Résumé
Artificial intelligence (AI), powered by data and trained with machine learning (ML) technologies, is rapidly playing a key role in advancing society in vital areas. However, there are often irrational, biased, and discriminatory results associated with the deployment of ML systems. Consequently, value-based design methodologies have emerged in recent years to foresee and lessen immoral misbehavior by highlighting ethical and epistemological difficulties in the creation of AI systems.
In this work, a participative data-centric strategy to AI morality by design is described. This strategy was developed by identifying and refining concepts that originated from efforts inside the group of value-sensitive design techniques.
This method offers a realistic perspective for tackling epistemological and ethical concerns with data activities from the outset of an ML development project. Therefore, this chapter aims to improve prospects for morally oriented AI architecture by emphasizing the need to bridge the gap between system programmers and domain specialists in order to foster a common understanding of a specified information domain and its relation to a certain practice.