The Responsible Organisation

The University of Bologna (Alma Mater Studiorum Università di Bologna) is a public research university in Bologna, Italy. Prof. Monica Palmirani is a professor of Legal Informatics at the University of Bologna and leads the "AI for Law and Governance" unit the the Research Institute for Human-Centered Artificial Intelligence (Alma AI) of the university, focused on the intersection of law and technology, with a particular emphasis on the use of AI in the legal domain. The research team has worked on the GENAI4LEX-B project, which aims to develop AI tools to support the legislative process in the Italian Chamber of Deputies.

The Italian Chamber of Deputies (Camera dei Deputati) is one of the two houses of the Italian Parliament. It is composed of 400 members elected every five years.  The Chamber holds the power to initiate and enact laws, approve, or reject legislation, and exercise oversight over the government. The GENAI4LEX-B tool was nominated by the Chamber as one of the three projects selected through a call for proposals to enhance legislative efficiency.

For the preparation of this article, insights were drawn from an interview conducted with Prof. Monica Palmirani, and Assistant Professors Salvatore Sapienza and Michele Corazza, University of Bologna.

The problem

The GENAI4LEX-B project was initiated in 2024 in response to several critical challenges observed within the legislative process, particularly in the Italian Chamber of Deputies. These challenges include the extensive time required for research, the complexities of legal language, the dynamic nature of laws and regulations, the effort involved in summarising committee amendments, the intricacies of compliance verification, and the difficulties in managing and accessing diverse legal sources.

A primary challenge is the extensive time dedicated to preparatory research by parliamentarians and legislative staff, who must investigate existing laws, regulations, and Constitutional Court decisions to ensure the consistency of new, bills and avoid conflicts with established legislation. This research process is complicated due to the technical nature of legal language, whose precise interpretation is highly time consuming. 

Additionally, laws and regulations are subject to constant legislative revisions, amendments and updates. Staying beside the latest legislation poses a considerable challenge in ensuring that drafted bills remain compliant with current legal frameworks. Together with this, committee amendments are often numerous and complex and require substantial effort to understand their implications fully and summarise them.

The verification of compliance, which involves ensuring that proposed bills adhere to all relevant policies, strategies, and constitutional principles, is yet another time-consuming aspect of the legislative process. 

Finally, effectively managing and accessing diverse legal sources, encompassing national and European legislation, Constitutional Court decisions, and parliamentary documents, presents another difficult task.

The solution and its implementation

The GENAI4LEX-B project, led by Professor Monica Palmirani from the University of Bologna, is an initiative focused on supporting the legislative process within the Italian Parliament, and tackling the challenges inherent to the drafting of a new bill by integrating Generative AI technologies. 

Its objective is to assist parliamentarians in the drafting of bills. For this purpose, the tool automates various tasks, such as the identification of relevant legal documents and the summarisation of committee amendments. Being still under development and testing, it has the potential to impact the legislative process to support a more efficient, transparent and accessible lawmaking processes. 

The GENAI4LEX-B project was one of three winning projects selected from 28 proposals submitted to a call for proposals by the Italian Chamber of Deputies seeking innovative projects that leverage generative AI to support parliamentary processes.

Partnerships and Collaborations

The GENAI4LEX-B project was developed through a rich network of partnerships and collaborations, drawing expertise from various sectors based on 25 year-research efforts. Coordinated by the University of Bologna, the consortium comprises also other academic institutions: LUISS CESP - Center for Parliamentary Studies, University of Verona, and the University of Turin, in collaboration with the National Research Council (CNR)'s Institute of Legal Informatics and Judicial Systems, and three spin-off companies, BitNomos, Aptus, and ASIMOV AI. Crucially, the project maintains close ties with governmental institutions, primarily the Italian Chamber of Deputies, the main intended user of the developed tools. This collaboration ensures the practical relevance and applicability of the solutions within the parliamentary legislative process. Additionally, potential collaboration with the European Commission's DG Informatics group points towards the project's ambition for broader European impact.

Technical features and functionalities

GENAI4LEX-B employs cross-technology applications and combines legal standards with legal ontologies and advanced AI techniques. It uses a hybrid approach that combines the strengths of symbolic AI, which operates by encoding legal knowledge like laws and regulations into a structured, machine-readable format (Law as Code), with the capabilities of generative AI, which can produce human-quality text for drafting and summarising legal documents and committee amendments. This allows the system to generate legal texts that are both accurate and contextually relevant. 

Additionally, GENAI4LEX-B utilises legal ontologies, specifically EuroVoc, to classify and thematically organise legal documents. This permits the model to understand the meaning of legal terms and concepts within their specific context. AI-powered analysis techniques, including clustering and similarity/relevance calculations, are also employed to identify relevant related regulations, laws, and Constitutional Court decisions.

