decisions, enabling affected individuals to better understand the rationale behind judicial outcomes. By addressing the technical, epistemic, and normative dimensions of AI in judicial contexts, the chaoter underscores the importance of designing AI tools that do not merely optimize efficiency but also uphold the principles of fair trial central to the rule of law. Achieving this balance is imperative to ensure that AI's role in legal decision-making strengthens, rather than undermines, the legitimacy of the judiciary and the fundamental rights it is tasked to protect.

Czech name

—

Czech description

—

Type

C - Chapter in a specialist book

CEP classification

—

OECD FORD branch

50501 - Law

Project

—

Continuities

I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Publication year

2025

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Book/collection name

Research Handbook on Epistemologies of Law

ISBN

9781035347995

Number of pages of the result

13

Pages from-to

355-367

Number of pages of the book

456

Publisher name

Edward Elgar Publishing

Place of publication

Cheltenham

UT code for WoS chapter

—