The Gap Between Trustworthy AI Research and Trustworthy Software Research: A Tertiary Study

Abstract

With the increasing application and complexity of Artificial Intelligence (AI) systems, the trustworthiness of AI has garnered widespread attention across various fields. An AI system is a specific type of software system with unique trustworthiness requirements due to its distinctive characteristics in data and algorithms. Our objective is to investigate the state of the art in trustworthy AI and trustworthy software separately and to analyze the connections and gaps between them. To this end, we conducted a tertiary study, which is a systematic literature review of existing secondary studies. These secondary studies are divided into two groups: one focuses on trustworthy AI and the other on trustworthy software. We developed frameworks for both trustworthy AI and trustworthy software, summarized the definitions of quality attributes in a structured format, and analyzed the similarities of these attributes between the two areas. Additionally, we created a swimlane diagram illustrating trustworthy practices throughout the development life-cycle and in relation to specific quality attributes. Researchers in these two areas originate from distinct research communities, leading to a significant gap between the trustworthiness of AI and software. However, we believe that existing research on trustworthy software can effectively address some gaps in trustworthy AI research, and we have identified evidence of connections between the two areas.