Transparency as Answerability: Role-Dependent Information Flow Modeling

Mattias Brännström; Themis Dimitra Xanthopoulou; Lili Jiang

cs.AIOct 12, 2026

Transparency as Answerability: Role-Dependent Information Flow Modeling

Authors: Mattias Brännström, Themis Dimitra Xanthopoulou, Lili Jiang

Organizations: Umeå University, Sweden

Abstract

Transparency in AI-supported decision systems is widely required but poorly defined. Existing approaches treat transparency as a property of systems or components, producing characteristic failures: assessments that do not transfer across organizational boundaries, stakeholder questions that cannot be formulated within the available frame of analysis, and component-level findings that say nothing about system-level behavior. We propose treating transparency as a relation: a system is transparent to a stakeholder to the degree that their legitimate questions are answerable from its structure. We formalize this using information flow networks, in which transparency questions are expressed in terms of network structure and answerability is the condition that the system settles them. The framework supports three dimensions of analysis: structural answerability from the network alone, compositional answerability across organizational boundaries for deployed systems, and role-dependent answerability given a stakeholder's accessible sub-network. We demonstrate the framework on a real pre-integration assessment of an AI recruitment tool in a public sector municipality. The assessment uses participatory design and interviews to model the decision system structure. We show how structural and role-dependent transparency questions can be formally stated and analyzed across the Provider and Deployer boundary where the AI system is composed together with the deployer's decision process.

Figures & tables

#	Channel	Performed by	Operation	Stage
d1	$\mathit{D_{0}}\xrightarrow{\text{Application}}\mathit{CV}$	Candidate	CV submission	Sourcing
d2	$[\mathit{RP}],\mathit{CV}\xrightarrow{\text{Initial scoring}}\mathit{Score_{R}}$	Recruiter	4-grade candidate scoring	Assessment
d3	$[\mathit{RP}],\mathit{CV}\xrightarrow{\text{Initial scoring}}\mathit{Score_{HM}}$	Hiring Manager	4-grade candidate scoring	Assessment
d4	$[\mathit{RP}],\mathit{D_{0}}\times\mathit{Score_{R}},\mathit{Score_{HM}}\xrightarrow{\text{Selection meeting}}\mathit{D_{1}}$	Recruiter, Hiring Manager	Collaborative candidate selection	Selection
d5	$\mathit{Candidate}\xrightarrow{\text{Personality test}}\mathit{Score_{PT}}$	External Provider	Personality testing	Interview
d6	$[\mathit{RP}],\mathit{Candidate},\mathit{Score_{PT}}\xrightarrow{\text{Interview}}\mathit{Score_{comp}}$	Recruiter, Hiring Manager	Competence-based interview	Interview

Table 1. Deployer recruitment process: channels, operations, and stages.

ID	Example legitimate question	Class
Candidate Subject of assessment; seeks fair evaluation
C1	Was I hired, and at what step was I eliminated?	specific run
C2	What criteria was I assessed against?	process structure
C3	How did those criteria affect my outcome?	specific run
C4	What would have changed my result? (structure/logs)	counterfactual
C5	What happened in my assessment? (logs)	specific run

Table 2. Illustrative roles and legitimate questions in the Deployer recruitment process. Roles reflect theoretical information demands and accountability obligations rather than job titles. Questions are illustrative and grounded in the case-study context. Classes distinguish process structure , specific run , counterfactual , statistics over runs , and derived oversight questions. Parenthetical annotations indicate when a question is evaluated specifically against preserved logs.

Panel of Figure 1 — Figure 1. An abbreviated information-flow model of the Deployer recruitment process prior to AI tool integration. For readability, the figure omits some secondary details represented in Table 1 .

Panel of Figure 2 — Figure 2. An abbreviated version of the information flow model of the Provider’s AI system for recruitment. Inputs are a Client recruitment request, captured as a requirement configuration, and a set of candidates. The output is a list of ranked and scored candidates.

#	Channel	Performed by	Operation	Stage
p1	$[\mathit{Template}],\mathit{C_{0}}\xrightarrow{\text{Hard req. filter}}\mathit{C_{1}}$	AI-Filter	Hard requirement filtering	Filtering
p2	$[\mathit{TQ}],\mathit{Candidate}\xrightarrow{\text{Simple questions}}\mathit{A_{rule}}$	Candidate	Rule-based questioning	Interview
p3	$[\mathit{TQ}],\mathit{Candidate}\xrightarrow{\text{Free text questions}}\mathit{A_{text}}$	Candidate	Open response collection	Interview
p4	$[\mathit{TQ}],\mathit{Candidate}\xrightarrow{\text{Competence questions}}\mathit{A_{comp}}$	Candidate	Competence-based questioning	Interview
p5	$[\mathit{TQ}],\mathit{A_{rule}}\xrightarrow{\text{AI rules}}\mathit{S_{rule}}$	AI-Scoring	Rule-based scoring	Scoring
p6	$[\mathit{TQ}],\mathit{A_{text}}\xrightarrow{\text{AI Free text}}\mathit{S_{text}}$	AI-NLP	Semantic text analysis	Scoring

Table 3. Provider AI system: channels, operations, and stages.

