Epistemic curiosity, the intrinsic desire to seek knowledge, plays a pivotal role in fostering engagement and deep learning, especially when students navigate complex topics such as artificial intelligence (AI). This study examines secondary school students’ epistemic curiosity patterns as they engage with machine learning for text classification in five one-hour lessons. At the end of each lesson, students shared their wonderings about key concepts, allowing us to trace the evolution of their epistemic curiosity over time. Mixed methods analyzed temporal changes in curiosity patterns and thematic topics with gender-based differences. We found students’ epistemic curiosity declined significantly after transitioning to more technical AI topics, marking a turning point in engagement. More than one-third of students maintained curiosity, especially at the medium level, while more students experienced downward than upward shifts. Moreover, students who maintained low curiosity or declined from high/medium to low curiosity were at higher risk of struggling with AI concepts. Students were curious about a wide range of technical and non-technical topics throughout the model life cycle - design, implementation, deployment, and evaluation. Notably, female students tended to be curious about how AI makes decisions, asking about feature representations, suggesting a deeper interest in AI’s reasoning process and interpretability. In contrast, male students focused more on model design, inquiring about the original purpose, rationale behind the design, and aspects of model evaluation such as accuracy and scalability. This study provides insights on designing AI education to foster curiosity and engage a broader range of students.