• workshop_table17#tag-band S1–S2 · 20–40 min

The project emerged from a practical and theoretical mismatch identified in the original Responsible AI route. The managerial decision-making paper frames organisations as operating in volatile, high-stakes settings marked by uncertainty, incomplete data, information asymmetry, misleading inputs and rapidly changing conditions. In that context, AI was attractive because it could process complexity, but it was also problematic because opaque outputs weakened trust, ethical accountability and governance compliance. Responsible AI supplied the normative starting point, yet the paper also shows that most existing work stayed at the level of principles, taxonomies or compliance language, with too little operational guidance for real decision support under uncertainty. The project therefore began from the need to make AI recommendations explainable, uncertainty-aware and usable for managers who still had to justify consequential decisions.

The later RAIDT logic and governance papers sharpen that problem into an Information Systems governance question. They argue that policy statements, model-level documentation and periodic audits are too coarse once outputs are materially shaped at run time by prompts, retrieved context, tools, adapters and review actions. That governance pressure explains why RAIDT formalises run as the unit of governance, requires a run-level evidence pack, and evaluates readiness through the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). In this framing, a score profile with anchors 1=missing / 3=partial / 5=audit-ready gives organisations a way to inspect whether one configured use can be reconstructed, challenged and improved. It also treats influence methods as governance interventions, not merely technical tweaks. Accordingly, the project emerged from Responsible AI because uncertainty made trustworthy decision support necessary, and it evolved into RAIDT because accountability demands required evidence-based governance at the level of each consequential run.

A suitable illustration is healthcare note summarisation, which the RAIDT logic paper uses as a worked scenario. Suppose a hospital deploys GenAI to draft a discharge summary from a partial clinical note. The note omits some medication details, the patient history is incomplete, and the system retrieves a guideline excerpt that may not fully fit the present case. A fluent answer may still look authoritative, even though the uncertainty is material and the clinician remains accountable for the final record.

In a principles-only Responsible AI approach, staff may know that transparency matters, but they may still be unable to show which prompt version, retrieval snapshot, model configuration or review step shaped the output. RAIDT addresses precisely that governance pressure by treating run as the unit of governance and preserving a run-level evidence pack for the event. The resulting score profile can reveal whether the case is merely partial or genuinely audit-ready across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability), allowing oversight to focus on the exact run rather than on abstract policy claims.

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