PaperlayerView exact revision on arXiv

Source-linked AI summary

Executive Judgement in AI-Mediated Decision-Making Environments: A Process Theory of Formation, Qualification Attrition and Authorisation

Richard Hill

arXiv:2609.39442v1cs.CY

TL;DR

Distributed human and AI contributions can produce an inadequately warranted organisational commitment when consequential grounds do not survive transformations before authorisation. The article develops a process theory linking qualification attrition and a formation–authorisation gap to governance arrangements; its central practical conclusion is to examine the transformation chain rather than infer sound governance from adoption, fluency, accuracy, or nominal oversight.

  • Problem

    The article asks whether consequential grounds survive the transformations through which distributed judgement becomes an authorised commitment, rather than treating hybrid cognition itself as the unresolved question.

    Source: §1 · p. 2

  • Method

    It develops a process theory of executive judgement governance that links qualification attrition and the formation–authorisation gap to boundary setting, interpretive challenge, reliance calibration, and authorisation with answerability.

    Source: §5.2 · p. 11, §5.3 · p. 12, §9 · p. 21

  • Results

    The theory argues that technical competence and appropriate local reliance do not guarantee that consequential qualifications survive aggregation into an organisational commitment.

    Source: §9 · p. 21

  • Takeaways & Limitations

    Examine the transformation chain between judgement formation and authorisation rather than infer sound governance from adoption, fluency, accuracy, or nominal oversight.

    Source: §9 · p. 21

    The claim that executive value lies primarily in governing warranted commitments is normative and theoretical, not an empirically established ranking of skill value.

    Source: §1 · p. 3

Abstract

from arXiv · show

Generative artificial intelligence can contribute to the representations, alternatives and evaluations through which executive judgements are formed. Existing research already explains important aspects of hybrid cognition, reliance, managerial agency and accountability. This article develops a narrower process challenge: locally competent human and AI contributions can still culminate in an inadequately warranted organisational commitment when established operational assumptions, uncertainties or dependencies are lost or transformed before authorisation. It proposes qualification attrition as the process through which such epistemically consequential content is weakened across successive transformations, and a formation--authorisation gap as the resulting discrepancy between the grounds required for warranted commitment and those that remain accessible, intelligible and challengeable to the authorising role. Executive judgement governance is defined through four dimensions: boundary setting, interpretive challenge, reliance calibration, and authorisation with answerability. Four propositions connect transformation, qualification attrition and governance arrangements to epistemic risk, accountability risk and judgement quality. The contribution is an integrative process theory of how distributed judgement formation can fail as a composition even where individual contributions, local reliance decisions and formal accountability arrangements appear competent.

1 Introduction

The article asks whether the consequential grounds of distributed judgement survive the transformations that turn provisional analysis into authorised organisational commitment. It proposes a governance process focused on preserving those grounds, rather than on executive personality or a broad claim that AI participation is theoretically unrecognised.

1 Introduction

  • Locally competent contributions can still produce an inadequately warranted commitment when assumptions, uncertainty, dependencies, or dissent are weakened as analysis moves toward authorisation.No single contributor need make an obvious error; the problem can arise from how reasonable contributions compose.Source: §1 · p. 2
  • Research already theorises hybrid cognition, managerial phronesis, and organisational knowing, so the article rejects a broad claim that AI participation in judgement remains theoretically unrecognised.The cited work addresses conjoined agency, generative AI in organisational decisions, context-sensitive moral judgement, and LLMs’ effects on organisational knowing.Source: §1 · p. 2
  • The theoretical issue is whether consequential assumptions, uncertainties, and dependencies remain accessible, intelligible, and challengeable when analysis becomes commitment.The article distinguishes this issue from distribution itself: a distributed process can preserve epistemic access, while a simple process can conceal a decisive assumption.Source: §1 · p. 2
  • The article introduces qualification attrition for weakened consequential qualifications and the formation–authorisation gap for the resulting discrepancy in grounds available to the authorising role.Executive judgement governance concerns configuring, challenging, calibrating, and rendering answerable the processes through which distributed contributions become commitments.Source: §1 · pp. 2–3
  • The article argues that executive value lies primarily in making and governing warranted commitments, while treating this as a normative theoretical position rather than an empirical ranking of skill value.It neither treats optimisation as intrinsically undesirable nor claims humans necessarily outperform AI; when executive intervention adds value remains conditional.Source: §1 · p. 3

2 Conceptual approach and scope

The article uses an integrative conceptual synthesis and a targeted literature search to develop a process account of executive judgement. Its unit is a consequential organisational decision episode, with distinctions among judgement formation, decision, cognition, authority, and accountability delimiting the scope.

