S7.09 - Socio-technical_systems

S7.09 ? Socio-technical systems

flowchart LR
    A[Background problem:
model-only governance
misses roles, routines, context] --> B[RAIDT
run-level evidence framework]
    H[Practical fields:
healthcare, finance, education,
public services, enterprise productivity] --> C[[Socio-technical systems
configured GenAI use across
people, tools, routines, context]]
    B --> C
    C --> D[Run-level evidence pack]
    C --> E[RAIDT five-pillar
score profile]
    C --> F[Reviewer reconstruction
contestability and audit readiness]
    D --> G[Governance readiness
and organisational learning]
    E --> G
    F --> G

? Star S7 - Academic Theory and Design Logic

Star context: Positions RAIDT as an Information Systems and design-science contribution in which GenAI governance is treated as an organisational arrangement of people, tools, routines, evidence, and authority rather than as a model-only technical problem.

Academic picture

Definition / background

A socio-technical system is a system in which technical components and social components jointly shape how work is performed and how outcomes are produced. The concept originates in organisational and Information Systems thinking that rejected the idea that technology can be understood apart from users, roles, routines, incentives, institutional constraints, and managerial control. In that tradition, performance and failure are produced by the interaction between artefacts and organisational arrangements, not by either side alone.

This matters directly for GenAI governance. A large language model may generate text, suggestions, summaries, or classifications, but the real organisational outcome depends on who framed the task, what prompt was used, what retrieval or tools were attached, who reviewed the result, what escalation path existed, what deadline pressure was present, and what policy or regulatory context shaped acceptable use. RAIDT therefore treats a run as a socio-technical event: a configured use of GenAI by identifiable actors within a specific task and governance setting.

The concept differs from narrower terms such as model performance, user experience, or workflow automation. Those terms illuminate parts of the picture, but socio-technical systems explain the whole configuration. In RAIDT, that wider configuration is essential because governance cannot stop at model documentation or technical logging. It must also account for human review, accountability allocation, exceptions handling, evidence completeness, and the organisational routines that determine whether outputs are trusted, challenged, or acted upon.

Within RAIDT, socio-technical systems belong to the theory and design logic layer because they justify why run-level evidence is necessary. If GenAI use is socio-technical, then evidence packs must cover both technical traces and organisational context. The five-pillar score profile likewise becomes meaningful only when interpreted as a judgement about a socio-technical arrangement rather than as a verdict on a model alone. Responsibility, Auditability, Interpretability, Dependability, and Traceability all depend on how people and systems are configured together.

Why this concept matters

Socio-technical systems thinking solves a basic governance problem: organisations often treat GenAI risk as if it sits only inside the model, while many real failures arise from poor integration into human work. A technically capable model can still produce weak governance outcomes if prompts are ad hoc, reviewers are under-trained, accountability is ambiguous, or exceptions are not logged.

The concept also avoids a common confusion between using AI and governing AI. Using AI concerns task completion. Governing AI concerns whether that use is reviewable, contestable, reconstructable, and aligned with organisational obligations. RAIDT needs the socio-technical concept because it moves discussion away from generic principles and toward the actual design of work practices, controls, and evidence around each run.

If this concept is missing, organisations can produce false assurance. They may believe they are safe because they have a policy, a model card, or a procurement checklist, while the day-to-day configured use remains opaque and weakly supervised. A socio-technical lens exposes where responsibility is diffuse, where review is nominal rather than real, and where evidence is too thin to support audit readiness.

Key idea: Socio-technical systems matter in RAIDT because GenAI governance depends on how models, people, roles, routines, and evidence are configured together at run level.

What this item explains

Why RAIDT governs configured use in context rather than assessing a model in the abstract.
Why run-level evidence must include both technical traces and organisational process evidence.
Why human oversight is not an optional add-on but part of the system being governed.
Why governance quality varies across tasks even when the same model is used.
Why the evidence pack and score profile must reflect the interaction of prompts, outputs, review roles, escalation paths, and local policy constraints.
Why organisational learning from GenAI depends on understanding recurrent socio-technical patterns, not isolated output errors alone.

Practical example / likely audience question

Audience question

Why describe RAIDT as socio-technical when the system is obviously built around a GenAI model and its logs?

Answer

The concern behind this question is usually that socio-technical language sounds broad, abstract, or unnecessarily sociological. The direct answer is that GenAI outputs do not enter organisations by themselves. They enter through tasks, interfaces, instructions, review routines, authority structures, and accountability arrangements. If governance only inspects the model and its logs, it misses the practical conditions that determine whether a generated output is accepted, challenged, corrected, or escalated.

A simple example is a drafting assistant used in a policy team. Two teams may use the same model with the same temperature settings, yet one team produces defensible outputs because prompts are templated, sensitive claims are checked by a named reviewer, edits are logged, and unresolved issues trigger escalation. The other team may operate informally, without consistent checking or documented responsibility. Technically the model is the same, but the socio-technical system is different, and so is the governance quality.

RAIDT handles this better than a generic AI governance approach because it asks for evidence at the level where those differences actually appear: the run. Instead of assuming that one policy or one model assessment covers all uses, RAIDT reconstructs how a specific configured use operated, what controls were present, and whether the arrangement was governable in practice.

Practical example in RAIDT terms

Consider a healthcare setting in which a hospital uses a GenAI assistant to draft discharge summaries from clinician notes. The run-level issue is not simply whether the model can generate coherent text. The key governance question is whether a particular discharge-summary run can be shown to have used the correct patient context, the correct prompt template, the correct reviewer role, and the correct sign-off path before information is entered into the clinical record.

The evidence needed would include the prompt or template version, the model and configuration used, the time and task context, the clinician or staff role initiating the run, the identity and timing of the reviewer, records of edits or overrides, and the final disposition of the output. Responsibility is affected because the hospital must show who remained accountable for the summary. Auditability and Traceability are affected because the run must be reconstructable. Interpretability matters because reviewers need to understand what the draft is for and how it should be checked. Dependability matters because unreliable or inconsistent drafting creates safety and quality risks.

In governance-readiness terms, socio-technical systems thinking improves the hospital's position because it turns the discharge-summary workflow into an evidence-bearing arrangement. RAIDT can then distinguish between a technically available GenAI tool and a clinically governable socio-technical practice.

Detailed link to RAIDT

Socio-technical systems link to RAIDT in four ways.

First, they support RAIDT's core idea that responsible GenAI governance must focus on configured use in organisational context rather than on abstract model claims.
Second, they explain why the run is the correct unit of governance: each run is a situated interaction among tool settings, task conditions, human actors, review routines, and institutional constraints.
Third, they shape the contents of the evidence pack and the interpretation of the score profile, because the quality of governance depends on technical and social evidence taken together.
Fourth, they strengthen reviewability, contestability, audit readiness, and organisational learning by making it possible to reconstruct not only what the model produced, but also how the surrounding work system handled that output.

Socio-technical systems ? Run-level evidence ? Evidence pack ? RAIDT score profile ? Governance readiness

In this chain, socio-technical systems provide the theoretical reason for collecting structured run evidence. Run-level evidence populates the evidence pack. The evidence pack supports scoring across the five pillars. The score profile then becomes a practical governance instrument for review, challenge, audit preparation, and continuous improvement.

Link to the five RAIDT pillars

Responsibility

Socio-technical systems strongly affect Responsibility because accountability in GenAI use sits across humans, tools, and organisational roles. RAIDT asks who initiated, reviewed, approved, and acted on a run, rather than letting accountability disappear into a vague claim that the model produced the output.