S1.04 - Information_disorder

S1.04 ? Information disorder

flowchart LR
    A[Background problem
misinformation, disinformation, stale sources, noisy retrieval] --> B[RAIDT
run-level evidence framework]
    H[Practical fields
source provenance, dates, prompts, retrieval logs, verification notes] --> C[[Information disorder
information-integrity risk within a run]]
    B --> C
    C --> D[Evidence pack]
    C --> E[RAIDT score profile]
    D --> F[Reviewer reconstruction
and contestability]
    E --> G[Governance readiness
and organisational learning]

? Star S1 - Origins, Background and History

Star context: Explains why RAIDT emerged from Responsible AI, managerial uncertainty, IS governance, audit traditions, and GenAI operational pressure by showing that organisational use of GenAI is vulnerable to misinformation, disinformation, noisy retrieval, stale context, and weak source control.

Academic picture

Definition / background

Information disorder refers to the condition in which unreliable, deceptive, stale, partial, contradictory, or weakly contextualised information enters a communicative or decision process. In the broad literature, the term is often associated with misinformation and disinformation. In organisational GenAI governance, however, the issue is wider. It includes noisy prompts, poor retrieval results, outdated knowledge-base entries, copied text of uncertain provenance, inconsistent internal records, and plausible but unsupported generated claims. The concept therefore matters because generative AI systems can absorb, combine, and restate problematic information with speed and fluency, making defects harder to notice once they appear in an apparently coherent output.

Within RAIDT, information disorder is important because the framework is built around the run as the unit of governance. RAIDT does not assume that unreliable information can be eliminated in advance. Instead, it asks whether one specific run contains enough evidence about sources, prompts, retrieval, checking, and human intervention to show how information quality was managed. This makes information disorder a run-level governance issue rather than only a general epistemic concern.

The concept also helps distinguish several related problems. Misinformation concerns inaccuracy without necessary intent to deceive; disinformation implies deliberate deception; noisy inputs may simply be low-quality, irrelevant, duplicated, or weakly matched to the task. RAIDT groups these together under the practical governance problem of whether a run relied on information that could not be trusted, contextualised, or adequately reviewed. That is why the concept belongs in this star: it helps explain the background pressures that make evidence-based governance necessary.

Information disorder is directly connected to RAIDT's two outputs. A run-level evidence pack should show what information entered the run and what checks were applied. The five-pillar score profile should reflect whether responsibility for verification was clear, whether the run was auditable, whether the use of information was interpretable, whether outputs were dependable given the input conditions, and whether sources and transformations were traceable. In this sense, information disorder is not peripheral to RAIDT. It is one of the reasons RAIDT exists.

Why this concept matters

Information disorder matters because organisations increasingly use GenAI in contexts where information quality cannot be assumed. Staff may work with mixed internal and external materials, rapidly retrieved documents, user-provided text, and model-generated summaries that look confident even when the underlying informational basis is weak. Without a governance concept for this problem, organisations risk mistaking fluency for reliability.

The concept also prevents a common governance error: assuming that if a model is approved, or if a workflow is documented, then the information used in a particular run is automatically fit for purpose. In practice, many failures arise not because the model is intrinsically defective but because the run absorbed misleading sources, stale references, contradictory documents, or unverified claims. RAIDT makes that problem visible by requiring evidence about what informed the run and what was done to test it.

If this concept is missing, organisations are left with vague statements about quality assurance but limited ability to explain why a problematic output appeared. They may struggle to contest a misleading answer, learn from a failure, or defend their process to supervisors, auditors, or stakeholders. RAIDT uses information disorder to move governance from broad concern to operational scrutiny.

Key idea: Information disorder matters because RAIDT makes unreliable information a reviewable property of a specific GenAI run rather than a vague background risk.

What this item explains

How unreliable or weakly sourced information can enter a GenAI run through prompts, retrieval, source materials, or user inputs.
Why fluent model outputs can conceal misinformation, disinformation, missing context, or stale evidence.
Why source visibility, provenance checks, and verification actions are necessary for run-level governance.
How information quality affects the evidential strength of a RAIDT evidence pack.
Why weak information conditions can lower confidence in a RAIDT score profile across several pillars.
How organisations can convert a general concern about misinformation into a concrete review process tied to one run.

Practical example / likely audience question

Audience question

How does RAIDT handle unreliable inputs?

Answer

The concern behind this question is that a governance framework may identify the problem of unreliable information without offering a practical response. The direct answer is that RAIDT does not claim to remove all uncertainty or guarantee truth automatically. Instead, it makes the informational basis of a run visible enough for review. That means capturing evidence about where inputs came from, how retrieval worked, what prompt framing was used, what checks were applied, and who judged the output safe or unsafe for use.

Consider a team using GenAI to draft a policy briefing from a mixture of internal reports, web extracts, and previously generated notes. A generic governance approach may say that staff should verify outputs, but it may not preserve enough detail to reconstruct what information the model actually used. RAIDT is stronger because it ties governance to run-level evidence. If the briefing later contains an inaccurate claim, reviewers can examine the specific source set, retrieval path, prompt wording, generated text, and verification decisions rather than relying on post hoc recollection.

This is why RAIDT handles unreliable inputs better than a generic principles-only approach. It does not treat information quality as a background aspiration. It turns it into something inspectable, contestable, and auditable within a specific use event.

Practical example in RAIDT terms

Consider a public-service setting in which a local authority uses GenAI to draft a housing-support case summary from citizen emails, uploaded tenancy documents, internal guidance notes, and retrieved policy extracts. The GenAI use case is attractive because it saves staff time, but the run-level issue is information disorder: one retrieved policy extract is out of date, one email contains an unsupported allegation, and the prompt asks for a concise summary without requiring explicit uncertainty flags.

The evidence needed includes the task definition, the source files and their dates, retrieval results, the prompt template, the model and configuration used, the generated summary, reviewer annotations, and a note of what information was verified, rejected, or escalated. Responsibility is affected because the caseworker or supervisor must be clearly accountable for checking source validity before using the output. Auditability is affected because a reviewer should be able to reconstruct how the obsolete policy extract influenced the summary. Interpretability is affected because the relationship between source material, prompt framing, and output wording has to be understandable. Dependability is affected because output quality is unstable when source quality is unstable. Traceability is affected because the authority needs a clear record of which documents and retrieval results shaped the run.

In governance-readiness terms, this item improves practice by showing that the problem is not simply that GenAI can hallucinate. It is also that the run may be fed disordered information. RAIDT makes that condition visible and manageable through evidence capture, review checkpoints, and score-based reflection on whether the run was governed well enough for organisational use.

Detailed link to RAIDT

Information disorder links to RAIDT in four ways.

First, it supports the RAIDT core idea that GenAI governance must address actual conditions of use, including the quality and provenance of information entering a run.

Second, it links directly to the run because information disorder is not meaningfully governed in the abstract; it must be examined in relation to one task, one context, one source set, and one output episode.

Third, it shapes both the evidence pack and the RAIDT score profile because source reliability, retrieval quality, verification steps, and uncertainty handling all affect the evidential basis on which a run is judged.

Fourth, it strengthens reviewability, contestability, audit readiness, and organisational learning by allowing reviewers to identify where unreliable information entered the process and how the organisation responded.

Information disorder ? Run-level evidence ? Evidence pack ? RAIDT score profile ? Governance readiness

Link to the five RAIDT pillars

Responsibility

Information disorder affects Responsibility because organisations must show who was expected to validate sources, notice uncertainty, and decide whether an output could be used safely.