Saudi Arabia SDAIA AI Ethics Principles (Version 2.0, 2025) 1.PD: Fairness in Plan and Design: fairness-aware design, harm and benefit analysis, chosen fairness metrics and defined sensitive attributes with thresholds

What does 1.PD require?

At the outset the design team pins down the system's objectives and plans a fairness-aware design with precautions across algorithm, processes and mechanisms so bias cannot produce discriminatory or skewed outcomes. Technical and non-technical members together identify the likely harms and benefits, who is affected including vulnerable groups, and whether the impact is justified by the system's purpose. Fairness assessment metrics are selected now, suited to the algorithm type, the kind of decision it drives (punitive or selective, for instance) and the harm and benefit of correct and incorrect predictions, and sensitive personal attributes of systematically or historically disadvantaged persons or groups are identified with the threshold that separates fair from unfair, to be measured in later steps.