LLMs in Parliaments: Fostering Technological Innovation, Cooperation and Democratic Principles

Large Language Models (LLMs) are gaining attention worldwide, including in parliaments. Their ability to draft legislation, manage amendments, transcribe, translate and summarise texts has been well demonstrated. However, the use of LLMs is not neutral and these tasks greatly impact democracy. This paper explores, from both the legal and technological perspectives, a crucial problem when using LLMs to streamline parliamentary activities, namely assuring explainability and reliability. Focusing on the Retrieval-Augmented Generation and Chain-of-Thought solutions, the paper analyses how these technical mechanisms facilitate the compliance with the legislation in force at the European level. In doing so, the paper envisages a feasible use case, which is an LLM for assessing whether a bill complies with the distribution of legislative powers between the State and Regions, focusing on the Italian constitutional system. Accordingly, the paper outlines the systematic implications for Parliament of using such an LLM, highlighting the related challenges and benefits.