2 Conceptual approach and scope

  • The article combines foundational theories with a targeted search of scholarly publications and repositories on human–AI collaboration, organisational decision-making, reliance, learning, knowing, governance, and accountability.The search covered literature available before 29 September 2026, particularly publications from 2022 onward, and distinguished different publication and evidence types.Source: §2 · p. 3
  • The search was directed by the theoretical problem, included sources that challenge the contribution, and was not an exhaustive or reproducible systematic-review sample.Implications from laboratory tasks and particular occupations are treated as theoretical extensions to executive work.Source: §2 · p. 3
  • The focal unit is a consequential decision episode in which an organisational problem is framed, responses are evaluated, an authorised commitment is made, and its grounds may later be reconsidered.An executive role has delegated authority to commit organisational resources or approve policies, and may be enacted collectively by a board or executive team.Source: §2 · p. 3
  • Judgement formation develops an appraisal, while a decision selects or authorises action; cognition concerns representations, authority permits commitment, and accountability entails answering to an evaluative forum.These distinctions prevent distributed production from being mistaken for collective consciousness or a complete allocation of responsibility.Source: §2 · pp. 3–4

3 Foundations and the contemporary theoretical problem

The theoretical foundations frame AI-mediated judgement as functional participation in distributed cognition, while distinguishing performance, reliance, and organisational authorisation. The article locates its narrower contribution in whether warrant-relevant qualifications survive the process from judgement formation to commitment.

3.1 Judgement under constraint

  • Bounded rationality conditions choice on limits to knowledge and computation, while expertise and intuition do not by themselves justify rejecting an accurate recommendation.AI may relax some constraints while leaving objective specification and evidence evaluation unresolved; the sources also frame executive commitments as shaped by political and organisational processes.Source: §3.1 · p. 4

3.2 Functional participation in cognition

  • AI participates functionally in cognition when its outputs affect decision representations, without that participation implying consciousness, moral agency, or authority to act.Participation can range from a one-off drafting request to a recurring advisory workflow or delegated agent sequence.Source: §3.2 · p. 5

3.3 Performance and warranted reliance

  • Human–AI combinations do not on average outperform the better human-only or AI-only alternative in the experimental comparisons synthesised by Vaccaro et al. (2024), though effects vary across tasks.Improving on unaided humans is not the same comparator as improving on the strongest available arrangement, and the finding does not establish a particular executive team’s performance.Source: §3.3 · p. 5
  • Advice use varies by setting, and appropriate reliance depends on whether behaviour matches system suitability for the task rather than on trust alone.Useful assistance in professional writing or customer support does not establish improved executive judgement; reported associations between GenAI confidence and self-reported critical thinking do not establish causal decline.Source: §3.3 · pp. 5–6

3.4 From adjacent theories to a narrower contribution

  • Adjacent research already addresses distributed cognition, hybrid agency, organisational knowing, validation, and accountability, making the paper’s contribution narrower than a claim of theoretical absence.The remaining question is whether warrant-relevant qualifications survive the transformations leading to authorised commitment.Source: §3.4 · pp. 6–7
  • Qualification attrition concerns selective degradation of consequential qualifications, not simply reduced information volume or variation, and may occur with or without substantial compression.The construct adds value only if it provides explanatory or predictive leverage beyond information compression theory.Source: §3.4 · p. 6
  • The formation–authorisation gap concerns grounds available to the organisational authoriser, not model opacity or source unverifiability alone.An opaque source can coexist with a comparatively small gap when validation evidence, limitations, uncertainty, and dependencies remain usable at authorisation.Source: §3.4 · pp. 6–7

4 From distributed formation to authorised commitment

The section links distributed judgement formation to authorisation through an asymmetry of participation and authority, then specifies qualification attrition as a process that can create a role-relative gap in warranted grounds. It distinguishes this epistemic account from questions of responsibility and defines risks and judgement quality without reducing them to accuracy, documentation, or eventual success.

