Explained or Certified? Examining the Influence of XAI and AI-Seals on Users’ Trust and Understanding

Abstract

Explainable artificial intelligence (XAI) aims to increase users’ understanding and trust in AI-powered systems. However, while XAI has shown benefits, it also comes with shortcomings such as shifting the cognitive burden to users. Certifications, so-called AI-Seals, pose an alternative (or addition) to XAI by signaling a system's trustworthiness through a label. Despite their potential, empirical investigations of AI-Seals remain scarce. In an online experiment (N = 436), a 2 (XAI vs. no XAI) x 2 (AI-Seal vs. no AI-Seal) between-subject design was conducted to analyze the effects of XAI, AI-Seals, or a combination of both on users’ perceived and factual understanding and trust in the AI-application. Results show that XAI led to more perceived and factual understanding but had no effect on trust. Against our assumptions, an AI-Seal could neither promote perceived understanding nor trust. Likewise, the combination did not increase trust but has potential to increase factual understanding.

Ähnliche Arbeiten