To ensure integration and exchange of information, the tool is compliant with the Akoma Ntoso – LegalDocML OASIS standard, which is the universal standardized XML format for representing legal documents digitally. This standard ensures that legal documents created within the system can be easily shared and used across different platforms and software, regardless of how they were created. Furthermore, it allows legal information to be presented in a standardised way, making it machine-readable and easier for practitioners to navigate and understand.

To verify compliance with existing policies and strategies, GENAI4LEX-B utilises the Houdini deductive-logical engine, which employs the LegalRuleML OASIS standard to model legal norms and automates the compliance verification process. Specifically, Houdini verifies whether a proposed law complies with existing legislation by analysing the text of the proposed law, identifying the relevant legal norms, and checking for any conflicts or inconsistencies. Additionally, Houdini uses logical reasoning to deduce the implications of the proposed law and identify any potential unintended consequences. This helps decision-makers to understand the full impact of the proposed legislation before it is enacted.

Intended users and adoption

While the GENAI4LEX-B tool is primarily intended for use by legislative offices within the Italian Chamber of Deputies, its potential user base extends beyond parliamentarians. The developers envision the tool being particularly useful for legislative staff who are responsible for researching, drafting, and analysing legal documents. 

Additionally, the component focused on summarising amendments is designed to assist the secretariats of political parties in their work of compiling and understanding these changes. Other potential use cases involve general law practitioners and policy makers that need to study compliance aspects and similarities among different law pieces.

Furthermore, there are plans to conduct a pilot test with a user group within the European Commission's DG Informatics, suggesting that the tool could eventually be implemented on a broader European scale.

Expected benefits

A series of positive outcomes of this tool have been preliminarily identified. The main benefits include:

• Improved efficiency: By automating time-consuming tasks, such as legal research and amendment summarisation, the tool can significantly accelerate the legislative process, allowing for faster drafting and review of bills. As an example, the automated summary feature helps to prepare for Chamber discussions and committees’ presentations, tasks that are traditionally developed within the party’s secretariat and are very time intensive.
• Enhanced accuracy: Additionally, the introduction of AI-powered tools can help reduce errors and inconsistencies in legal drafting, which contribute to more precise and coherent legislation in the long term. Additionally, as the tool provides real-time updates on legislative changes, it can help ensure that parliamentarians always have access to current information and avoid related human errors.
• Support for democratic debate: Prof. Palmirani sustains that the tool can also support strengthening the democratic debate. As the system facilitates access and understanding to relevant legislative information and analysis, it can support a more informed and robust debate on legislative proposals. 

Main challenges 

As the project team tested and learned along the development and implementation of the tool, several challenges have also been identified:

• Model training and accuracy: Training accurate AI models is very challenging, especially considering that the nuances of legal terminology can have different interpretations and require specific technical training and constant refinement. 
• Cost of resources: Developing and deploying advanced AI solutions, particularly those involving Generative AI, demands significant investment in robust and scalable infrastructure to handle the computationally intensive tasks. As Prof. Palmirani explained, the technological infrastructure and its related cost represented a challenge the team faced during the development of the tool: "Now we have only taken small parts also for reasons of machine costs [and] there have been crashes. Once a week, while we were doing the test, there was always someone of us who was monitoring the performance of the machine because if it dropped in performance, we would pull it up". This poses challenges also to the tool’s sustainability, as maintaining and updating AI models and infrastructure over time requires continuous resources and expertise.
• Explainability and transparency: For the project team, it is fundamental to provide clear explanations for non-technical users of the tool on its functioning. Prof. Palmirani also highlighted that it was particularly relevant to explain the quality of data that were used to train the model, stressing that they strongly communicated that the tool used exclusively “data coming from safe sources, parliamentary sources [and respecting] the hierarchy of sources".
• User adoption and training: Achieving successful and trustworthy implementation relies also on user adoption and acceptance. Training and support for parliamentarians and legislative staff to effectively utilise the tools is considered necessary. The GENAI4LEX-B team prepared a small chapter or manual covering the skills needed for parliamentarians and the offices’ staff to interpret the outcome of the tool.
• Ethical considerations: The use of AI in lawmaking raises ethical concerns, such as the potential impact on decision-maker autonomy and the need to avoid biases. Prof. Palmirani reflected, as well, potential questions regarding the separation of powers: “if the same tool is used by both the executive offices and the parliamentarians, how do we divide the roles and the tripartition of powers between who proposes, who governs, who legislates and who executes?". To support the ethical and legal assessment of the proposed solutions, especially under the EU AI Act, the institutions could establish task forces aimed at monitoring the implementation of AI in Parliaments.

Link: Generative artificial intelligence in the Parliament: a project coordinated by the University of Bologna Receives an Award — UniboMagazine 

Contact: monica.palmirani@unibo.it