Introduction

The rapid advancement of artificial intelligence (AI) has ushered in a new era of technological innovation characterized by the development of models capable of automating and enhancing human tasks with remarkable efficiency. AI systems, such as ChatGPT, DALL-E, and Perplexity, have gained widespread popularity, transforming how society interacts with technology [1,2]. These platforms and others have made cutting-edge AI capabilities accessible to a broad audience, demonstrating their potential to impact nearly every facet of human life. In education, the influence of AI is particularly profound, as it introduces novel opportunities and challenges that are reshaping the educational landscape [3,4].

Integrating AI into education brings forward the potential to enhance teaching methodologies and learning experiences. For educators, AI offers tools to design personalized instructional materials tailored to the needs of individual students, including those with disabilities [5]. Recent studies have highlighted how AI-driven systems support teachers in automating time-intensive tasks such as grading and assessment, allowing them to focus more on pedagogical strategies [6,7,8]. From the learners' perspective, intelligent tutoring systems provide personalized, realtime assistance, fostering improved engagement and skill acquisition across diverse contexts [9]. These AI applications extend beyond traditional curricula, aiding in developing everyday competencies and mitigating risks associated with digital environments [10].

One significant area where AI intersects with education is addressing societal and global challenges, such as those outlined in the United Nations Sustainable Development Goals (SDGs) [11]. For instance, AI has shown promise in promoting inclusive education by generating adaptive materials for students with special needs, including automatically generated concept maps [12,13]. Moreover, AI-driven intelligent companions can help students navigate online risks, such as exposure to toxic content on social media, while fostering digital literacy and critical thinking skills [14]. Such applications underscore AI's role in advancing SDG 4, emphasizing inclusive and equitable quality education for all [15].

Despite these benefits, the adoption of AI in educational contexts is not without its challenges. Recent literature has widely discussed concerns about the misuse of generative AI models, the amplification of biases, and the dissemination of incorrect information [16,17,18]. Generative models capable of performing complex tasks such as essay writing and image creation raise ethical and pedagogical questions about their impact on student learning and academic integrity [19]. Thus, educators and students alike must develop a nuanced understanding of AI's capabilities and limitations to ensure its ethical and practical use in educational settings [20].

Building on the principles outlined in the UNESCO Beijing Consensus, the AIxEDU workshop seeks to address these issues by providing a platform for exploring the transformative potential of AI in education. It aims to foster interdisciplinary dialogue and encourage contributions investigating the development and deployment of AI systems in educational contexts. The workshop highlights the need for innovative approaches to human-AI collaboration, promoting inclusive practices, and the ethical management of educational data. Furthermore, it emphasizes the importance of explainable AI in fostering trust and transparency, adaptive lifelong learning, and personalized educational experiences [21,22,23]. By bringing together researchers and practitioners, the workshop aspires to chart a path forward that maximizes the benefits of AI in education while addressing its inherent challenges.

Workshop's Contibutions

We received 37 submissions for the AIxEDU workshop, and after a rigorous review process, 16 were accepted for presentation. In addition to submissions from Italian authors, accepted papers included contributions from researchers in Sweden, Slovenia, Ukraine, France, and Finland. This highlights the international significance of the topic of AI in education and underscores the Italian Conference on Artificial Intelligence as a pivotal event for discussing AI-related issues, not only for Italian researchers but for the global academic community.

As shown in Figure 1, the workshop contributions span diverse topics, each reflecting critical aspects of education and learning. At the core, learning and education remain the central themes, emphasizing the processes and systems through which knowledge is acquired and shared. Generative AI, a transformative technology, plays a pivotal role by enabling the creation of personalized content and innovative instructional materials while raising important ethical and pedagogical questions. Assessment emerges as another vital focus, with AI offering solutions to automate grading and provide insightful feedback, thereby enhancing the evaluation process. Educators are integral to this landscape, supported by AI-driven tools that simplify routine tasks and allow for tailored teaching approaches. The workshop also highlights the importance of tools and technologies that improve accessibility and engagement, as well as challenges that must be addressed to ensure equitable and effective adoption of AI in educational settings. Participants signify the collaborative and interdisciplinary nature of these discussions, engaging diverse stakeholders in shaping the future of AI in education. Together, these topics underscore the multifaceted impact of AI on the educational landscape and the need for collaborative exploration to harness its potential responsibly.

Session1: AI in Teaching and Learning Practices

The paper Enhancing Instructional Design: The Impact of CONALI Ontology and ChatGPT in Primary Education Training [24] explores the combined use of the CONALI Ontology and ChatGPT in enhancing the instructional design process for primary education. The study involved 110 students at the University of Foggia who utilized these tools to design learning units (LUs). Findings revealed that the structured framework of the CONALI Ontology facilitated the clear articulation of SMART objectives, while ChatGPT augmented the process by promoting efficiency and creativity in idea generation. Participants noted both benefits and challenges, including the need for critical evaluation of AI-generated content. The study underscores the transformative potential of AI when integrated with pedagogical frameworks to train educators in developing technology-enhanced learning experiences.

