How to Imagine Educational <scp>AI</scp> : The Filling of a Pail or the Lighting of a Fire?

Abstract Recent advances in artificial intelligence (e.g., machine learning, generative AI) have led to increased interest in its application in educational settings. AI companies hope to revolutionize teaching and learning by tailoring material to the individual needs of students, automating parts of teachers' jobs, or analyzing educational data to optimize the delivery of content. This article outlines the role of imaginaries in shaping concrete practices and understandings of educational AI. By building on an expanded notion of sociotechnical imaginaries, we show, through a review of science‐fiction narratives about technology and education, that the imaginaries underlying educational AI are lacking or limited. When imagined, educational tools are most often depicted as systems for the efficient transmission of information. We contend that this poor imaginary affects why, how, and what educational AI systems are developed. As we demonstrate by analyzing the paradigmatic ideas espoused by Salman Khan, educational AI is designed to optimize the acquisition of knowledge and largely ignores other aspects of teaching and learning, such as care, play, or curiosity. We observe that educational AI is not an object of lively philosophical debate—unlike, for example, self‐driving cars or autonomous weapons, which are, incidentally, also much more present in the popular imaginary. We conclude by discussing imaginaries of educational AI that would offer an alternative to the dominant paradigm, focusing on Neal Stephenson's The Diamond Age , Kazuo Ishiguro's Klara and The Sun , and Ian McEwan's Machines Like Me . We argue that to develop socially beneficial and desirable educational technologies, we need to look beyond AI as an information system, and we point to three narratives that present alternative imaginaries of educational AI that can allow us to conceptualize new desirable forms of incorporating technology in teaching and learning, as well as point to its limitations.