METISiA – a Generative AI Prototype for Epistemic Translation in Legal Disputes

Abstract

Introduction

Artificial Intelligence (AI) in the form of “legal AI” is rapidly transforming legal practice, offering not just enhanced prediction and analysis but also the capacity to redefine legislation, legal reasoning and dispute resolution approaches [1-2]. However, effectively incorporating multistakeholder epistemics and value-based decision hierarchies in dynamic complex legal contexts remains a major challenge for achieving just outcomes. A special area of concern is family law, especially post-separation custody conflicts [3]. Those frequently involve conflicting interpretations of children's needs, driven by differing values, knowledge gaps, and fragmented paradigms among parents and legal professionals. This thesis addresses this challenge by focusing on the design and development of a specialized generative AI assistant, METISiA, to promote more just and child-centered outcomes exemplified in Austrian custody disputes. The core of this research lies in modeling an AI system capable of navigating the complex human dynamics of legal negotiation. This involves two key innovations: first, the construction of a comprehensive epistemic knowledge base that holistically models children’s developmental needs, grounded in interdisciplinary frameworks including 4E cognition. Second, the design of the generative AI prototype itself, which utilizes retrieval-augmented generation (RAG) to ensure its guidance is transparent and grounded in evidence. METISiA is engineered to support value-based argumentation, helping to bridge communication gaps and foster mutual understanding. Its development incorporates critical monitoring to address inherent AI risks, including algorithmic bias. The primary function of the AI is to act as a neutral intermediary that facilitates a more constructive and empathetic negotiation process.

Methodology

A mixed-methods research design will guide the project. An initial phase of systematic literature reviews, semi-structured interviews, and focus groups with parents, legal professionals, and child psychology experts will establish the foundational requirements for the AI model. Subsequently, the METISiA prototype will be built and its effectiveness evaluated through user engagement, assessing its impact on shared understanding, dialogue quality, and overall user acceptance based on its perceived neutrality and accuracy.

Expected Impact

The expected impact of this research is twofold: a functional AI prototype designed for practical use in custody mediation and a robust framework for embedding epistemic justice into future legal AI systems. Ultimately, this project aims to make legal negotiation more evidence-based, understandable, and empathetic, thereby supporting vulnerable stakeholders in achieving more equitable resolutions.

References

[1] E. L. Rissland, K. D. Ashley, and R. P. Loui, “AI and Law: A fruitful synergy,” Artificial Intelligence, vol. 150, no. 1–2, pp. 1–15, 2003.

[2] D. Schwarcz, S. Manning, P. J. Barry, D. R. Cleveland, J. J. Prescott, and B. Rich, “AI-powered lawyering: AI reasoning models, retrieval augmented generation, and the future of legal practice,” Minnesota Legal Studies Research Paper No. 25-16, Mar. 2, 2025. [Online]. Available: https://ssrn.com/abstract=5162111 [Accessed: April 23, 2025]

[3] B. Rešetar and R. E. Emery, “Children's rights in European legal proceedings: Why are family practices so different from legal theories?,” Family Court Review, vol. 46, no. 1, pp. 65–77, 2008.