Co-Constructive Interaction with Language Models

Modern large language models (LLMs) have made interaction with artificial intelligence accessible through everyday language. In educational contexts, they can generate explanations, reformulate content, provide examples, answer follow-up questions, and adapt contributions to an evolving dialogue. Yet fluent outputs are not necessarily accurate, relevant, or conducive to learning. Whether LLMs can contribute to jointly developing explanations therefore depends on their capabilities and on how users formulate goals, evaluate responses, and shape the interaction.

This project asks to what extent LLMs enable the joint development of explanations and how such interaction can be described theoretically. It examines which contributions support mutual reference, monitoring, and scaffolding, how users elicit and respond to them, and which possibilities and limitations arise from the complementary but asymmetric relationship between humans and language models. It also investigates the knowledge and skills required for purposeful, critical, and responsible use and the implications for AI literacy and computer science education.

The project combines three theoretical perspectives. The theory of hybrid interaction systems directs attention to the reciprocal influence of humans and digital artifacts within specific contexts of use (Schulte & Budde, 2018). Research on explanation as a social practice conceptualizes explanations as processes in which participants negotiate their object, form, and aim (Rohlfing et al., 2021). AI literacy provides a framework for identifying the cognitive, practical, affective, and ethical competencies required for reflective engagement with AI. Building on these perspectives, the project introduces co-construction as an approach to human–LLM explanatory interaction and develops the concept of co-constructive prompting. Prompting is understood not as formulating a single optimal input, but as an iterative process in which provisional goals and generated explanations are examined, refined, corrected, and extended.

The exploratory project comprises three studies intended to develop theory from complementary perspectives. The first examines participants’ accounts of their experiences with LLMs. The second observes two people using an LLM together and investigates how they organize the interaction, interpret responses, and determine subsequent actions. The third uses a workshop to discuss the joint construction of explanations with an LLM. Comparing interactions before and after the workshop allows changes following reflection to be explored. Across the studies, analyses address conversations as well as individual prompts and responses. They consider whether model outputs provide contributions on which users can build, particularly connections with earlier dialogue, opportunities to monitor understanding, and appropriate support. These functions do not imply human-like understanding, intention, or responsibility. Users remain responsible for assessing plausibility, consulting other sources, and deciding how generated content should be used.

The expected contribution is a theoretically grounded, analytically applicable model specifying under which conditions LLMs can participate in co-constructive explanatory interaction. Connecting model capabilities with communicative practices and educational aims clarifies the potential and boundaries of co-constructive prompting. The project seeks to inform learning opportunities beyond prompt templates. Learners should monitor emerging explanations, evaluate outputs using epistemic and ethical criteria, and deliberately shape human–AI interaction. Interaction with LLMs thus becomes both a means of learning and an object of critical reflection.

References

#

• Rohlfing, K. J., Cimiano, P., Scharlau, I., Matzner, T., Buhl, H. M., Buschmeier, H., Esposito, E., Grimminger, A., Hammer, B., Häb-Umbach, R., Horwath, I., Hüllermeier, E., Kern, F., Kopp, S., Thommes, K., Ngonga Ngomo, A.-C., Schulte, C., Wachsmuth, H., Wagner, P., & Wrede, B. (2021). Explanation as a Social Practice: Toward a Conceptual Framework for the Social Design of AI Systems. IEEE Transactions on Cognitive and Developmental Systems, 13(3), 717–728. https://doi.org/10.1109/TCDS.2020.3044366
• Schulte, C., & Budde, L. (2018). A Framework for Computing Education: Hybrid Interaction System: The need for a bigger picture in computing education. Proceedings of the 18th Koli Calling International Conference on Computing Education Research, 1–10. https://doi.org/10.1145/3279720.3279733