Vortrag auf der 3rd International Conference on Frontiers of Artificial Intelligence, Ethics, and Multidisciplinary Applications in Norwegen (FAIEMA 2025)    [13.08.25]

Koautoren des Papers: Thorsten Zylowski, Tarek Bahiani & Matthias Wölfel

Abstract:

This paper presents a user study conducted to examine the effects of explainable artificial intelligence (XAI) interfaces on trust in artificial intelligence (AI) systems. The study is based on an online pub quiz scenario where participants collaborated with an AI system to find correct answers. The evaluation focused on variables derived from self-determination theory (SDT) to examine the extent to which these variables are useful in predicting trust using technical-focused XAI methods such as local interpretable model-agnostic explanations (LIME). Through a controlled experiment with a sample of 94 participants, we gathered data on the perceptions of trust, a trust proxy to determine the trusting behavior and variables from SDT, namely autonomy, competence, and relatedness. Additionally, it was elaborated how trusting the AI was affected by the participants’ uncertainty with the answer. We found that perceived trust is not influenced by the presence of a technical XAI component. However, the actual trust behavior was significantly influenced. The technical XAI component was unable to exert a significant impact on the perceived competence, autonomy, and relatedness, although visual scatterplot analysis revealed higher values when the XAI component was available. Specifically, increased competence and relatedness resulted in increased perceived trust.

Keywords: Explainable artificial intelligence · Trust · Self-determination theory · User study

Das Paper (peer review) wird im Anschluss an die Tagung in der Springer Buchreihe "Frontiers of Artificial Intelligence, Ethics and Multidisciplinary Applications" veröffentlicht werden.

Die FAIEMA 2025 findet in diesem Jahr vom 18.-19. September 2025 in Starvanger, Norwegen, statt.

Zurück zu Aktuelles