• integrated_82#Q2.5

Uncertainty matters in RAIDT because the framework is designed for organisational uses in which decisions are made with incomplete, ambiguous, conflicting, or fast-changing information. In the managerial paper, this is the condition under which conventional AI support becomes risky: outputs may look fluent and authoritative even when data are weak, misleading, or unstable. For RAIDT, uncertainty is therefore not a peripheral technical issue; it is a governance trigger. It explains why interpretability alone is insufficient and why uncertainty communication, limitation disclosure, and human override are required if managers are to calibrate trust rather than over-rely on opaque recommendations.

RAIDT addresses this by treating the run as the unit of governance and by requiring a run-level evidence pack for each material use. Uncertainty matters across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability): Responsibility depends on whether limitations and oversight are made explicit; Auditability depends on whether a contested run can be reconstructed; Interpretability depends on explanation plus uncertainty communication; Dependability depends on repeat-run stability under variable conditions; and Traceability depends on preserved provenance for prompts, retrieved context, tools, and review actions. The score profile, using anchors 1=missing / 3=partial / 5=audit-ready, makes uncertainty governable rather than implicit. It also clarifies why influence methods as governance interventions must be logged, because structured prompting, retrieval, alignment, or adapters can either improve or degrade governance readiness under uncertain conditions.

In a healthcare note-summarisation workflow, a clinician asks a GenAI assistant to summarise a chest-pain consultation before referral. The case is uncertain because symptoms are incomplete, the patient history may be fragmented, and the summary could influence further action. Under RAIDT, the system should not simply produce a neat narrative. The run-level evidence pack should record the prompt template, model or deployment identifiers, any retrieved protocol text, the output, an explicit limitation or uncertainty statement, and the clinician?s oversight decision.

If the summary omits red flags or sounds overconfident despite weak evidence, the run will score poorly on Interpretability, Responsibility, and possibly Dependability. If structured prompting requires a fixed schema and uncertainty disclosure, and the clinician remains the decision maker, the score profile becomes stronger and closer to audit-ready. The value of RAIDT here is that uncertainty is surfaced, evidenced, and reviewable after the event.

• 01-Responsible_AI_for_Managerial_Decision-Making_Under_Uncertainty-V3
• 11-RAIDT_Academic_Logic_M_v11
• 15-RAIDT-IS-Governance_M_v07