ICT, Specialised AI Applications and Digital Governance
16:05 - 17:25 | Aula Prima (Level 2)
Chair: Dr Ingrid Galea
Prof. In摹. Saviour Zammit
Department of Communications and Computer Engineering, Faculty of 福利在线免费 and Communication Technology
The ESA-funded SMARTEN6G project and the SINO-Malta–funded TEKAID6G project investigated the impact of AI/ML on 5G, 5G-Advanced, and emerging 6G mobile communication systems, spanning both Non-Terrestrial Networks (NTN) and Terrestrial Networks (TN). This impact was examined from two complementary perspectives: ‘AI/ML for the network’ and ‘the network for AI/ML’.
The former addresses the use of AI/ML techniques to optimise the communication network itself, including improvements in performance and energy efficiency. The latter examines how mobile networks are affected by AI/ML-driven user applications, particularly real-time, latency-sensitive workloads that are best deployed at the network edge, commonly referred to as Edge AI applications.
Within SMARTEN6G, both perspectives were explored in the context of NTNs incorporating satellites, high-altitude platforms (HAPs), and unmanned aerial vehicles (UAVs). A key outcome was the development of an NTN channel emulator based on Digital Twin methodologies, enabling realistic experimentation without direct access to satellite constellations, HAPs, or UAV platforms.
The TEKAID6G project focused on terrestrial networks and enabled the development of the EDGE AI Low Energy (EDGEAILE) testbed for 6G research. Both projects leveraged a bespoke power-measurement toolkit that we designed and implemented to quantify energy efficiency impacts of AI/ML from both network and end-user perspectives.
These presentations summarise the key results from both projects and outline ongoing and future work based on the EDGEAILE platform.
Prof. Ivan Mifsud
Department of Public Law, Faculty of Laws
At the intersection of legal expertise and cutting-edge AI, WATTS represents a transformative leap in technology, transforming predecessor AMPS from TRL4 to TRL-7 deployment-ready status; where AMPS demonstrated bilingual RAG, WATTS merges this with specialised knowledge graph, creating a hybrid graph-RAG engine, promising to revolutionise judicial workflows across higher courts and complex litigation. It will deliver three distinct role-specific dashboards for adjudicators, advocates, and law students. It will extend its data pipelines to cover the court of magistrates, first hall and appeal, develop a graph-RAG architecture that firmly grounds language model outputs in verifiable structured evidence, benchmarks and fine-tune open source LLMs optimised for Maltese legal tasks, and integrate these components into a GDPR and FITA-compliant containerised platform. By combining a transparent, open model and structure role-based access, WATTS will provide an explainable, accessible and ethically sound AI toolset, strongly supporting Malta's digital justice agenda and offering a valuable, replicable blueprint for other small or bilingual jurisdictions.
Dr Jose Pacheco de Almeida Prado
Institute of Aerospace Technologies
This work presents the Natural Language Understanding (NLU) framework of CLEAR (Cockpit Linguistic Engine for Assisted Responses), a hybrid cognitive–computational system developed within the ARTIAP project to support pilots through natural language interaction in safety-critical cockpit environments. CLEAR is designed to operate with on-board, resource-constrained Large Language Models (LLMs), which necessitates a robust interpretation pipeline that combines symbolic, statistical, and constrained generative reasoning.
The NLU architecture adopts a layered approach. At its first level, symbolic linguistic analysis employs finite-state patterns, operational phraseology, and domain-specific grammars derived from FCOM, QRH, and standard cockpit communications to provide deterministic classification of well-formed procedural queries. Frame-based semantics and ontology grounding further map pilot utterances to structured operational frames composed of domain-specific slots (e.g., weather variables, aircraft systems, runway identifiers), ensuring explainable and certifiable interpretation.
When symbolic evidence is insufficient, a distributional layer based on embedding similarity performs intent and domain inference using prototype representations, enabling robust handling of paraphrases, noisy language and informal phrasing. For ambiguous or context-dependent queries, a constrained LLM layer performs pragmatic inference under strict output restrictions, selecting only from validated domains, intents and slot values to prevent hallucination and unsafe interpretations.
