Cognitive Offloading or Effective Practice? Exploring the Future of Learning with GenAI

A/Prof Tim Fawns, Monash Education Academy, Monash University

Many in Higher Education worry that cognitive offloading is eroding student learning, agency, and academic integrity. However, evidence from the AIinHE.org project reveals a more complex story. This talk explores how students navigate a landscape of evolving technologies, rules and trust relations, developing complex practices, and weighing practical needs against institutional requirements and personal beliefs and values. At a time of significant challenges for student engagement, the need for ongoing, nuanced dialogue between educators and students is increasing in urgency. Drawing next on the 4Ps assessment framework (product, process, performance, practice), I will talk about how reconfiguring assessment (and, therefore, teaching) to elicit a range of forms of evidence of learning over time is important not only to assurance but to helping students navigate learning in a time of widely available AI. Finally, I will ask how, through such change, we might reframe the challenge in front of us from a crisis of integrity and assurance to an opportunity to learn with students about what AI could mean for learning, student identity, agency, disciplinary knowledge, and society more broadly.

Dr Vi Truong, Dr Winn Chow, Andrew Chester, Dr Stella Peng, Dr Zoey Li, Dr Arzoo Atiq and Prof Eun-Jung Holden, University of Melbourne

Higher education has entered a post-adoption reality for Generative AI, shifting attention from whether students use these tools to how they manage the cognitive and identity tensions that follow. This presentation introduces the Diffusion, Competence, and Identity (DCI) Framework, developed from qualitative and cluster analysis of 188 reflective essays by master's students. The analysis reveals four distinct AI adopter typologies, each shaped by different fears: skill erosion, loss of authentic voice, and uncritical dependence. By understanding these internal struggles, the presentation offers targeted pedagogical strategies for educators seeking to support effective, critically reflective engagement with AI, rather than passive offloading.

Dr Winn Wing-Yiu Chow, Dr Stella Peng, Dr Arzoo Atiq, Vi Troung, and Muqing Guo, University of Melbourne

While students frequently claim Generative AI enhances their critical thinking, recent research warns of skill atrophy and metacognitive laziness. This study explores this paradox using a mixed-methods analysis of 188 postgraduate student workflows. We introduce a novel Four-Tiered Cognitive Offloading Framework (4COF) to reveal how students distribute cognitive load. Our findings show that while approximately half of the cohort believes AI acts as an intellectual partner, they actually fall into an efficiency trap, bypassing foundational learning and abandoning metacognitive self-regulation. We define this state as Epistemic Confinement, where students experience an illusion of competence, feeling as though they are thinking independently while operating entirely within AI-constructed analytical boundaries.

Dr Ha Nguyen, University of Melbourne

The ICAP (Interactive>Constructive>Active>Passive) framework for cognitive engagement (Chi & Wylie, 2014) has been influential in the education sector for providing clear behavioural differentiation between levels of learning. It also expounds the distinct knowledge-change processes involved in functioning at each level. The interactive mode is underpinned by co-inferring, or co-creating, knowledge with others through dialogue, while the constructive mode is achieved through inferring new knowledge by going beyond the learning material. These modes result in deeper learning than the passive and active mode, where the main knowledge-change processes are the storage and integration of given information, respectively (Chi & Wylie, 2014).

Jennifer Zhu, Saw Hoon Lim, Angelina Fong, University of Melbourne

This presentation examines how biomedical science students navigate the integration of GenAI into their learning. Drawing on a mixed-methods study across both Biochemistry and Physiology cohorts, it combines assignment-level GenAI declarations, survey, and focus groups to explore patterns of use and underlying decision-making. The findings reveal that GenAI functions as both a tool for offloading aspects of thinking, and a resource that can support learning through continuous cross-referencing. Despite differing behaviours, both groups share uncertainties around appropriate use. The study highlights that effective practice lies in student’s ability to make informed judgements about when and how to engage with GenAI.

Anne-Marie Chase & Kelly Galvin, Swinburne University of Technology

Generative AI has intensified long-standing challenges in assessment design, particularly risks of cognitive offloading where students outsource core thinking to technology. This presentation brings together a conceptual framework, early empirical insights, and an emerging design response to explore how assessment can preserve genuine learning in GenAI contexts. Drawing on our Thinking to Learn work, preliminary findings from professional learning workshops with academics, and a developing learning progression model, we argue for a shift from reactive regulation toward developmental, design-led approaches. The session will offer participants practical and conceptual tools to rethink where learning, thinking, and AI appropriately belong.

