How AI can be used to simplify small claims proceedings

In November 2023 the University’s Faculty of Laws’ Public Law Department and the Faculty of Information & Communication Technology’s Artificial Intelligence Department commenced an 18-month project, funded by Xjenza Malta and supported by the Ministry for Justice and Reform of the Construction Sector, the outcome of which being the creation of a working prototype application called AMPS, intended as a tool to help the adjudicators of the Small Claims Tribunal (SCT) in the process of deciding the cases before them.

The SCT was chosen because in deciding money claims of up to €5,000 it is a microcosm of the other civil courts and the issues the adjudicators face bear much similarity with those faced by magistrates and judges presiding over the inferior and superior civil courts respectively. The rulings of the SCT, like those of the other civil courts, are in Maltese (subject to the odd exception catered for at law).

The Maltese language presents a challenge in itself, namely, how to instigate machine learning in a low resource language, apart from the fact that AMPS must learn not just any Maltese but legal Maltese terms as used in the Maltese courts and tribunals. The technical challenges to be overcome in the process of creating this prototype application are the same as the other courts, or very similar, thus making the SCT a good model for the exercise at hand.

Initially, the people behind this project were inspired by the writings of authors such as Susskind, and envisaged the creation of a system which learned from data fed to it, and based on its acquired knowledge, would predict the outcome of yet undecided SCT cases.

Approach promotes user trust through greater transparency and interpretability

However, meetings with stakeholders namely SCT current and former adjudicators for the large part revealed an aversion for prediction: indeed there was a near unanimous sentiment that they, and they alone, should conclude the outcome of the particular case and that a predictive system, even a system which merely offered the option of prediction, would likely either make the human adjudicator complacent because they would start to use it to save time by simply seeing how AMPS decides a case and then drafting their ruling accordingly, or such prediction would cast doubt in the adjudicator’s head if they decided one way and AMPS came up with a different solution.

On the other hand, these same interviewees welcomed the idea of a user-friendly, reliable tool which could cut down research time and possibly even give them access to information which would otherwise be hard to obtain, or which they might have overlooked completely. Crucially, AMPS has to earn their trust, in the sense that it not only has to prove reliable by providing accurate and correct information when prompted, but it also has to put the users’ minds at rest that it will not in any way breach rules of ethics and confidentiality, for example by requiring the user to divulge confidential information or by sharing information with third parties.

AMPS leverages the bilingual AMPS-JuST (short for Justice) dataset, a meticulously curated corpus consisting of Maltese SCT judgements. This dataset, developed through a hybrid approach combining GPT-4-generated annotations with subsequent verification by expert lawyers, ensures accurate rhetorical role labels, concise summaries, and bilingual translations.

Moreover, AMPS integrates advanced eXplainable AI (XAI) techniques to generate transparent and context-aware explanations, clarifying why specific results or recommendations are provided. By combining the high-quality AMPS-JuST dataset with explainability methods, AMPS enhances both the efficiency and reliability of legal research. This approach promotes user trust through greater transparency and interpretability, enabling adjudicators to quickly identify pertinent information, save valuable time, and maintain informed control over their decision-making processes.

It is extremely important that adjudicators always feel in control and never get the impression that AMPS is in any way vying to substitute them, to be crowned a ‘Computer Judge’. The creators of AMPS have no such intention, as emerges from this first working prototype concerning the Maltese Courts and machine learning in the Maltese language including Maltese legal terminology.

Ivan Mifsud is the dean of the Faculty of Laws. Charlie Abela is a senior lecturer at the Department of AI, Faculty of ICT. Joel Azzopardi is an associate professor at the Department of AI, Faculty of ICT.