Yael Feldman-Maggor

Generative artificial intelligence (GenAI) has entered science classrooms, teacher preparation, and research, carrying promise and risk. This editorial introduces a special issue of nine studies, spanning biology, chemistry, physics, and science teacher education across four continents, that trace the arc along which GenAI reaches learners. Where data and models are built, the studies show that synthetic student texts, automated scoring, and formative feedback become trustworthy only when experts encode disciplinary theory into prompts and correct the model's reasoning. Where GenAI meets classrooms, they show that when it drafts lesson plans, differentiates instruction for under-resourced schools, and answers students’ questions, quality varies so widely that human mediation is essential. Beyond the classroom, they document how teacher education courses and eight national systems build educators’ capacity for responsible use. Synthesising this evidence, we read the collection through a Human-in-the-Loop framework: GenAI not as an autonomous instructional authority, but as a socially embedded, fallible partner requiring sustained human judgment. Cross-cutting analysis reveals four tensions: affective uptake outpacing cognitive gain; generic AI literacy versus discipline-specific pedagogy; trade-offs among accuracy, equity, and data sovereignty; and teacher agency that scales up, not away, as models improve. Together, their open questions set the agenda for future research.