| Abstract: |
Fairness is a crucial concept in the context of artificial intelligence (AI)
ethics and policy. It is an incremental component in existing ethical
principle frameworks, especially for algorithm-enabled decision systems. Yet,
unwanted biases in algorithms persist due to the failure of practitioners to
consider the social context in which algorithms operate. Recent initiatives
have led to the development of ethical principles, guidelines and codes to
guide organisations through the development, implementation and use of fair
AI. However, practitioners still struggle with the various interpretations of
abstract fairness principles, making it necessary to ask context-specific
questions to create organisational awareness of fairness-related risks and how
AI affects them. This paper argues that there is a gap between the potential
and actual realised value of AI. We propose a framework that analyses the
challenges throughout a typical AI product life cycle while focusing on the
critical question of how rather broadly defined fairness principles may be
translated into day-to-day practical solutions at the organisational level. We
report on an exploratory case study of a social impact microfinance
organisation that is using AI-enabled credit scoring to support the screening
process of particularly financially marginalised entrepreneurs. This paper
highlights the importance of considering the strategic role of the
organisation when developing and evaluating fair algorithm-enabled decision
systems. The paper concludes that the framework, introduced in this paper,
provides a set of questions that can guide thinking processes inside
organisations when aiming to implement fair AI systems. |