The paper Generative AI for Teaching Latin and Greek in High School [25] examines the application of generative AI, particularly ChatGPT, in teaching Latin and Ancient Greek to high school students, focusing on its potential in Italian Liceo Classico settings. The study highlights the dual response of resistance and acceptance in educational institutions regarding generative AI. By using examples and prompts, the research demonstrates how ChatGPT can serve as a valuable ally, aiding in grammatical analysis, translation, and contextual understanding of classical texts. Emphasizing a constructivist approach, the paper advocates for the cautious yet proactive integration of AI to complement traditional pedagogies, thereby enriching the study of classics without replacing teacher-student interactions.

The paper Fostering Metacognitive Skills in Programming: Leveraging AI to Reflect on Code [26] introduces the Reflective AI Programming Lab (RAP Lab), a novel pedagogical framework designed to enhance metacognition and reflective learning in computer science education. By having students collaborate on programming tasks using AI-generated code under structured constraints, the lab encourages them to articulate problem-solving strategies and critically evaluate AI assistance. AI tools act as impartial observers, fostering a safe environment for experimentation and error analysis. The approach bridges the gap between educational and industry practices, emphasizing the development of higher-order cognitive skills crucial for AI-assisted programming.

The paper Speeding Up Design and Making to Reduce Time-to-Project and Time-to-Market: An AI-Enhanced Approach in Engineering Education [27] details the integration of AI tools such as ChatGPT and GitHub Copilot in an advanced engineering course focused on embedded systems. Students employed AI for tasks including design optimization, coding, debugging, and documentation, significantly reducing development time. The approach emphasized promptbased learning, which honed students' ability to communicate effectively with AI systems. While AI enhanced efficiency and supported decision-making, students retained control over critical aspects of their projects, demonstrating the synergy between human creativity and AI-driven solutions in accelerating innovation.

The paper Enhancing Digital Object-Based Learning with Haptic Gloves and Virtual Reality in the Virtual Museum of Tor Vergata [28] explores the RESTART project at the University of Tor Vergata, which integrates digital object-based learning (OBL) with virtual reality (VR) and haptic gloves to enrich cultural heritage education. The project employs advanced immersive technologies to foster the development of transversal skills, including creativity and critical thinking, while also enhancing participant well-being. Through AI-adapted experiences and inclusive design, the initiative ensures accessibility for diverse users, including those with disabilities. The research highlights how technology-driven OBL promotes engagement and understanding of cultural content while fostering social inclusion and psychological well-being.

Session 2: AI in Assessment and Evaluation

The paper EULER: Fine-Tuning a Large Language Model for Socratic Interactions [29] presents an innovative approach to enhancing educational utility in large language models (LLMs) by fine-tuning them to support Socratic dialogue rather than providing direct answers. Using Direct Preference Optimization (DPO) and datasets containing Socratic conversations, the study demonstrates that models can be trained to guide students toward discovering solutions independently. The methodology involves processing educational datasets to align with Socratic principles and evaluating the tuned models against advanced systems like GPT-4o. Results suggest that this fine-tuning approach significantly improves the models' ability to foster critical thinking in educational contexts.

The paper Large Language Models for the Assessment of Students' Authentic Tasks: A Replication Study in Higher Education [30] explores the use of LLMs in evaluating student-written tasks within a higher education context. Building on previous research, this replication study employs updated models to assess student submissions using rubrics developed by educators. By comparing evaluations from LLMs and human experts, the study examines the accuracy and reliability of AI in academic assessment. Findings indicate that specific adaptations and guidelines are essential for achieving reliable assessments, highlighting the potential of AI to enhance scalability and sustainability in educational evaluation.

The paper Uninvited Generative AI Has Joined Our Students: Tackling Disinformation and Creating Content with the Help of Generative AI Apps [31] details a workshop aimed at enhancing students' information literacy and critical thinking skills through generative AI applications. Conducted across multiple high school grades, the workshop involved students using AI tools to address disinformation and create digital content. The study highlights students' native interaction with generative AI and their growing awareness of its capabilities. Results underscore the need for integrating media literacy and ethical considerations into curricula to address the challenges and opportunities presented by AI in education.

The paper Redefining Education: A Personalized AI Platform for Enhanced Learning Experiences [32] discusses the PROSPETTIVA project sponsored by the Sicilian region (Italy), which aims to enrich secondary education by incorporating AI technologies. The web-based PROSPET-TIVA platform leverages advanced Large Language Models to create personalized learning environments that actively engage students and enhance AI literacy. Key functions of the platform include summarization, simplification, and the creation of concept maps, which are selected for their ability to support in-depth learning and critical thinking. The platform's design facilitates seamless switching between different LLMs, allowing it to adapt to various educational needs efficiently. This flexibility, combined with targeted functionalities, ensures that the platform provides high-quality educational experiences. Initial findings suggest this approach significantly enhances learning outcomes and helps bridge educational inequalities. Future work will refine these functionalities and expand the platform based on user feedback.

The paper Artificial Intelligence in the Professional Activity of a University Lecturer in Ukraine: Realities and Prospects [33] examines the integration of AI tools into the professional tasks of Ukrainian university lecturers. Through a comprehensive survey of 205 educators, the study evaluates the current use and attitudes toward AI in handling routine educational, methodological, and scientific tasks. Findings reveal varying levels of AI literacy and adoption, emphasizing the potential of AI to streamline administrative functions while encouraging deeper exploration of its pedagogical applications. The paper calls for targeted training and guidelines to support effective AI integration in academia.

The paper Teacher-AI interaction in the selection of target texts [34] investigates the integration of an Artificial Agent (AA) in the student assessment correction process and the delivery of online recursive feedback within educational settings. The main objective is to enhance the efficiency and effectiveness of feedback to improve the learning process. This research primarily examines the interaction between the AA and a human agent during the correction phase. Methodologically, the study employs an AI-driven approach, utilizing a BERT model specifically tailored for the Italian language to analyze text-based student responses. The AI system categorizes and ranks student submissions by comparing them with predefined target texts (TTs), using measures like cosine similarity to assess the degree of resemblance. The main results indicate that about 80% of student responses were reliably assessed using this AI approach. The AI system is particularly adept at identifying which texts require further human evaluation and helps highlight the uncertain elements in the evaluations. It suggests that while the AI can significantly support the correction process, there are still areas, particularly regarding subjective assessments, where human intervention remains crucial.

Session 3: AI in Educational Systems, Data Analysis, and Challenges

The paper Co-Designing Media Education Strategies: A Workshop on AI and Information Literacy [35] examines a pilot workshop conducted under the TADAM project to address the role of generative AI in media education. This workshop, hosted at the MED Summer School, engaged educators and researchers in exploring critical aspects of AI's impact on media production and misinformation. Using Merrill's instructional design principles, participants engaged in hands-on tasks, such as generating and analyzing fake news, to foster critical thinking and media literacy. The outcome included collaboratively developed guidelines for "critical prompting," promoting informed citizenship and the effective use of AI in educational and media contexts.

The paper From Training KPIs to Learning KPIs: Ensuring Effectiveness in Learning Processes Through Predictive Analytics and Data-Based Tutoring Actions [36] presents the LearnalizeR framework, a data-driven model for enhancing online course effectiveness. This model employs predictive analytics and macro performance indices to identify critical issues in learning environments and guide tutoring interventions. The framework's application in two real-world case studies demonstrates its potential to improve participation, mitigate risks, and optimize course design. The findings underscore the importance of transitioning from traditional trainingfocused KPIs to learning-centered evaluation metrics for more impactful digital education.

The paper Does Time Matter in Analyzing Educational Data? A New Dataset for Streaming Learning Analytics [37] introduces a novel temporal dataset derived from the Open University Learning Analytics Dataset (OULAD). This dataset incorporates time-based interaction data, enabling advanced stream algorithms to analyze learning patterns in real-time. Experiments with Online Bagging algorithms reveal that temporal structures significantly enhance the predictive capabilities for student outcomes, particularly for the majority class. However, challenges remain in addressing minority classes. This research highlights the dataset's potential to support dynamic and responsive interventions in Virtual Learning Environments.

The paper NoVAGraphS: Towards an Accessible Educational-Oriented Dialogue System [38] discusses the development of NoVABOT, a dialogue system designed for visually impaired individuals to explore graphical structures like UML diagrams and circuits. Utilizing AIML technology, the system translates graphical data into interactive textual dialogues accessible through screen readers. Preliminary evaluations with visually impaired students and their teachers indicate the system's effectiveness in facilitating understanding and problem-solving in various academic disciplines. This project underscores the importance of inclusive educational technologies in ensuring equitable learning opportunities.

The paper Evolution of LMS Design and Implementation in the Age of AI and Large Language Models [39] critiques the outdated architecture of current Learning Management Systems (LMSs) and proposes a shift to graph-based models integrated with AI functionalities. By transforming content storage into a knowledge graph, the proposed architecture enables advanced AI features such as semantic searches and personalized learning. The study highlights challenges in integrating multi-domain ontologies and adapting to evolving educational demands while showcasing the transformative potential of AI-driven LMS platforms.

Organizing Committee