The Role of Generative AI in Enhancing Primary School Science Achievement: A Conceptual Paper

Generative artificial intelligence (GenAI) has attracted sustained interest in science education for its capacity to personalise explanations, render submicroscopic processes visible, and maintain learner engagement. This conceptual paper theorises how GenAI can strengthen primary pupils’ conceptual understanding in abstract topics (for example, the particulate nature of matter) and specifies the conditions under which it improves science achievement. Drawing on constructivist and sociocultural perspectives (Piaget; Vygotsky) together with the Cognitive Theory of Multimedia Learning (CTML) and Cognitive Load Theory (CLT), the paper advances the AAE framework (Adaptive scaffolding, Active multimodal processing, and Equitable personalisation) as the core set of mechanisms that link GenAI design features to learning outcomes. From this framework, we derive testable propositions: (a) aligning narration with animation and enabling learner controlled segmenting reduces extraneous load and increases germane processing; (b) adaptive sequences of predict, observe, and explain repair common misconceptions; and (c) personalised pacing or modality combined with structured collaborative scripts promotes equitable participation and achievement. A staged research agenda is outlined, including curriculum aligned two tier concept measures. The paper concludes with pedagogical and ethical implications for teachers (orchestration, verification of AI generated content), institutional leadership and policy makers. By consolidating theory, design principles, and propositions, the paper provides a roadmap for rigorous and context sensitive evaluation of GenAI in primary science.