User Interface Design Based on Human-Centered Explainable AI Methods

Abstract

Artificial intelligence (AI) has demonstrated its considerable influence on every aspect of human life. However, algorithms are getting rather complex, and there are more black-box models as tasks that AI deals with increase and become more complex. Therefore, the eXplainable AI (XAI) attempts to solve this problem by making the algorithm understandable and trustworthy for human beings. Although numerous explanation methods are provided, most answer the Why question (why does the algorithm generate such decisions) from the technical experts’view. In contrast, the leading target group of explanations for AI is non-technical end users. As the calling for human-centered XAI gets stronger, researchers have proposed a set of requirements and design guidelines for human-centered XAI. However, these requirements merely stay at an abstract level and have not gone into a detailed design context. Only limited research provides clear guidance on how to implement and fulfill those instructions, nor does much research present the actual practice in design work. Moreover, the evaluation research for the current human-centered XAI still needs to be enriched. This thesis work cooperates with the company EatMyRide (EMR) which assists cycling enthusiasts in customizing and evaluating their nutrition plans. This research aims to make the working principles and algorithm of the application more understandable and trustworthy so that users will stick to this application. The main contribution of this work is that it will compensate for the deficiency of authentic practice and evaluation of human-centered XAI in an actual design context. There will be research on the current EMR application and interviews for its potential users to acquire more profound insights, especially regarding the aim of this thesis. Based on previous findings, the practice of human-centered XAI will be presented as new user interfaces in low-fidelity and high-fidelity prototypes, and user testing will be conducted to evaluate the effect of the design work. The design practice and evaluation based on the findings from the literature review and previous research are the main contributions to the current human-centered XAI field because it implements the XAI guidelines and evaluates the real effects of the actual practice. After that, the paper delivers discussions and conclusions regarding the research questions, the limitation of this thesis work, and insights and suggestions for future explorations.