Despite the surge in research on the ethical risks of Artificial Intelligence (AI) there is still a clear need for methodologies and practical strategies to assess ethical risks of AI applications. As risk assessment becomes a cornerstone of regulatory initiatives (e.g. EU AI Act) the question remains to what extent these methods are capable of addressing more complex normative issues related to voluntariness, justice and power imbalances in the deployment of AI. The current article examines three common categories of ethical risk assessment: (1) bilateral assessments, such as impact assessment for AI, (2) multilateral assessments, where the perspective of a group of stakeholders is included in the assessment and (3) foresight assessments, where future impacts and risks are assessed. Subsequently, it will make a case for relational risk assessment methods as a supplementary approach to the ethical analysis of AI. It takes as a starting point the three-party model as developed by Hermansson & Hansson (Risk Management 9(3):129–144, 2007) to identify salient ethical issues as they arise in the relationship between three critical parties or roles present in all risk related decisions: the decision-maker, the risk-exposed and the beneficiary. This model is then adjusted to better fit with AI contexts and applied to AI in credit risk scoring to evaluate its merits. Overall, the article seeks to contribute to the development of a more fine-grained understanding of the ethical risks of AI, emphasising the importance of addressing and interrogating relational dimensions such as power imbalances as ethical risks of AI applications.
Read full abstract