Awareness Without Agency
Designing and Testing Friction in Student-AI Interaction
R.G. Mihălăchiuţă (TU Delft - Electrical Engineering, Mathematics and Computer Science)
C.C.S. Liem – Mentor (TU Delft - Electrical Engineering, Mathematics and Computer Science)
L. Cavalcante Siebert – Graduation committee member (TU Delft - Electrical Engineering, Mathematics and Computer Science)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Students increasingly delegate reasoning, argumentation and writing to large language models, and these systems are built to make that delegation effortless. The same tools scaffold learning and erode the capacities education exists to develop, which is the structure Plato named pharmakon: remedy and poison in one substance, the difference lying in the conditions of use. This thesis asks whether deliberately introducing friction into LLM interaction changes how students work, and under what conditions they accept or abandon it.
Two studies address the question. A survey of ninety university students establishes that reflective engagement collapses situationally. Awareness is not the missing ingredient, since the students who most clearly recognise the model doing their intellectual work are those least able to stop. These findings, with the conditions for reflection drawn from learning theory, yield seven design requirements and a design space of seven friction-based interaction modes.
Three were combined into a single friction condition and tested against a conventional assistant, in a within-subjects experiment (N = 14). Friction did not slow students down or reduce what they produced. What changed was its destination, with the share directed to the assistant rather than into the document rising from 39% to 55%. The writing itself showed no structural difference. Participants divided in how they experienced the resistance, and what unified them was a request for control over when it applied.
Friction relocates student effort reliably. What it is worth remains open, and the ambivalence is not a gap in the evidence but the finding itself: the same design, in the same conditions, helped some students and obstructed others. What friction asks of a student is that they stay a little longer with not yet knowing.