Master thesis

From Error to Inspiration: Quantifying the Impact of AI Hallucinations on Creative Work

Generative AI systems increasingly mediate professional creative work in domains such as advertising, product design, and digital services. While these systems can support idea generation and productivity, prior research highlights risks such as over‑reliance on AI, difficulties in evaluating AI‑generated content, and potential creative stagnation. At the same time, recent work suggests that AI “hallucinations”, plausible but incorrect or surprising outputs , can stimulate divergent thinking by introducing unconventional ideas that humans might not generate on their own.

Recent research like Holmström et al. (2026) suggests that generative-AI hallucinations have a dual role. They enable rapid idea generation, diversity of concepts, and unconventional ideas (generative potential), but also foster dependency on AI, evaluation difficulty, and creative stagnation (reflection). Complementary qualitative work conceptualizes GenAI as a “spirited technology”, a responsive yet unpredictable system that produces outputs beyond the users’ intentions and reshapes how creatives engage with it and react to its suggestions (Retkowsky et al., 2026). However, there is a lack of quantitative evidence on when hallucination‑prone, polished AI outputs actually help or hinder divergent thinking and creative performance in realistic creative tasks.

This master’s thesis will use a quantitative, experimental design to investigate how different GenAI configurations (e.g., sketch‑like vs. polished outputs; wild/hallucination‑tolerant vs. accuracy‑focused prompting) influence divergent thinking and creative outcomes in professional‑style tasks such as writing tasks, product concepts brainstorming, or interface mockups. Participants will work on these tasks with or without AI support, and the study will measure dependent variables such as idea fluency, flexibility, originality, elaboration, perceived early closure, dependency on AI, and satisfaction with the creative process.

Exemplary Research Questions

How do different GenAI configurations (hallucination‑tolerant vs accuracy‑focused; sketch‑like vs polished outputs) affect divergent thinking (e.g., fluency, flexibility, originality, elaboration) in professional‑style creative tasks?
How do individual differences such as AI literacy, tolerance for ambiguity, and baseline creative self‑efficacy moderate the effects of AI hallucinations on creative performance and dependency?

Research Methods

Quantitative experimental design (e.g., laboratory experiment or online experiment with realistic creative tasks)

Requirements

This thesis proposal is suitable for students in business and economics, business informatics, industrial engineering or related fields, who meet the following requirements:

Strong interest in current research on generative AI, creativity, and human-AI interaction.
Experience with quantitative empirical research methods (e.g., experiments, surveys) and solid skills in statistical analysis (e.g., regression, mediation/moderation, structural equation modeling).
High degree of independence and personal responsibility.
Ideally, familiarity with GenAI tools (e.g., Google AI Studio) and basic scripting for API integration or logging (e.g., Python, R, javascript) for the implementation of the experimental conditions.

Literature

Holmström, J., Carroll, N., & Sundberg, L. (2026). What is the Creative Value of AI Hallucinations? Insights from Generative AI use in Educational Contexts. Information Systems Frontiers, 1-18.
Retkowsky, J., Hafermalz, E., & Huysman, M. (2026). Harnessing a" Spirited Technology": How Working with Generative AI Collapses the Creative Process. Academy of Management Discoveries, (ja).
Cai, W., & Gao, M. (2025). Beyond hallucination: generative AI as a catalyst for human creativity and cognitive evolution. ICCK Transactions on Emerging Topics in Artificial Intelligence, 2(1), 36-42.