Systematic Approach to Creating Educational Content Using Artificial Intelligence in Multi-agent Systems

Abstract

The article presents a systematic review of artificial intelligence methods for educational content generation in multi-agent systems. Modern approaches to the automatic generation of educational questions, lesson plans, and materials were analyzed using large language models. Four main groups of generation methods were distinguished: generation of educational questions and tasks using Bloom's taxonomy classification; generation of lesson plans using the step-by-step reasoning-prompting technique; generation with parametric control of complexity level; and pedagogically focused approaches that take into account the principles of constructivism in education. Proposed four types of classification of architectures of educational content generation systems: single-agent, hybrid, and multi-agent. Each type is characterized by the coordination approach, verification mechanisms, scalability, and content quality. It is shown that a significant proportion of systems positioned as multi-agent in fact belong to pipeline or hybrid architectures. Also, existing multi-agent content generation systems were analyzed, and the feasibility of using a multi-agent approach to build adaptive educational systems was demonstrated. A conceptual model of a multi-agent system for generating educational content is proposed, covering the ontological level, the content generation level, learner models, adaptation and personalization, agent coordination, quality assessment, and feedback. A reference architecture with five subsystems is proposed: planning, generation, evaluation, adaptation, and coordination. Key open issues are identified, critical risks of applying generative artificial intelligence in education are determined, and potential directions for further scientific research are outlined.