During study, students often offload cognitive operations, from the body to the environment (Clark & Chalmers, 1998), for example through tools such as pen and paper (Storm & Stone, 2015). This process allows them to reduce cognitive load when dealing with tasks that demand multiple cognitive operations (Burnett et al., 2025). For example, students may offload cognitive operations, such as coding and decoding in the case of notes (e.g., encoding ideas into bullet points or symbols), or comparison (e.g., using a spreadsheet to automatically compute differences and rankings), which can free resources for other operations and facilitate understanding (Kirsch, 2010). However, students' widespread use of generative AI (genAI) such as ChatGPT in their studies may have caused a major shift in the patterns of cognitive offloading, especially high-level cognitive operations, which have implications for developing crucial cognitive skills. Additionally, the use of genAI spans a spectrum of epistemic agency from integration (Gonsalves, 2024), to scaffolding (Darvishi et al., 2024) to complete outsourcing (Chen, 2025), with students exercising progressively less control over and responsibility for their knowledge work. Students offloading behaviors when studying with AI agents is yet understudied, leaving and important gap on the account of (a) the offloading of cognitive operations from students to genAI during study activity, and (b) the agency profile of student-genAI interaction as students engage in knowledge work. In our study, we aimed to clarify when genAI use functions as an extension that reorganizes cognition without displacing learners’ central epistemic role, and when it becomes a substitute that relocates substantial knowledge work outside the learner.

This study is grounded in two interconnected theoretical perspectives that together enable focused analysis of student-genAI interactions: Distributed Cognition and Epistemic Agency. Distributed Cognition (Hutchins, 1996) provides the foundational lens for understanding how cognitive activity emerges from coordinated interactions among components of a sociotechnical system, comprised of individual students, genAI tools, and their learning communities. This framework allows us to conceptualize genAI not merely as an external aid but as an integral component of a distributed cognitive system where knowledge work unfolds across human and technological agents. The concept of epistemic agency (Scardamalia, 2002; operationalized by Damşa et al., 2008) enables examination of the degree to which learners maintain responsibility for and control over their knowledge work, including setting epistemic goals, evaluating outputs, revising ideas, and regulating their learning processes. Together, these frameworks provide an adequate analytical structure for investigating both what cognitive operations students offload to genAI and how these interactions can be characterized in terms of agency.

The study addresses two interrelated research questions: (RQ1) what cognitive operations are higher education students offloading to genAI? This question is addressed through analysis of the cognitive operations embedded in students' prompts to genAI systems; (RQ2) How can their interactions be characterized, especially in terms of agency? This question is examined through an analytical framework that categorizes student-genAI interactions along a spectrum of epistemic agency– integration (highest agency, where students actively incorporate genAI outputs into their own knowledge work), scaffolding (moderate agency, where genAI provides support structures that students build upon), and outsourcing (lowest agency, where students delegate cognitive operations with minimal personal engagement or evaluation). By examining these questions through the lens of distributed cognition and epistemic agency, this study aims to illuminate the implications of genAI use for students' development of cognitive skills and their ability to engage as epistemic agents during study.

Methodology Participants and Data Collection Data were collected from 73 higher education students at Tampere University, representing 15 different disciplines. Participants engaged in independent study sessions using genAI for 1-3 hours at a time. The data consisted of chat logs from these sessions, comprising primarily text-based exchanges but also including images when students incorporated visual materials into their prompts or received visual outputs from the genAI systems. Data Analysis The analysis employed a two-level coding scheme to address the research questions. The first level of coding focused on identifying the cognitive operations that students offloaded to genAI through their prompts (RQ1). The second level of coding examined the degree of epistemic agency students exercised in their interactions with genAI (RQ2). First Level: Cognitive Operations Coding Student prompts were coded according to the cognitive operations being offloaded to genAI. These operations were categorized into three hierarchical levels: Low-level cognitive operations: identification, differentiation, coding and decoding, comparison, and classification. Mid-level cognitive operations: mental representation, mental transformation, projection of virtual relations, and analysis and synthesis. High-level cognitive operations: logical reasoning, divergent thinking, syllogistic reasoning, transitive reasoning, hypothetical reasoning, analogical reasoning, and logical inference. Second Level: Epistemic Agency Coding Interactions were coded along a spectrum of epistemic agency, reflecting the degree of control and responsibility students maintained over their knowledge work. The three primary categories were: Integration (strong agency): Students actively incorporate genAI outputs into their own knowledge work through critical evaluation, synthesis with existing knowledge, iterative refinement, and substantive modification of generated content. Scaffolding (moderate agency): Students use genAI to support their knowledge work through requesting explanations, seeking structured guidance, obtaining feedback on their ideas, or accessing procedural assistance while maintaining decision-making control. Outsourcing (weak agency): Students delegate cognitive operations to genAI with minimal personal engagement, accepting outputs without evaluation, requesting complete solutions, or transferring epistemic responsibility to the system.

