Three Blocks to Build Practical AI: Teachers’ Perceived Feasibility of Conversational Human–Machine Cooperation in K–12 Programming Education

Abstract

Education systems are bringing artificial intelligence (AI) into K–12 curricula, but block-based environments, as usually configured, keep young learners at “toy” exercises. In this study, we examine whether a minimal interface backed by expanded machine capability changes teachers’ judgments of what students can build. In the enhanced Tooee environment for Scratch, students “talk” to a virtual agent with only three blocks. Students delegate complex operations to the machine but write the program logic themselves. The environment supports four data types and three practical capabilities, namely large-scale data processing, automated data collection, and inter-program control. Twenty-five in-service teachers used it throughout a fifteen-session graduate course. In one end-of-course survey, they rated the perceived feasibility of students implementing eight types of AI applications with this environment and with the block-based environments they had used before. Ratings for Tooee were significantly higher for all eight types, and the gains were largest for the three practical capabilities. In the open-ended responses, teachers highlighted the conversational interface, expanded functionality, and immediate feedback, while a small number of them noted runtime errors and concerns about age-appropriateness. In the judgment of practicing teachers, conversational human–machine cooperation was associated with substantially higher perceived feasibility of practical AI application development. As the next step, classroom studies with students are necessary.