Explain AI-Based Essay Scorings without XAI - Empirical Investigation of an User-Centered UI Design for AI-Based AES Systems.

The lack of understandability of AI-based decisions is increasingly posing trust-related and regulatory problems. This also applies to the educational sector, where AI is a central element of modern automated essay scoring (AES) systems. However, current research on explainable AI primarily focuses on complex technical approaches. These explanations usually show a lack of understandability by the actual users, who often have no knowledge of AI. Based on an experiment with 245 students at a German university, we were able to show that even the basic principles of user interface design can improve understandability and hereby trustworthiness. Thus, the use of visual elements promotes understandability even when only little information is provided. Especially when providing further AI-specific information on the scoring of AES systems, however, it must be considered that in combination with visual elements an information congruency can be observed, leading to a cognitive overload in the worst case. [ABSTRACT FROM AUTHOR]