Expected Outcomes The analysis is currently underway and will be completed by the time of the ECR presentation. Preliminary findings from the first level of coding (cognitive operations) reveal several patterns in how students offload cognitive work to genAI. Initial analysis indicates that students frequently employ a sequential pattern of cognitive offloading, beginning with low-level operations such as identification and differentiation (e.g., through prompts like "define," "list," or "explain"), followed by requests for higher-level operations including analysis and synthesis, mental representation, and projection of virtual relations. This pattern appeared in approximately half of coded interactions in preliminary data, suggesting a systematic workflow where students delegate increasingly complex cognitive tasks to the AI system. Notably, the preliminary coding reveals substantial offloading of reasoning operations, including divergent thinking, logical reasoning, and hypothetical reasoning. Students appear to use genAI for generative exploration, argument construction, and scenario evaluation. Prompts beginning with open-ended "what" and "how" questions frequently offload divergent thinking, while requests to analyze scholarly arguments or evaluate models delegate logical reasoning to the system. The second level of analysis, examining epistemic agency, will determine whether these patterns of cognitive offloading reflect integration, scaffolding, or outsourcing modes of interaction. Full results will illuminate the extent to which students maintain control over their knowledge work and the implications for their development as epistemic agents in knowledge-work contexts involving genAI.

Burnett, L. K., & Richmond, L. L. (2025). Meta-analytic investigations of the effect of cognitive offloading on memory-based task performance and interindividual variability. Memory & Cognition. https://doi.org/10.3758/s13421-025-01743-8 Chen, B. (2025). Beyond Tools: Generative AI as Epistemic Infrastructure in Education (arXiv:2504.06928). arXiv. https://doi.org/10.48550/arXiv.2504.06928 Clark, A., & Chalmers, D. (1998). The Extended Mind. Analysis, 58(1), 7–19. https://doi.org/10.1093/analys/58.1.7 Damşa, C. I., Kirschner, P. A., Andriessen, J. E. B., Erkens, G., & Sins, P. H. M. (2010). Shared Epistemic Agency: An Empirical Study of an Emergent Construct. Journal of the Learning Sciences, 19(2), 143–186. https://doi.org/10.1080/10508401003708381 Darvishi, A., Khosravi, H., Sadiq, S., Gašević, D., & Siemens, G. (2024). Impact of AI assistance on student agency. Computers & Education, 210, 104967. https://doi.org/10.1016/j.compedu.2023.104967 Hutchins, E. (1996). The integrated mode management interface (Tech. Rep.). University of California at San Diego. Final report for project NCC 92-578, NASA Ames Research Center. Kirsh, D. (2010). Thinking with external representations. AI & SOCIETY, 25(4), 441–454. https://doi.org/10.1007/s00146-010-0272-8 Scardamalia, M. (2002). Collective cognitive responsibility for the advancement of knowledge. In B. Smith (Ed.), Liberal education in a knowledge society (pp. 67–98). Open Court. Storm, B. C., & Stone, S. M. (2015). Saving-Enhanced Memory: The Benefits of Saving on the Learning and Remembering of New Information. Psychological Science, 26(2), 182–188. https://doi.org/10.1177/0956797614559285