Tangible AI in the Elementary Classroom: Effects on AI Literacy, Computational Thinking, and Learner Dispositions

Artificial intelligence (AI) is increasingly ubiquitous, yet AI literacy remains underdeveloped in elementary education, where young learners often view AI in anthropomorphic or magical terms. This study examined changes in computational thinking (CT), AI concept mastery, learning motivation, and programming self-efficacy following a tangible AI intervention, and explored the teacher's implementation experience. Guided by a framework integrating the AI Literacy Framework and the SEAME Model, a four-week intervention (4 × 60 minutes) was implemented with 36 fifth-grade students and one teacher in a public elementary school, using a single-group pretest–posttest design within a mixed-methods approach. Quantitative data were analyzed using repeated-measures ANCOVA, with prior computing experience as a covariate; qualitative data from semi-structured teacher interviews (during and after the intervention) and observations of all four sessions were analyzed thematically. Significant pre–post gains were found for all four composite outcomes, largest for AI concept mastery (M = 8.4 to 15.1; η²p = .70) and total CT score (η²p = .56), although the CT abstraction subdimension did not improve (M = 2.3 to 2.1). Students able to explain how machine learning models learn from data rose from 28% to 83%. Anthropomorphic descriptions of AI decreased from 89% to 25% of students, with post-intervention explanations increasingly referring to sensors, data, and training. Three qualitative themes emerged: strong pedagogical confidence but limited AI content knowledge, persistent infrastructural constraints, and reduced technical anxiety through gradual mastery. Tangible AI may offer a promising bridge to early AI literacy; however, without a control group, changes cannot be causally attributed to the intervention, and these single-school findings are preliminary.