Current state of practice on AI ethics implementation in the industry

Abstract

This thesis consists of two articles discussing the current state of practice on artificial intelligence (AI) ethics implementation in the industry. As the usage of AI in systems affecting peoples’ everyday lives constantly increases, public demand for ethical consideration in AI development and deployment grows as well. AI ethics has been discussed in the academia for decades, but a consensus has yet to be found on what it means and contains. In this study, a framework for AI ethics research is constructed based on ethical principles frequently discussed in academic literature. Based on the framework, first, a multiple case study via semi-structured interviews in software companies developing AI systems is conducted, to gain a preliminary understanding on the AI ethics implementation in the industry. Second, a large-scale industrial survey is executed in order to achieve an in-depth understanding on the current state of practice on AI ethics implementation in the industry, and through larger sampling be able to generalize the results. The study found a clear gap between research and practice regarding the implementation of AI ethics. In particular, developers are not aware of the academic discourse on AI ethics and they consider AI ethics important only in principle. As a result, AI ethics principles are not implemented in practice. It was also discovered that the industry professionals do not utilize any tools or methods to implement AI ethics, and the industry does not have any common practices for the implementation of AI ethics either. In this study, means are suggested for narrowing the gap between research and practice for better implementation of ethics in AI system development.