The resulting structured representation – domain, intent and operational slots – guides downstream multi-agent reasoning and retrieval. This hybrid NLU design balances interpretability, robustness and computational efficiency, addressing both regulatory constraints and operational variability in cockpit dialogue. By integrating classical NLP, modern semantic embeddings and safety-bounded generative inference, CLEAR demonstrates a viable pathway to trustworthy natural-language interfaces for future human–AI teaming on the flight deck.
Dr Luca Nguyen and Prof. Stefano Moncada
Islands and Small States Institute
Small island states face unique challenges in balancing economic development with environmental protection. In Malta, construction and demolition activities are governed by stringent environmental regulations, particularly S.L. 549.161, requiring pre-demolition environmental audits. However, compliance presents significant barriers for small and medium enterprises: regulatory complexity, resource constraints, and interpretation challenges.
COMPLY addresses this challenge through a regulation-agnostic automated compliance platform. While initially developed to meet Malta's S.L. 549.161 pre-demolition audit requirements, the platform's modular architecture allows adaptation to other environmental regulations and jurisdictions. The system combines retrieval-augmented generation for regulatory interpretation with rule-based logic for structured compliance workflows, guiding users through requirements and generating compliant documentation. This scalability is particularly relevant for EU member states implementing Green Deal directives and environmental compliance requirements across diverse sectors.
Research findings from stakeholder consultations with regulators, environmental consultants, and professional users reveal that automated tools can reduce compliance processing time and costs, making environmental protection economically viable for SMEs. COMPLY demonstrates how regulatory technology can support sustainable development goals (SDGs 9, 11, 12) by reducing barriers to environmental compliance while maintaining regulatory integrity. The project contributes both a scalable tool applicable beyond Malta's construction sector and a governance framework for AI-mediated regulatory interpretation in broader EU contexts.
Prof. Joel Azzopardi
Department of Artificial Intelligence, Faculty of 福利在线免费 and Communication Technology
While Parliamentary Questions (PQs) are vital for democratic oversight, the Parliament of Malta's official archives are often difficult to navigate due to their unstructured nature, dependence on Maltese, and lack of semantic search capabilities. PQ-LENS addresses this challenge by building upon a proof-of-concept prototype (https://pq.ir.mt) that utilised open-weight LLMs, such as Gemma, to automate the scraping, translation, and thematic classification of parliamentary questions from the current (14th) legislature.
While the initial prototype demonstrated the feasibility of an AI-powered dashboard, this project aims to evolve the tool into a production-grade civic infrastructure. Funded by the Digital Technologies Programme 2025, PQ-LENS seeks to create a comprehensive, multilingual dashboard that provides structured insights into decades of parliamentary discourse. Key objectives include:
By hosting open-weight models on local infrastructure, PQ-LENS ensures data sovereignty and demonstrates how AI can process sensitive public data ethically without relying on third-party commercial providers. Ultimately, the project provides a ‘lens’ for researchers, journalists, and citizens to scrutinise governance and engage more effectively with the legislative process.
Prof. Vanessa Camilleri
Department of Artificial Intelligence, Faculty of 福利在线免费 and Communication Technology
The use of artificial intelligence in sports performance analysis has enabled increasingly detailed and automated measurement of athlete behaviour. However, many existing systems rely on opaque models that offer limited transparency to coaches and athletes, constraining trust, interpretability, and practical uptake. This paper presents SWIM-360, an explainable AI-driven framework for swimming performance analysis that places interpretability at the centre of system design.
SWIM-360 integrates monocular video capture with pose estimation and feature extraction to quantify biomechanically meaningful indicators of swimming performance, including stroke rate, left–right symmetry, body alignment, and temporal consistency across laps and turns. Rather than treating these features solely as inputs to black-box predictive models, the system embeds explainability throughout the analytical pipeline. Feature attribution methods, temporal visual overlays, and structured performance summaries are used to link model outputs directly to observable swimmer actions and coaching concepts.
