A participatory data-centric approach to AI Ethics by Design

Abstract

ABSTRACT Data-driven artificial intelligence (AI) based on machine learning techniques (ML) has increasingly become an enabler in critical societal domains. However, the introduction of ML systems is often accompanied by unjustified, biased, and discriminated outcomes with severe consequences for the individuals affected. Consequently, in recent years value-based design methods have sought to anticipate and mitigate moral wrongdoing by drawing attention to ethical and epistemic challenges related to the design of AI systems. This article presents a participatory data-centric approach to AI Ethics by Design by promoting and refining insights from contributions within the family of value-sensitive design methods. The approach provides a practicable outlook on addressing epistemic and ethical issues related to data activities in early ML development project stages. Hence, the article seeks to enhance opportunities for ethically informed AI design by stressing the need for bridge building to cultivate a shared understanding among system developers and domain experts about a given data domain and its relatedness to a specific practice.