Collaborative Making of a Boundary Object for Classifying Generative AI Learning Activities

Synthesis: Dohn, Markauskaite, Huber, et al. (2026) develop and critically test a taxonomy for classifying GenAI learning activities, built collaboratively through a postdigital dialogue and framed as a boundary object (after Bowker & Star). The taxonomy specifies six core categories of human–GenAI entanglement — Learning Objective, Content, Representation Format, Epistemic Engagement, Social Design, and Artefacts — that together describe the key relationships between human and GenAI elements in a learning activity. By making visible choices about why, what, how, with what, and with whom students engage GenAI, the taxonomy helps practitioners and researchers systematise, compare, and imagine GenAI learning activities — connecting to AI Literacy, Prompt Engineering, Human AI Collaboration, and Learning Design.

Dohn, Markauskaite, Huber, Wardak, Yang, Zeivots, et al. (2026) address a practical and scholarly problem: as publications on GenAI for learning proliferate, an overview of how GenAI is used for learning becomes increasingly hard to obtain because of the lack of a common framework and terminology. Practitioners report frustration that they are expected to engage students in learning with GenAI without clarity about what learning goals activities should support or what role GenAI should play. The article proposes a taxonomy to systematise GenAI learning activities and support comparison across them.

A taxonomy as a boundary object

Building on Bowker and Star (1999), the authors treat classifications as boundary objects — objects that are "customisable" and "ambiguous" yet have enough "common identity" to travel across borders while maintaining a constant identity. A boundary object provides a common point of information and coordination from which each user can take and impose different meanings relevant to their expertise and needs. Because classifications are "suffused with ethical and political values," proposing a set of categories puts an ethical and political stake in the ground — making distinctions and creating focal points for current debates about GenAI's role in learning.

The taxonomy was developed iteratively through a postdigital dialogue, with the initial prototype tested and reconfigured against concrete examples in a multi-voiced discussion. This responds to three challenges: the field's rapid change (risk of obsolescence), the inherent entanglement and situatedness of learning activities (which no taxonomy can fully capture), and the influence of pedagogical histories and traditions on interpretation.

The six-category taxonomy

The taxonomy combines the German-Scandinavian Didaktik tradition (Heimann's didactic theory, focused on justification of content choice and character-formation) with the Anglo-American Activity-Centred Analysis and Design (ACAD) approach (focused on socially and materially entangled conditions of learning). Six categories describe GenAI learning activities:

Notably, the Epistemic Engagement category draws on Bozkurt's (2024) three GenAI literacies — "know what" (theoretical/conceptual), "know how" (practical/operational), and "know why" (critical assessment of ethical, epistemological, ontological aspects) — while adding a further distinction in "know how" between using existing GenAI and constructing GenAI. This matters because the entanglement forms, and resulting ethical and political issues, differ substantially when constructing (where fairer, more justifiable training data may be possible) versus using. The authors note that techniques such as fine-tuning and Retrieval-Augmented Generation (RAG) occupy a meaningful intermediate position, raising questions about control, responsibility, and whether incorporated biases can be mitigated.

Outcomes and critical reflection

The collaborative testing produced both elaborations (discrete clarifications and additions that made the taxonomy more precise without changing its logic) and restructurings (encompassing changes that challenged the scheme's underlying logic). The authors conclude that the taxonomy, as a generalised artefact, inevitably falls short of adequately portraying all aspects of specific entangled GenAI learning activities. However, it can be concretised for each particular situation, and the level of generality of the categories allows comparison across specific learning activities. The dialogue also shows how the taxonomy can serve as a prompt to stimulate educators' imagination about GenAI activities.

The taxonomy is explicitly not a design method or pedagogical framework — it aims to help teachers and researchers orient their search for good practices by specifying important traits and elements to look for, offering an accessible overview of possible GenAI roles in learning activities. It is positioned as one possible way to look at the landscape, acknowledging Global North bias in its theoretical foundations and inviting consideration of how it might differ from Global South pedagogical traditions.

Connections to the knowledge base

The taxonomy's Epistemic Engagement category (understanding / using / critiquing / constructing) maps directly onto AI Literacy dimensions and to Critical Thinking, and its person-plus-GenAI labour-distribution categories connect to Human AI Collaboration and debates about Cognitive Offloading. Its Content category situates Prompt Engineering as a distinct learning content area. The paper's framing of learning activities as entangled human–GenAI systems resonates with Activity Theory AIED, and its practical orientation connects to Learning Design and Curriculum Design.

Connected Concepts

• Generative AI
• AI Literacy
• Learning Design
• Curriculum Design
• Human AI Collaboration
• Activity Theory AIED
• Prompt Engineering
• Critical Thinking
• Cognitive Offloading
• Higher Ed

Connected Articles

• Student LLM Interaction Taxonomy Review 2026 — Taxonomy review of student-LLM interactions
• Dierickx Taxonomy LLM Tasks Critical AI Literacy Journalism 2026 — Task-based taxonomy of LLM tasks for critical AI literacy
• Socratic AI Physics Tutor Taxonomy 2026 — Taxonomy of Socratic AI physics tutor dialogue
• Sposato AI Educational Leadership Taxonomy 2025 — Taxonomy of AI educational leadership
• Metacognitive Learning Scenarios Taxonomy — Taxonomy of metacognitive learning scenarios

Citation

Dohn, N. B., Markauskaite, L., Huber, E., Wardak, D., Yang, H., Zeivots, S., Casey, A., van Diggele, C., Dohn, N. B., Mannix, K., Mantai, L., Pressick-Kilborn, K., Spence, N., Vallis, C., & Wilson, S. (2026). Collaborative Making of a Boundary Object for Classifying Generative AI Learning Activities. Postdigital Science and Education. https://doi.org/10.1007/s42438-026-00671-3