Publications
Permanent URI for this collectionhttps://repositorio.grial.eu/handle/123456789/34
Browse
884 results
Search Results
Item First-Month LMS Behaviours Associated with Success in an AI-Permissive Software Engineering Course(Asociación para el Desarrollo de la Informática Educativa (ADIE), 2026-09-02) Vázquez-Ingelmo, Andrea; Paule-Ruiz, M. Puerto; García-Peñalvo, Francisco José; García-Holgado, AliciaThis extended abstract profiles the highest-performing behavioural cluster during the first milestone of a Software Engineering course governed by an explicit artificial-intelligence (AI) usage policy. Moodle traces from two post-AI cohorts (n = 146) yielded three clusters: an integrator group (n = 32), an intermediate group (n = 73), and an at-risk group (n = 41). Integrators achieved a higher mean milestone grade than at-risk students (6.37 versus 4.81; Hedges’ g = 0.92) and a higher pass rate (71.9% versus 44.0%). Their profile combined access volume and anticipation: they accessed theoretical materials, task specifications, rubrics, and modelling tools more often, and reached theoretical materials about ten days earlier. A logistic model based on theoretical-material volume, task volume, and task first-access timing separated integrators from the remaining students with an ROC AUC of 0.86. The results support first-month onboarding that introduces rubrics and modelling tools early, triggers early task access, uses analytics for positive recognition, and treats forum activity as a complement rather than a substitute for task- and rubric-anchored engagement.Item Asistentes de aprendizaje con inteligencia artificial en Ingeniería del Software: diseño y percepción del alumnado(Asociación para el Desarrollo de la Informática Educativa (ADIE), 2026-09-02) García-Peñalvo, Francisco José; Conde, Miguel ÁngelEste trabajo analiza el uso de dos asistentes de aprendizaje basados en inteligencia artificial en una asignatura de Ingeniería del Software. El primero adopta una estrategia socrática y guía al alumnado mediante preguntas, pistas y reflexión progresiva. El segundo, construido con LAMB, proporciona respuestas directas fundamentadas en materiales docentes seleccionados por el profesorado. A partir de los registros de uso y de 53 respuestas anónimas, se estudian la adopción, la utilidad percibida y las preferencias de aprendizaje. Los resultados muestran que ambos asistentes fueron valorados positivamente, aunque respondieron a necesidades diferentes: el enfoque socrático favoreció el razonamiento conceptual, mientras que LAMB fue preferido para el repaso y la preparación de exámenes. Los hallazgos subrayan la necesidad de ajustar el diseño de estos sistemas al momento del aprendizaje, la presión evaluativa y las expectativas del alumnadoItem Hybrid intelligence and educational agency: Towards an ecology of human-AI collaboration(Ediciones Universidad de Salamanca, 2026-09-23) García-Peñalvo, Francisco JoséThe emergence of generative artificial intelligence has revived an enduring question in educa-tional technology: Should artificial intelligence (AI) supplant teachers or support them? The evi-dence does not sustain either pole. This article advances hybrid intelligence, an integrated mode of human-AI collaboration in which agency, judgment, and accountability are shared rather than lodged in just one side, as a more useful lens for understanding AI in education. Using a theoretical and argumentative method, the paper defines the concept. It separates hybrid intelligence from adjacent ideas, including blended learning and educational AI, and links its philosophical lineage to extended mind theory, man-computer symbiosis, and actor-network theory. The article then examines three major conceptual frameworks: Molenaar’s six-level automation model, Cukuro-va’s AIED-HCD taxonomy, and Holstein, Aleven, and Rummel’s hybrid adaptivity model. These are brought together in a five-level architecture for hybrid learning ecosystems: people and roles, hybrid space, data, models, and governance. In addition, it sketches three scenarios for the grad-ual incorporation of AI into university teaching, together with the nine persistent challenges (the “elephants in the room”). The discussion highlights shared patterns across the frameworks and the teaching competencies required for hybrid intelligence. The article ends by arguing that, in education, hybrid intelligence can realise its potential only through a human-centred orientation: technology must adapt to education, not education to the technology.Item Model of Characterization of Teamwork Competence Based on Three Types of Capabilities(Springer, 