Department of Mathematics and Computer Science

The rapid integration of AI tools into everyday life has introduced a new class of cognitive challenges that the general public is largely unaware of. The Cognitive Load, Fatigue & Decision Offloading 2025 dataset, published by the Human Clarity Institute and collected from 503 participants across five English-speaking countries, documents how people experience mental strain when crafting AI prompts, how AI blurs the boundary between personal judgment and automated responses, and what cognitive after-effects, such as mental replay and difficulty switching off, emerge from sustained AI interaction (Human Clarity Institute, 2025). Despite the significance of these findings, the underlying concepts of cognitive load, decision fatigue, and cognitive offloading remain abstract and inaccessible to non-specialist audiences. Serious games, designed primarily for educational or awareness purposes, offer a compelling vehicle for bridging this gap, transforming abstract psychological constructs into embodied, felt experiences that players can reflect upon (Michael & Chen, 2006; Plass et al., 2015).

Most public discourse around AI focuses on capability or systemic risk, leaving the individual's internal cognitive experience largely invisible. People may regularly offload decision-making to AI tools, feel mentally drained after prolonged interaction, or lose confidence in their own judgment, without any conceptual vocabulary to describe these experiences. This thesis asks whether a well-designed serious game can make these invisible cognitive phenomena tangible and personally meaningful — and what design principles make such a game effective.

Research Questions

Three interlocking questions structure the thesis. First, how can empirically measured constructs from the HCI 2025 dataset, cognitive load, decision fatigue, AI judgment blurring, and post-interaction after-effects, be translated into meaningful game mechanics? Second, does playing the game demonstrably increase players' conceptual understanding of and reflection on AI-related cognitive offloading? Third, what design tensions arise when balancing scientific fidelity against playability, and how should these be resolved?

Game Concept

The game, provisionally titled Mind the Load, places the player in escalating daily-life scenarios — writing a work email, choosing between job offers, responding to a personal dilemma — each of which can be completed independently or delegated to an embedded AI assistant. As players offload more decisions, a "cognitive clarity" meter recovers short-term, simulating the immediate relief of delegation. However, a secondary "agency confidence" meter degrades over time, reflecting the judgment blurring captured in the HCI dataset's ai_judgment_blurring variable and grounded in Sweller's cognitive load theory (Sweller, 1988) and Baumeister's ego depletion model (Baumeister et al., 1998). A "mental residue" mechanic operationalises post-interaction fatigue, where AI-assisted tasks leave lingering cognitive noise that interferes with subsequent decisions — directly reflecting Risko and Gilbert's (2016) framework of cognitive offloading costs. The game concludes with a personalized debrief mapping the player's in-game behavior onto real-world HCI findings.

Methodology

The thesis follows a research-through-design methodology across three phases. The first phase involves analysis of the HCI dataset alongside foundational cognitive science literature — including Sweller's cognitive load theory, Baumeister's ego depletion model, and Clark and Chalmers' (1998) extended mind framework — and a survey of existing serious games addressing cognitive bias and digital wellbeing. The second phase produces an iterative browser-based prototype, documented through design diaries and two rounds of structured playtesting with working adults. The third phase uses a mixed-methods evaluation combining pre- and post-play questionnaires, think-aloud protocols, and thematic analysis of open-text responses, mirroring the methodological approach of the source dataset itself.

Procedure

The thesis begins with a review of existing evaluation metrics for conversational tone, empathy, and naturalness in AI-generated dialogue. A set of evaluation criteria specific to the emergency room waiting context is then defined in collaboration with relevant stakeholders, for instance, distinguishing appropriate reassurance from inappropriate symptom minimization, and used to construct an LLM evaluator prompt following rubric-based design principles. A sample of chatbot conversations is rated by both the LLM evaluator and a group of human raters, including laypeople and domain experts. Quantitative comparison of scores and qualitative thematic analysis of rater reasoning are used to identify where the LLM evaluator diverges from human judgment. Findings inform iterative improvements to the evaluator design and surface broader implications of automating emotional language assessment in high-stakes settings.

References

• Baumeister, R. F., Bratslavsky, E., Muraven, M., & Tice, D. M. (1998). Ego depletion: Is the active self a limited resource? Journal of Personality and Social Psychology, 74(5), 1252–1265. https://doi.org/10.1037/0022-3514.74.5.1252
• Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1), 7–19. https://doi.org/10.1093/analys/58.1.7
• Human Clarity Institute. (2025). Cognitive Load, Fatigue & Decision Offloading 2025 (Dataset, v3.0). https://github.com/humanclarityinstitute/HCI-CognitiveLoadDecision-2025 
• Michael, D., & Chen, S. (2006). Serious games: Games that educate, train, and inform. Thomson Course Technology.
• Plass, J. L., Homer, B. D., & Kinzer, C. K. (2015). Foundations of game-based learning. Educational Psychologist, 50(4), 258–283. https://doi.org/10.1080/00461520.2015.1122533
• Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002
• Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257–285. https://doi.org/10.1207/s15516709cog1202_4