Artificial Legal Intelligence

ALIKE and XAIcred.

In increasingly complex legal systems legal information retrieval is an expensive and difficult task. Drawing on open access databases from the European Court of Justice (Curia and EUR-LEX) and various domestic databases on administrative legal decisions, the project leverages Machine Learning, Natural Language Processing, Network Analysis and Graph Neural Networks to build and test various approaches to legal information retrieval, primarily with a focus on precedent finding. This project also seeks to use the same approach to find patterns in the way asylum seekers’ testimony are assessed as either trustworthy or untrustworthy. The aim of this latter part of subproject 1 is to be able to retrieve informal information signals in a set of case law (here, the Danish Refugee Council’s decisions).

ALIKE in collaboration with the IUROPA Project.

Large Language Models (LLMs) have been shown to perform well in generating text that is perceived as meaningful to humans. LLMs are already being used for many different tasks including summarisation, translation, editing, idea generation, etc.  LLM’s are also increasingly used in the legal service sector. Retrieval Augmented Generation (RAG) systems are used to combine LLMs with knowledge bases (consisting of case law and statutory law datasets) to enable legal argument generation. However, most RAG systems are operated by private companies who do not provide much technical information about their products and until today only few or none benchmarks for legal arguments generation exists. This project will build a working state-og-the-art RAG based on an open source LLM and using open access legal information (Curia and EUR-LEX) to test the ability of machines to produce legal text of a quality that is indistinguishable from human generated text. Based on the construction of original new benchmarks and empirical testing, the project will contribute to ways of measuring the quality of machine generated legal text.