Implementing AI ethics in software development

Abstract

Technical advances in Artificial Intelligence (AI) have made AI–powered systems a part of our everyday life. This is seen across applications from targeted advertising to autonomous vehicles. AI systems have grown from the confines of laboratories and are now applied in diverse societal contexts. Alongside the success stories of AI applications, false promises of the technology and growing numbers of incidents related to AI systems have highlighted the need to address ethical considerations. With current technology, the previously hypothetical threats and ethical issues related to development, application and use of AI have now become a reality. It is foreseeable that more issues will arise as the level of maturity increases. AI systems are still far from perfect. This realization has led to a call for action towards the advancement of AI ethics in the field, resulting in a high demand for principle-based approaches to AI ethics. Various stakeholders have addressed AI ethics via guidelines, laws, and regulations. AI principles in the form of guidelines have lacked actionability and developers have struggled with implementing abstract ethical guidelines into concrete actions. Transferring principles into practice is a major challenge for AI ethics. The question remains about how to influence developers to identify and consider ethical issues in stages of development. Although methods in the area exist, empirically tested AI ethics methods are still in need to bridge the gap of research and practice. The lack of actionability in proposed high-level solutions has prompted this research. This dissertation offers a means of addressing this research problem via an empirically grounded understanding on how to implement ethics in software development processes. The focus is on operational tools used in software development that transform philosophical thinking tools and principles to development practices. The research includes five qualitative research articles: two conference papers, one magazine article and two journal articles. The results of the dissertation further our understanding of how to implement ethics in software development. The method ECCOLA (Article V) helps to raise awareness of ethical issues and offers a process to implement ethical considerations into software development.