2025-07-15) Fidalgo-Blanco, Ángel; Sein-Echaluce, María Luisa; Fonseca, David; García-Peñalvo, Francisco Joséhere are various teamwork models with different orientations regard-ing the conceptual model, member involvement, evidence management, and even the training process for acquiring teamwork-related skills. This research defines a hybrid model that integrates the two main theoretical models (focused on group achievements and team member involvement) and an open-box method (with continuous generation and verification of both group and individual evidence). Therefore, teamwork competence is associated with a set of capabilities of dif-ferent types, classified into three main categories: group, individual, and general (soft skills), which are related to teamwork but not exclusive to it. This paper also presents the evidence that allows for continuous and transparent training and evaluation of these three types of capabilities.Item Integrating Individual and Collective Skills: A Rubric-Based Model for Teamwork Competence Assessment(Springer, 2024-06-29) Sein-Echaluce, María Luisa; Fidalgo-Blanco, Ángel; García-Peñalvo, Francisco José; Fonseca Escudero, DavidThe competence of teamwork comprises a set of skills that enable the assessment of teamwork evolution (collective skills) and the involvement of each team member (individual skills). In most research works, these skills are grouped without making this distinction between collective and individual skills. In this study, collective skills are associated with the different phases that constitute the evolution of teamwork, allowing for the identification of the precise moment when such a skill should be applied. Individual skills are applied in all phases of teamwork, as they measure individual involvement and responsibility, along with the competencies necessary at an individual level to develop teamwork. This work presents a rubric that associates phases, evidence, technology, and indicators and allows educators to measure the degree of acquisition of each and collective skill. The method used for the development of teamwork has been the Comprehensive Training Model of the Teamwork Competence, which supports both the continuous and transparent creation of evidence of teamwork development by the teams and each of their members, as well as the continuous monitoring of this development by educators.Item Enhancing Learning Assistant Quality Through Automated Feedback Analysis and Systematic Testing in the LAMB Framework(Springer, 2025-06-22) Alier-Forment, Marc; Pereira-Valera, Juanan; Casañ-Guerrero, María José; García-Peñalvo, Francisco Joséhe Learning Assistant Manager and Builder (LAMB) is an open-source software framework that lets educators build and deploy AI learning assis-tants within institutional Learning Management Systems (LMS) without cod-ing expertise. It addresses critical challenges in educational AI by providing privacy-focused integration, controlled knowledge bases, and seamless deploy-ment through standard protocols. This paper presents major enhancements that enable systematic quality assurance and continuous improvement of these learning assistants. The new LAMB includes mechanisms for structured feedback on real-world assistant behavior, transforming it into a test suite with curated prompts and expected correct or incorrect responses. When changes are made—such as prompt engineering, retrieval-augmented generation optimization, or knowledge base expansions—this suite enables automated validation of their impact. A key innovation is using frontier large language models (LLMs) to evaluate responses automatically, generating detailed reports that reveal improvement areas and confirm performance gains. This systematic feedback-driven testing fosters continuous refinement while preserving quality standards. Validation studies show measurable boosts in reliability and consistency. In various educational contexts, the framework identifies edge cases, maintains con-sistency across iterations, and provides actionable insights. Automated testing is especially beneficial for assistants with extensive knowledge bases and complex interaction patterns. This work advances educational AI by providing a robust methodology for quality assurance and ongoing improvement of learning assistants. Its structured feedback and automated evaluations ensure alignment with educational goals while refining assistants over time. The enhanced LAMB framework offers a scalable and reliable solution for educators aiming to integrate AI-driven support into their LMS environments.Item Refactoring User Interfaces Through a Data-Driven Framework: a Case Study in the Health Domain(IEEE, 