AI ethical biases: normative and information systems development conceptual framework

Alongside the revolutionary benefits of AI, it can cause numerous problems across the system development process. AI ecosytem players have recently started to interrogate the ethical biases implicit in AI-enabled applications and agents. The contestable nature of ethics and the complexity of AI-enabled applications has led to incoherent literature around AI ethical biases. The numerous conceptions of AI ethics and a multiplicity of ethical biases has compounded matters for researchers, practitioners, and policy makers. The current study proposes a conceptual framework to organize AI ethical biases. A narrative literature review was conducted to identify and group the biases into data biases, method biases and implementation biases. The CRISP-DM framework was used to classify the ethical biases. The emerging conceptual framework has four clusters that represents: System development phases, scope of ethical bias, exemplars, and possible solutions. The study extends the existing AI ethical frameworks and provides a unified communication artefact for practitioners.