Institut für Operations und Information Systems
Asia Pittana
BSc MSc
DOKTORAT
Institut für Operations und Information Systems
Betreuer:
Univ.-Prof. Dr.
Stefan Thalmann
Assessing the Intersection of Bias, Interpretability and
AI Literacy in Automated Machine Learning
Automated Machine Learning (AutoML) is transforming access to artificial intelligence by enabling non-expert users to build and deploy predictive models through no-code platforms. While this democratization lowers technical barriers, it also heightens concerns about fairness in the generated models. AI literacy, the skillset and knowledge to critically engage with AI systems, has been applied in the traditional AI domain to address similar challenges and has shown promising results. However, little is known about how it shapes the responsible use of AutoML. This thesis investigates how advancing AI literacy influences users’ awareness of bias and their understanding of AutoML workflows. To this end, a vignette-based online experiment with 450 participants from the European labor market compared a short bias-awareness training with a control condition. Participants’ AI literacy levels are determined through pre- and post-tests, together with established instruments from literature to assess interpretability, bias awareness, and trust. The findings show that the short training did not improve AI literacy, bias awareness, or calibrated trust. However, participants with higher AI literacy consistently reported stronger interpretability, underscoring AI literacy as a key enabler of understanding, even in highly automated settings. By contrast, fairness perceptions and trust proved to be resistant to brief, passive interventions. These results highlight the limitations of one-off micro-trainings and suggest the need for more interactive, practice-oriented learning formats. They further point to the importance of embedding explanatory resources and fairness diagnostics directly into AutoML platforms. As one of the first empirical studies on human–AutoML interaction, this thesis clarifies how AI literacy can empower responsible engagement to ensure the ethical and trustworthy use of AutoML.