The presentation will focus on the design and implementation of the explainability layer within SWIM-360 and its role in supporting coach-centred interpretation. Preliminary findings from controlled pool trials demonstrate how explainable outputs align with established coaching heuristics and biomechanical reasoning, enabling coaches to interrogate system assessments and engage in reflective dialogue with athletes. Early evaluation suggests that the inclusion of explainability improves perceived usefulness and trust without reducing analytical capability.
By positioning explainable AI as a foundational principle rather than a post-hoc addition, SWIM-360 contributes to responsible, human-centred applications of AI in sport. The work illustrates how explainability can bridge automated analytics and embodied coaching expertise in swimming performance analysis.
Dr May Agius | Co-researchers: Dr Lucianne Zammit, Prof. Christian Colombo, Ms Christine Scholz Fenech and Mr Roger Tirazona
Department of Education Studies, Faculty of Education
As artificial intelligence becomes increasingly embedded in young people's daily lives - from homework assistance to emotional companionship – a critical question emerges: who equips students to navigate this technology ethically and safely? Research consistently identifies teachers as primary mediators of students' digital experiences. Yet, a significant gap exists in understanding whether educators feel prepared to guide learners through the complex terrain of digital trust.
Digital trust in AI is conceptualised as a multifaceted construct reflecting confidence in technology's reliability, ethics, security, and integrity. For teachers, trust encompasses concerns about algorithmic bias, transparency, data privacy, academic integrity, and diminished critical thinking. Research suggests that teacher trust is influenced by AI knowledge, self-efficacy, and cultural values, yet limited attention has been paid to how ethics teachers approach AI trust themselves.
Ethics teachers occupy a unique position as primary mediators of students' moral development, responsible for fostering critical engagement with emerging technologies while navigating their own uncertainties about AI's ethical implications. Understanding their perspectives is crucial, as teachers' attitudes directly influence their willingness to guide students toward responsible technology use.
This qualitative study employed semi-structured interviews with eight ethics teachers from Maltese government secondary schools to explore their perceptions of digital trust in AI across generic contexts. Data was analysed using thematic analysis to identify patterns in how participants conceptualise trustworthiness, what factors influence their confidence in AI systems, and how their digital trust relates to their readiness to teach students about AI ethics.
Findings inform pedagogical approaches supporting teachers' evolving mediating role in AI-integrated education.
Prof. Leonard Busuttil
Department of Technology and Entrepreneurship Education, Faculty of Education
The rapid emergence of generative artificial intelligence (AI) is reshaping teaching, assessment, and academic integrity in higher education. While student engagement with generative AI has attracted growing attention, comparatively less research has examined how university faculty develop the knowledge, confidence, and pedagogical strategies required to integrate these tools meaningfully into their practice. This paper presents the design and ongoing evaluation of a hands-on professional development workshop model to strengthen practical AI literacy among academics.
The workshop design is grounded in Kolb’s experiential learning theory and informed by Rogers’ diffusion of innovations framework. Participants engage in guided, hands-on exploration of generative AI tools alongside structured reflective discussions focused on pedagogy, assessment design, and ethical considerations. Nearly 250 academics have participated across multiple workshop iterations. Pre- and post-workshop questionnaires and facilitated reflections are used to capture participants’ immediate responses to the training.
The primary aim of this study is to determine whether and how faculty translate workshop learning into sustained, meaningful integration of generative AI in their teaching. To address this, a comprehensive follow-up questionnaire examines continued engagement, implementation patterns, and factors influencing adoption over time. The data reveal which approaches persist, which practices faculty attempt, and the specific challenges encountered when adapting assessment to generative AI.
By foregrounding sustained faculty development, this study presents an evidence-informed model that connects technical AI training with pedagogical reflection and examines how professional development is taken up in teaching practice over time.