2023-10-16) Vázquez-Ingelmo, Andrea; García-Holgado, Alicia; García-Peñalvo, Francisco José; Pérez-Sánchez, Pablo; Antúnez-Muiños, Pablo; Sánchez-Puente, Antonio; Vicente-Palacios, Víctor; Dorado-Díaz, Pedro Ignacio; Sánchez, Pedro LuisUser interfaces (UIs) play a crucial role in defining user experiences and influencing the success of software products. While UI design has traditionally been subjective and iterative, data-driven approaches are becoming increasingly popular to ensure that Uis meet user needs and expectations. However, contextual factors such as the application domain can present challenges for designing Uis that are both effective and efficient. This is particularly true in the health domain, where Uis must be adapted to specific tasks and user expertise to maximize the support provided by software systems. Moreover, the urgency of delivering fully functional systems in short periods can relegate UI design to a second plane. This paper presents a framework proposal for refactoring and improving Uis using a data-driven approach, providing an efficient and systematic methodology to address not solved UI issues introduced during previous software development processes. The proposed framework has been successfully applied to two medical platforms, demonstrating the importance of data-driven approaches for UI refactoring in domains with particular necessities.Item D-AI-COM: A DICOM Reception Node to Automate the Application of Artificial Intelligence Scripts to Medical Imaging Data(Springer, 2024-05-01) Vázquez-Ingelmo, Andrea; García-Holgado, Alicia; García-Peñalvo, Francisco José; Pérez-Sánchez, Pablo; Sánchez-Puente, Antonio; Vicente-Palacio, Víctor; Dorado-Díaz, Pedro Ignacio; Sánchez, Pedro LuisArtificial Intelligence (AI) has proven to be useful in several fields. The medical domain is one of the fields that benefits from the application of AI methods to automate and ease complex tasks including disease detection, segmentation, assessment of organ functions, etc. However, applying these kinds of methods to the variety of data formats involved in health contexts is not trivial. It is necessary to provide technologies that enable non-expert users to benefit from AI applications. This work presents a platform that acts as a DICOM reception node with the goal of automating the application of AI algorithms to medical imaging data. This platform is set to ease the process applying AI to their DICOM images by making the whole process transparent and straightforward for users without AI-related or programming skills.Item AI-Powered DICOM Image Segmentation: A Collaborative Platform for Continuous Expert Feedback(Springer, 2026-03-01) Santos-Blázquez, Pablo; Vázquez-Ingelmo, Andrea; García-Holgado, Alicia; García-Peñalvo, Francisco José; Sánchez-Puente, Antonio; Sánchez, P. L.his work presents the development of an interactive web platform that integrates deep learning techniques for the segmentation of cardiac ultra-sound (echocardiogram) images. The platform incorporates a Picture Archiving and Communication System (PACS) to facilitate the seamless visualization, anno-tation, and automated processing of DICOM images. The web platform features an intuitive interface that allows healthcare professionals to interactively annotate medical images, providing feedback that directly informs model improvements. The system’s retraining workflow ensures that AI-driven segmentation remains adaptable to real-world clinical needs. These findings underscore the importance of iterative AI model refinement through expert feedback, paving the way for more reliable and personalized medical image analysis.Item Management and Application of AI to DICOM Image Processing: A Systematic Mapping Literature Review(Springer, 2024-07-15) Fraile-Sanchón, Rubén; Vázquez-Ingelmo, Andrea; García-Peñalvo, Francisco José; García-Holgado, AliciaArtificial intelligence (AI) has the potential to bring unprecedented benefits to humankind. Therefore, it is worth investigating how to maximize these benefits while avoiding potential pitfalls. Given this context, the first task necessary to assess the potential of this approach is to understand the management landscape and the application of AI to DICOM image processing. In this case, the researchers employ a systematic mapping review. This paper presents this process and its main findings. 35 studies have been selected from a total of 154 analyzed. From them, in addition to obtaining a clear view of the application of AI to DICOM images, we can also conclude that pre-trained AI algorithms are used in a higher amount than non-trained algorithms in terms of DICOM image usage.