The integration of AI into interactive learning environments and digital learning platforms is seen as having great potential to enhance students' willingness and ability to learn. Applications such as ChatGPT or advanced AI tutoring systems are based on the assumption that they are able to follow and correctly interpret and analyze students' work in the environment. In this study, we present the results of eye-tracking 48 elementary school students while solving mathematics problems on a digital learning platform, followed by semi-structured interviews. The students' facial expressions were analyzed and interpreted as part of an emotion analysis while solving specific tasks. Some of the items in the digital learning platform were designed as digital game-based learning and included incentive systems such as punishments and rewards. Other items were designed as a maze where students could only progress by correctly solving certain tasks. Items provided a range of feedback, from confirmation to elaboration feedback, and some included rich graphics or problem-solving scaffolding. The results of this extended eye tracking and emotion analysis study showed that many students interacted with the items differently than expected, such as not paying attention to the incentive system in the game-based items or not using the help button or instructional videos. Sometimes students did not look or read what they were expected to read, and other times they did not use the problem-solving scaffolding tools as expected. This study provides evidence of the complexity of students' responses and interactions with digital learning items and tasks in learning environments and raises awareness of what we know and what we do not know and assume happens.

Keywords: Digital learning platform, eye tracking, facial emotion analysis, mathematics digital items, primary education.