Scenario	$\mathbf{D}_{0}$ removes	$\mathbf{D}_{1}$ adaptation	$\mathbf{P}$ replaces	Human evaluation
$\mathbf{Min}^{\mathbf{D}}_{\mathbf{P}}$	d2, d3	$\mathit{C_{1}}$ feeds d2, d3	d1	All candidates passing hard req.
$\mathbf{Alt}^{\mathbf{D}}_{\mathbf{P}}$	d2, d3, d4	Selection meeting receives	d1, d2, d3	All candidates,
		$\mathit{C_{2}}\times\mathit{ScoredTemplate}$		AI-ranked input
		instead of $\mathit{Score_{R}},\mathit{Score_{HM}}$
$\mathbf{Max}^{\mathbf{D}}_{\mathbf{P}}$	d2, d3, d4	Threshold channel added,	d1, d2, d3	$\mathit{top}_{x}$ only
		producing $\mathit{top}_{x}$ , $\mathit{rest}$ ;		proceed to d5, d6

Table 4. Three integration scenarios as instantiations of

C^{\mathbf{D}}_{\mathbf{P}}

					$\text{Max}^{D}_{P}$
Role	Legitimate question	Pre	$\text{Min}^{D}_{P}$	$\text{Alt}^{D}_{P}$	rest	top
Candidate	C1: Was I hired, and at what step was I eliminated?	A	A	A	A
	C2: What criteria was I assessed against?	P	P	P	P
	C3: How did those criteria affect my outcome?	(A)	(A)	(P)	(P † )
	C4: What would have changed my result? (structure/logs)	(A/P ‡ )	(A/P ‡ )	(A/B)	(P † /B)
	C5: What happened in my assessment? (logs)	(P ‡ )	(P ‡ )	(P)	(P † )

Table 5. Answerability of legitimate questions before AI integration and across the three integration scenarios. A = answerable, P = partially answerable, B = blocked; parentheses indicate deferral to another role, and X/Y structural/operational answerability.

\dagger

marks limitations arising from the undisclosed human fallback path, including uncertainty about when it was used and the consequences of outsourced human processing.

\ddagger

marks limitations arising from informal note taking of decisions and meetings.

\text{Max}^{D}_{P}

is split by candidate cohort of top ranking and the rest where needed; – means missing presuppositions where the question is not applicable.

AI-Assisted Decision Making

AI-assisted decision making studies how computational systems can inform human choices while accounting for context, uncertainty, and consequences.

See all 144 papers

AI Transparency: Governance Compliance or Stakeholder Requirements?
Jun 16, 2026Muneera Bano, Didar ZowghiAI Governance
CSIRO Melbourne, Australia · CSIRO Sydney, Australia
Transparent Information Flow Modeling for Role-Aware AI Act Assessment
Oct 12, 2026Mattias Brännström, Themis Dimitra Xanthopoulou, Lili JiangAI Governance Algorithmic Auditing
Umeå University, Sweden
"So There's a Catch-22 Here": How Early Adopters Who Build Multi-Agent LLM Systems Conceptualize Transparency
Jun 6, 2026Suchismita Naik, Samir Passi, Mihaela Vorvoreanu +2Multi-Agent LLM Systems AI Agent Governance
Purdue University, USA · Cornell University, USA · Microsoft Research, USA

Jun 16, 2026J/K move · Enter open · S save

AI Transparency: Governance Compliance or Stakeholder Requirements?

Muneera Bano, Didar Zowghi

CSIRO Melbourne, Australia · CSIRO Sydney, Australia

AI Governance

Open paper AskPDF arXiv

Transparency is increasingly mandated for public-sector AI systems, with organisations required to publish statements describing their AI use and oversight arrangements. However, the existence of such artefacts is often treated as equivalent to transparency itself, despite limited evidence that they proportionately serve relevant stakeholder groups. From a requirements engineering perspective, this raises a validation concern: compliance with mandated disclosure criteria does not necessarily ensure transparency adequacy for stakeholders with different levels of risk exposure, decision control, and involvement. This paper presents an empirical analysis of 92 publicly available AI transparency statements published by Australian Government agencies under the national AI governance mandate. We introduce the stakeholder Risk--Control--Involvement--Need (RCIN) framework to differentiate stakeholder classes according to their structural position and transparency needs. Using a structured rubric derived from the mandated criteria, we evaluate how both the mandate and published statements are calibrated to each stakeholder class. The findings show that while structural compliance is widespread, transparency calibration is uneven. Criteria serving high-control stakeholders are consistently realised, whereas criteria most critical for high-risk, low-control stakeholders are fewer and less substantively addressed. We conceptualise this as the Transparency Illusion: a condition in which transparency appears satisfied through compliant artefacts yet remains unevenly calibrated to stakeholders bearing the greatest exposure to AI-supported decisions. The study frames transparency as a stakeholder-calibrated validation problem, demonstrating that artefact-level compliance does not constitute requirements validation in this context.

Abstract

AI-Assisted Decision Making

Explore similar work

AI Transparency: Governance Compliance or Stakeholder Requirements?

Transparent Information Flow Modeling for Role-Aware AI Act Assessment

"So There's a Catch-22 Here": How Early Adopters Who Build Multi-Agent LLM Systems Conceptualize Transparency