The Influence of AI Literacy on User Preferences for Explainable AI in Recommender Systems

Sammanfattning

The increasing reliance on AI-driven recommender systems, presents a challenge due to the lack of transparency. This challenge raises societal concerns such as misinformation and lack of critical thinking in decision-making, which can lead to users’ over relying on AI. Explainable AI (XAI) seeks to address this issue by offering clear explanations, thereby improving understanding and trust. However, there is a limited amount of research on how users with different levels of AI knowledge interact and prefer these explanations. This study investigated how users’ AI literacy levels influence their preferences for different explanation formats and complexities in recommender systems. Using a quantitative method with a questionnaire of 104 participants, the study assessed AI literacy using the Artificial Intelligence Literacy Scale (AILS) and gathered preferences for various explanation styles. The Chi-Square Test of Independence revealed no statistically significant association between AI literacy and the participants’ most preferred explanation style. A Linear Mixed Model (LMM) analysis showed that while AI literacy did not significantly influence format explanation preferences, it had a significant interaction with explanation complexity. Specifically, low AI literacy participants significantly preferred simple explanations over detailed ones. The other interactions AI literacy, format, and complexity did not result in significant results. These findings highlight the need for personalized XAI explanations tailored to users’ AI literacy levels, particularly concerning the complexity. This research contributes to the XAI research field by highlighting the end user’s perspective as well as contributing to the development of a more accessible recommender systems that encourage decision-making with AI.