Legality of Technical Explainable AI Methods

mapping XAI methods and legal justification

Research Question: Can XAI methods satisfy legal obligations of transparency, reason-giving and legal justification?

Originally part of a European Lighthouse on Secure and Safe AI, this project aims to answer whether technical eXplainable AI (XAI) methods can meet legal obligations for transparency, reason-giving and justification when algorithmic decision-making systems are deployed within organisations.

Collaborators:

  • Paul Waller— University of Bradford and Thorney Isle Research.
  • Prof. Karen Yeung — Interdisciplinary Professorial Fellow in Law, Ethics and Informatics, Birmingham Law School and School of Computer Science.

You can find the original short report here (ELSA Deliverable 3.3, 28th August 2024):

This research is still ongoing and any input or discussions are very welcome.

Research from this project will be presented at the Law, AI and Regulation (LAIR) conference in June 2026 and the The 4th World Conference on Explainable Artificial Intelligence (XAI-2026) in July 2026. More details of presentations/publications to come.

Contributions

Organisational decision-making architecture

A key contribution of the project is the proposal of an organisational decision-making architecture to be able to disentangle what needs explaining in algorithmic decision-making systems. This was proposed by Paul Waller, see more details here.

Organisational decision-making architecture

Figure: Organisational decision-making architecture (ODMA). This diagram shows how an algorithmic decision-making (ADM) system is embedded within organisational roles, formal chains of authority, responsibilities for carrying out designated tasks and duties, and points for human oversight and intervention.

Other key contributions of the project:

  • A mapping of common XAI methods to legal concepts of explanation and reason-giving.
  • Discussion of limits of current technical approaches and recommendations for organisational and regulatory practices.
  • Case analysis of algorithmic decision-making systems embedded in organisational workflows and the points at which human intervention and oversight are required.