Although people who explain technical objects take their listeners’ interests into account, they still primarily focus on how these objects work. This is shown by a recent study from the A04 project of Transregio 318 “Constructing Explainability.” The findings were published under the title “Navigating the Dual Nature: Do Explainers Adapt to Explainee Interests When Explaining Technical Artifacts?” in the International Journal of Technology and Design Education. The article is freely accessible.
The starting point for the study was a theory originating in the philosophy of technology, according to which technical artifacts—that is, all kinds of tools developed by humans, including algorithms—can be viewed from two perspectives: first, in terms of their architecture, i.e., how they are structured and function; and second, in terms of their relevance, i.e., what purpose they serve and why they are important. This dual perspective is referred to as “dual nature.” Researchers from the fields of computer science education, educational psychology, and linguistics investigated whether both perspectives are reflected in explanations. The focus was on how flexibly explainers respond to the interests of their listeners.
People adapt their explanations—but not completely
For the study, the researchers conducted a controlled experiment with a total of 72 participants. They used the board game “Quarto!” as an example. The game served to highlight typical mechanisms of explanation. Participants were asked to explain the game in such a way that their conversation partner could subsequently win. In doing so, conversation partners who had been briefed on the study deliberately showed interest either in the game’s structure or in its relevance. To do this, they pretended not to be familiar with the game.
The results show that people do indeed respond to their conversation partner’s interests. When listeners showed particular interest in the meaning or utility of an artifact, the explainers more frequently addressed these relevance-related aspects. In these cases, the proportion of relevance-related statements rose from 28 to 40 percent. Nevertheless, statements about architecture—that is, how something works and is structured—dominated overall. Depending on the experimental conditions, architecture-related content continued to account for around 60 to 72 percent of the explanations. Apparently, many explainers view this knowledge as the foundation for being able to understand the board game Quarto! at all. Relevance-related statements can further enhance understanding.
Implications for Education and Explainable AI
The study also makes it clear that explaining is not a one-sided process. Instead, understanding emerges through mutual interaction. “We observed that explainers paid close attention to their interlocutors’ reactions and adapted their explanations accordingly, provided that this was consistent with their own explanatory plan,” says lead author Lutz Terfloth. This so-called “monitoring” enabled co-constructive communication, meaning that both conversation partners actively shape the explanatory process.
These findings are relevant both for educational research and for the development of explainable AI systems. They highlight that good explanations should include information not only about how technical systems work but also about their significance and application.
“Our findings can help design teaching and learning processes in a more targeted manner and further develop adaptive explanation systems,” says Terfloth. “Especially for explainable AI, it is crucial that explanations are not only correct but also tailored to the interests and needs of users. It is not enough to promote a superficial understanding. Users should become capable of taking action when interacting with technology.”
To the publication (open access):
Terfloth, L., Buhl, H.M., Lohmer, V. et al. Navigating the dual nature: do explainers adapt to explainee interests when explaining technical artifacts. Int J Technol Des Educ (2026). doi.org/10.1007/s10798-026-10084-9: https://link.springer.com/article/10.1007/s10798-026-10084-9
To the project:
