Evaluating Explainable AI (XAI) in Terms of User Gender and Educational Background

Abstract

Artificial intelligence (AI) and machine learning (ML) have become ubiquitous tools in the modern era. As more technologies become dependent on AI and ML, there is a greater push to understand how to use Explainable AI (XAI) to help human users understand the underlying decision processes of the technology. We report on a laboratory-controlled experiment focused on trust and understanding comparing different types of XAI explanations with a recommendation system while controlling for participant gender and educational background. We found statistically significant interactions of both gender and educational background. As a result, we conclude that there is not one particular way to show explanations that is best for all audiences because both gender and the person’s education and professional background play a part in determining their trust and understanding in the system. In addition, although a number of publications have promoted visual word clouds as a way to show XAI, our participants rejected that type of explanation and preferred textual explanations.