AI Ethics Impact Assessment based on Requirement Engineering

Abstract

This paper proposes a methodology for evaluating the ethical impact of artificial intelligence (AI) systems on people and society based on AI ethics guidelines. The ethical impact of AI has been recognized as a social issue, and countries and organizations have formulated principles and guidelines on AI ethics, and laws and regulations will be enforced in Europe. Because these principles and guidelines are written in terms of philosophy and law, AI service providers, developers, and business users have the challenge of how they should practice the principles and guidelines to their AI systems. To address this challenge, we first analyzed cases of ethical problems caused by AI in the past and assumed that ethical problems could be linked to interactions between components of AI systems and stakeholders related to such systems. On the basis of this assumption, we then developed a methodology to comprehensively extract the ethical risks that an AI system poses. This methodology consists of two approaches. The first approach is to develop an AI ethics model that embodies ethics guidelines as necessary requirements for ethical AI systems and correlates these requirements with interactions. The second approach is an impact assessment process that uses the AI ethics models to extract ethical risks for individual AI systems. In this paper, we discuss the details of this methodology and show the results of an initial validation to verify the above assumption and the ease of the impact assessment process.