Looking for the Right Paths to Use XAI in the Judiciary - Which Branches of Law Need Inherently Interpretable Machine Learning Models and Why

Aug 5, 2026

In judicial decision support, tracing the reasons behind outcomes is central, making explainability a key requirement for XAI systems. Yet clear standards for explainability—understandability and transparency—remain insufficiently defined, especially across different legal fields. This paper examines which branches of law may demand algorithmically transparent methods, contrasting white-box, black-box with post-hoc explainers, and full black boxes against legal values to identify development paths tailored to each domain.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Introduction

Machine learning in critical contexts

XAI needs in judicial decision support

Explainability standards across legal branches

Levels of explainability: white boxes, post-hoc explainers, full black boxes

Legal values and tailoring model transparency

Standards for ML systems in applying the law

Protecting rights and judicial guarantees

Looking for the Right Paths to Use XAI in the Judiciary Which Branches of Law Need Inherently Interpretable Machine Learning Models and Why? Andrzej Porębski Jagiellonian University, Gołębia 24, Kraków 31-007, Poland Abstract In a legal context, it is often particularly important to be able to trace the reasons why a decision was made; therefore, there may be an intuition that explainability is extremely important in judicial support systems. However, the standard of explainability (understandability, transparency) that machine learning technologies used to assist judges should meet has not yet been described. Defining this standard is even more complicated because, when considering it, it is necessary to take into account not only the specifics of the legal context in general but also of individual branches of law (for example, criminal or civil law) . In this paper, I consider which branches of law, due to their specificity, seem to require the use of the most algorithmically transparent – and thus inherently interpretable – methods. Juxtaposing three general levels of explainability (white boxes, black boxes with post hoc explainers, and full black boxes) with legal values, I consider the paths that the development of machine learning models supporting judicial reasoning should follow in order to be tailored to the specifics of each legal field. Keywords AI & Law, right to a fair trail, XAI in the judiciary, inherent interpretability, decision support systems 1. Introduction Machine learning came into use years ago in areas of life that can be best described as “critical contexts”, that is, contexts in which a special role is attached to maintaining sufficiently high standards because of the need to ensure, among other things, the security or protection of individual rights. Machine learning, therefore, drives things like autonomous cars or various medical technologies. In these areas, the need to take into account the values of eXplainable Artificial Intelligence (XAI) [1] is rightly recognised [2, 3] . Another ”critical context” in which machine learning methods can find application, and which is central to the presented paper, is in assisting the judiciary. Research on this topic clearly signals the possibility of using machine learning in supporting the operation of lawyersand the application of the law, for example, through risk prediction, prediction of likely court rulings, or warning of potential bias in judgements [4, 5] . Also, LLM models — including those designed for legal applications such as SaulLM-7B [6] — can hypothetically support lawyers’work on documents or related to jurisprudence. However, the application of the law by state Late-breaking work, Demos and Doctoral Consortium, colocated with The 2nd World Conference on eXplainable Artificial Intelligence: July 17–19, 2024, Valletta, Malta [Envelope-Open](Envelope-Open and.porebski@uj.edu.pl)[ and.porebski@uj.edu.pl](Envelope-Open and.porebski@uj.edu.pl) (A. Porębski) GLOBE https://www.researchgate.net/profile/Andrzej-Porebski-2 (A. Porębski) Orcid 0000-0003-0856-5500 (A. Porębski) © 2024 Copyright for this paper by its author. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0) . CEUR ~~ ~~[Workshop](Workshop ceur-ws.org)[ ceur-ws.org](Workshop ceur-ws.org)[ ](Workshop ceur-ws.org)[Proceedings](Proceedings ISSN 1613-0073)[ ISSN 1613-0073](Proceedings ISSN 1613-0073) bodies, and in particular by the judiciary, is one of the particularly critical contexts of social life, especially as it often represents the last opportunity for an individual to obtain protection of his or her rights. This is why, for example, the COMPAS tool used in the United States to estimate the risk of recidivism —which operates in the absence of more precise regulation, although it may have had a real impact on the decisions and sentences handed down — has received far-reaching criticism in the prevailing literat

FAQs

Why is explainability particularly important in judiciary decision support systems?

Judicial contexts require traceable reasons for decisions because courts are often the last protection of individual rights. Explainability helps ensure those reasons are understandable and transparent.

What challenge does the paper highlight about current explainability requirements?

The paper notes that a specific standard for explainability has not yet been adequately described for machine learning technologies used by or to assist judges, and the issue becomes more complex when considering different branches of law.

How does the paper relate transparency of machine learning models to legal values?

It juxtaposes white boxes, black-box models with post-hoc explainers, and full black boxes with legal values, then outlines paths for developing judicial reasoning support models that match the specifics of each legal field.

Aug 5, 2026

Upload by River Wang