Abstract
Conceptual alignment is the process of building a shared understanding of concepts. It is particularly relevant in group discussions, as it requires involved parties to reflect critically on concepts and negotiate meanings. While Generative Artificial Intelligence tools have been proposed to facilitate this process, it is still unclear how Large Language Models (LLMs) can support it without raising major concerns, especially in educational contexts. This paper reports an exploratory study on how group discussions around Artificial Intelligence (AI) can be informed by LLM-inspired activities and supported by an LLM-based interface. The study involving 24 middle-school pupils and was divided in 3 activities, each focusing on a different aspect of how LLMs work, with the last activity focused on conceptual alignment through an LLM-based interface.
Prelimiary results, based on qualitative and observational data, indicate that the LLM-based interface did support critical reflection on concepts within groups and that the activities allowed for a more critical understanding of how LLMs work.