Educational software with large language models
Εκπαιδευτικό λογισμικό με μεγάλα γλωσσικά μοντέλα
View/
Keywords
Next-generation educational software ; AI ; Cognitive learning ; Large language models ; Generative Artificial Intelligence ; Retrieval augmented generation ; AI agents ; AI flows ; EducationAbstract
Rapid technological and social changes, coupled with increasing time pressures, have defined a new reality for modern societies and extend to heavily affect the educational process. For example, educators often struggle with time limitations that restrict their capacity to redesign the teaching material in a more engaging, creative, and responsive way that better addresses the individuals’ learning requirements. At the same time, rapid technological advancements such as the advent of Generative AI, have become deeply embedded in everyday life and have fundamentally transformed learners' cognitive patterns and educational preferences (e.g. younger learners, who have typically grown up in digital environments, naturally prefer learning experiences that incorporate technology that allows them to interact with the learning content instead of passively receiving information).
These evolving patterns of both educators and learners impose the need for the development of educational tools that benefit from advanced technologies and enable adaptability to the diverse learning preferences while efficiently supporting teachers in delivering personalized content. Towards this direction, in the present thesis we explore how Generative and (agentic) Artificial Intelligence can empower the next generation of education software design, bridging this gap and ensuring that the teaching and learning processes remain effective and relevant in the face of ongoing societal and technological shifts.
In particular, we developed a series of AI-based tools that dynamically adjust the learning content into different formats, meeting each user's unique learning preferences. These tools employ the use of Large Language Models (LLM) and Retrieval Augmented Generation (RAG) in the context of AI agents and AI flows, that enable the retrieval of information from the given education curriculum and the provision of grounded responses in multimodal forms. For example, we introduce agentic designs that:
• enable the creation of interactive quizzes and quests, gamifying the educational content while empowering knowledge learning and assessment,
• allow for the effective development of learning plans and summaries (with visual and audio support) for addressing the time pressure bottlenecks as well as for covering for the diverse learners’ cognitive styles.
• empower the interactive Q&A sessions, fueled by agentic tools that can handle multimedia input and leverage planning and reasoning in order to assist the learner address inquiries on the learning subjects of their preferences, in a grounded and reliable way
Combined with a prototype implementation of such AI agents, we also provide details on the architectural backbone and functional interplay of these tools and conclude this study by sharing ideas for prospect future further research.