• qa_deck_100#slide 36 · Socio-technical governance and accountability lineage

RAIDT treats organisational GenAI use as a human-AI hybrid because organisational work is produced through joint configuration rather than machine autonomy. The papers draw on the human-AI hybrids literature to show that once AI systems draft, recommend, summarise, or explain, the governance problem sits in the hybrid arrangement itself. Humans define the task, select prompts, provide or approve context, decide whether to rely on the output, and bear organisational accountability for the consequences. AI contributes generation, speed, and structure; it does not remove managerial or professional judgement.

This matters methodologically and practically. If governance focused only on the model, it would miss the hand-offs through which responsibility is actually exercised. RAIDT therefore records both sides of the hybrid in a run-level evidence pack: prompt templates, retrieval snapshots, model and tool settings, human review notes, oversight decisions, and escalation actions. The score profile then assesses whether the hybrid run was governable across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). In effect, RAIDT treats collaboration itself as the object of evidence. That is why the framework is stronger than policy declaration or model documentation alone: it shows how people and AI acted together in one use, what evidence supported reliance, and whether the organisation can defend that reliance afterwards.

In a hospital discharge-summary workflow, a GenAI assistant drafts a summary from clinician notes and an internal guidance excerpt. The clinician reviews the draft, corrects omissions, and decides whether it enters the patient's record. This is a human-AI hybrid: the AI produces the first draft, but clinical responsibility, reliance, and record entry remain human acts. The governance issue therefore sits in the collaboration, not in the model alone.

A RAIDT run-level evidence pack would preserve the prompt schema, retrieved guideline snippet, output, uncertainty statement, clinician edits, and the final sign-off or escalation decision. The score profile then shows whether the hybrid run was reviewable and dependable enough for this high-stakes task. If the clinician's review decision or the retrieved source is absent, governance is weaker even if the text reads well.

• 17-RAIDT-Sociotechnical_M_v6
• 12-RAIDT_DSR_Theory_M_v8