Elaboração de itens de matemática para avaliações de larga escala via inteligência artificial generativa

Abstract

This research investigates the feasibility of using Generative Artificial Intelligence, specifically the Gemini model, as an auxiliary tool in the development of mathematics items for large-scale assessments, aligned with the Reference Matrix of the Educational Assessment System of Pernambuco (SAEPE). Faced with the scarcity of teachers' time and the high technical rigor required to formulate these questions, the study aimed to analyze whether the machine can optimize the creation of evaluative material. Characterized as applied, exploratory, and qualitative research, the methodology included a literature review and an experimental phase. In this empirical stage, restrictive prompts were developed to generate items focused on 9th-grade elementary school descriptors. The results demonstrated that the tool is capable of mimicking the structure required by official guides, formulating contextualized stems and distractors with precision by simulating real student errors. However, structural flaws and algorithmic hallucinations were observed in some responses. It is concluded that Artificial Intelligence is a powerful and viable educational tool to optimize planning time, provided its use is conditioned to the teacher's curation. The teacher's role is transformed, making their role essential in the validation and technical-pedagogical refinement of the machine-generated content.