Stephen Campitelli, University of Melbourne

Drawing on nationally conducted research and extensive professional experience as an Academic Skills Adviser engaging students in 1-1 dialogic conversations, this presentation considers the opportunities and risks of AI-assisted study within a beneficial versus detrimental cognitive offloading framework. It positions human connection as pivotal highlighting the critical influence of educators on students’ choices re generative AI use in four key areas: understanding of assignment requirements; feedback provision; communication opportunities; and foregrounding good scholarly practice. It proposes a hybrid learning model of beneficial generative AI offloading for extraneous tasks alongside a recognitive model with human teachers/advisers as critical points of engagement.

Dr Stella Peng and Dr Winn Wing-Yiu Chow, University of Melbourne

As GenAI threatens the construct validity of traditional open assessments, distinguishing genuine student learning from cognitive offloading is a critical challenge. This presentation introduces a dual-focus assessment weighing a final analytical report equally with a rigorous audit trail of human-AI collaboration. Analysing 41 student group reports, we investigate how students demonstrate epistemic agency when mandated to justify their intellectual contributions against AI-assisted work. Findings reveal that learning value concentrates at points of epistemic friction, where students actively challenge and override AI logic. This study offers an empirically tested framework for authenticating learning and surfacing invisible cognitive processes in AI-saturated environments.

Dr Wasana Karunarathne and Ashley Hanson, University of Melbourne

Generative AI can enhance the quality of students’ work and support learning, but it also raises concerns about over-reliance and reduced engagement with feedback. This presentation examines how students engage with feedback in an iterative assessment where AI use was permitted with restrictions. Using a data analysis report completed across multiple submissions, we analyse patterns of improvement and distinguish between deeper learning, surface-level revision, and patterns consistent with possible cognitive offloading. A reflective survey on feedback engagement and group contribution provides additional context. The findings highlight the importance of evaluating the nature of improvement and the role of students’ trust in feedback in shaping how they act on it.

Scott Tetley, University of Melbourne

As GenAI tools have become embedded in student workflows, many students respond by offloading thinking to AI to reduce effort and ensure correctness. Rather than attempting to prohibit or detect this behaviour, this presentation reframes the challenge as one of assessment redesign. It presents a redesigned undergraduate marketing assessment structured so that offloading becomes inefficient, and thinking becomes the easiest path to completion. Drawing on evidence from two cohorts in a large, multi-instructor subject, it demonstrates how targeted design changes can shift student behaviour and re-centre learning in a GenAI environment.

Dr Peter Matheis and Ms Aneta Delevska, University of Melbourne

The rise of generative AI is transforming how students think, learn, and complete assessment tasks, challenging traditional assumptions about authorship and evidence of learning. This presentation introduces Cognitive Offloading–Informed Assessment Design (COIAD), a framework that reconceptualises assessment as a system of offloading conditions shaped by task design. Moving beyond binary notions of academic integrity, COIAD highlights how visibility and regulation of cognitive processes determine assessment validity. Through a practical 2×2 model, the presentation demonstrates how educators can redesign assessments to make offloading visible, structured, and pedagogically meaningful, supporting assured learning while aligning with contemporary, tool mediated cognitive practices.

A/Prof James Birt, Bond University and A/Prof Thomas Cochrane, University of Melbourne

Generative artificial intelligence is changing how educators design learning activities, but faster content generation does not guarantee better learning. This presentation introduces the Prompt to Practice Extended Reality Learning Design Framework, a pedagogy led model for using generative artificial intelligence and vibe coding to design immersive simulation scenarios. The framework begins with the learning problem, uses structured prompting to externalise scenario logic, translates that logic across extended reality platforms, and evaluates learner judgement through design-based research cycles. It argues that generative artificial intelligence should support, rather than replace, educator judgement by making learning design visible, testable and transferable.

Dr Haseen Akhtar, Indian Institute of Technology, Hyderabad

This paper examines how foundational creativity training in design education can be reinterpreted in the presence of Generative AI. Drawing from a third year Bachelor of Design course on Design Creativity and Innovation, the study reflects on assignments focused on observation, constraints, and reflective enquiry. Rather than positioning AI as a shortcut, the paper argues for preserving slow, human-centered processes such as seeing, interpreting, and reframing. Using a qualitative, reflective approach, it explores how these exercises resist cognitive offloading and cultivate deeper creative engagement. The paper proposes a pedagogical framing where GenAI augments but does not replace core creative capacities in design learning.