Navigating the N-shaped path: Nonlinear development of pre-service PE teachers' professional competencies in AI-driven STEAM instruction.

Abstract

This study investigates the longitudinal developmental trajectories of pre-service physical education (PE) teachers' core competencies within a 16-week generative AI-driven STEAM basketball curriculum. Employing a parallel-group randomized controlled trial in which 80 participants were randomized and 75 completed the intervention (experimental n = 37; control n = 38), the study addresses the absence of longitudinal evidence on how Intelligent-TPACK develops over time by identifying a non-linear N-shaped trajectory across the three core competency domains, with two sub-dimensions (Professional Responsibilities, Ethics) as theoretically informative exceptions. Piecewise linear mixed-effects models show a Week-8 decline at peak instructional complexity followed by recovery; the between-group phase contrasts describing this pattern remain significant after correction for multiplicity. Two further patterns are reported as exploratory and hypothesis-generating: higher baseline AI proficiency was associated with a steeper Week-8 decline in classroom management (feature creep), and recovery co-occurred with reduced in-class AI activation (pedagogical simplification rebound). Intelligent-TPACK, interdisciplinary teaching ability and observed teaching practice were associated within persons, but because they were measured on the same occasions this is not evidence of a causal cascade. We offer the Generative AI-TPACK Fatigue Model as a hypothesis-generating framework rather than confirmed theory. The study offers preliminary, context-sensitive guidance for teacher education programs working in comparable settings, and identifies the conditions under which a mid-course performance decline might prove productive as a question for future work.