4.1 The conditional asymmetry

  • Judgement may be formed across many people and systems while authorisation and answerability remain comparatively concentrated; the relevant inquiry is the dependencies among grounds for action, not contributor counts.AI can change the volume, speed, and recombination of contributions, but a single well-supported analysis may offer a shorter, more inspectable route to commitment.Source: §4.1 · p. 7
  • The formation–authorisation gap is the discrepancy between grounds relevant to warranted commitment and those accessible, intelligible, and challengeable to the authorised role.The gap can be small in a highly distributed process and large in a seemingly simple one; authority may also be exercised through prior approval of a policy rather than approval of each transaction.Source: §4.1 · pp. 7, 9

4.2 Qualification attrition as the linking mechanism

  • Qualification attrition occurs when transformations weaken or remove qualifications that matter to whether a commitment is warranted, even though compression itself is not necessarily defective.The process can involve evidence becoming analysis, model output, recommendation, briefing, or delegated policy; reasonable local summaries can compose into an overqualified conclusion.Source: §4.2 · p. 9
  • Qualification attrition is distinct from generic information loss, and the formation–authorisation gap is its possible relational state at commitment rather than the same process or an inevitable risk.Whether consequential content survives in usable form matters more than distribution itself; AI may intensify the mechanism, but it predates AI and warrants comparison with conventional processes.Source: §4.2 · pp. 9–10

4.3 Responsibility is not a single residual obligation

  • Authority, accountability, moral responsibility, organisational responsibility, professional obligations, and legal liability are distinct, and accountability concerns an actor–forum relationship.Authority may sit with someone who has limited information, while expertise may come from someone without authority; responsibility should not be assigned solely to the visible operator where effective control lies elsewhere.Source: §4.3 · p. 10

4.4 Why the gap matters for judgement quality

  • Epistemic and accountability risks can diverge, and neither documentation nor predictive accuracy alone establishes judgement quality.Judgement quality concerns appraisal and commitment relative to evidence, objectives, constraints, alternatives, consequences, and capacity to revise; outcome and justification quality should be assessed separately.Source: §4.4 · pp. 10–11

5 Executive judgement governance

Executive judgement governance regulates how distributed contributions become warranted, answerable commitments through four connected dimensions. These dimensions govern task boundaries, representations, evidential reliance and accountable commitment.

5.1 Construct definition and level of analysis

  • Executive judgement governance is a process capability, not an executive personality type, and its four formative dimensions are boundary setting, interpretive challenge, reliance calibration, and authorisation with answerability.The dimensions are not interchangeable symptoms of one trait, and the model does not claim a validated scale or necessary-and-sufficient taxonomy.Source: §5.1 · p. 11

5.2 Boundary setting

  • Boundary setting aligns delegated AI use with task demands and available performance evidence, while revising scope as performance emerges and ensuring workflows can enforce limits.In novel domains, experimentation may be permitted while irreversible commitment is limited; overly restrictive boundaries can exclude useful evidence or preserve poor human practice.Source: §5.2 · pp. 11–12

5.3 Interpretive challenge

  • Interpretive challenge tests assumptions, exclusions and value choices, preserving distinct representations; it needs independent knowledge and permission to contest, but excessive reframing can delay commitment.Generated disagreement alone may repeat the original model's premises, and maximum dissent is not equivalent to maximum judgement quality.Source: §5.3 · p. 12

5.4 Reliance calibration

  • Reliance calibration adjusts human and AI contributions by competence, uncertainty and task relevance, distinguishing acceptance, checking, modification or rejection.It rejects blanket trust, indiscriminate rejection and plausibility-only checks; cognitive forcing reduced overreliance in one experimental task but carried subjective-evaluation costs. Calibration also depends on boundaries, interpretive challenge and feedback.Source: §5.4 · pp. 12–13

5.5 Authorisation and answerability

  • Authorisation with answerability links a commitment to identifiable authority, inspectable reasons, intervention rights and an account-demanding forum, with access to uncertainty and the ability to alter or refuse action.A signature alone may not expose accepted grounds or actual authority; answerability should track influence and control rather than require executives to reconstruct every model operation. If adequate access is unavailable, reducing commitment scope or irreversibility may be preferable to more documentation.Source: §5.5 · pp. 13–14

6 Conceptual model and research propositions

The model locates risk in consequential qualifications weakened across transformations before authorisation, rather than in distributed participation itself. It proposes governance dimensions and propositions linking qualification attrition to epistemic and accountability risks.

6.1 Process relationships

  • The model distinguishes participation and dependency structure, epistemic transformations, qualification attrition, the formation–authorisation gap, and resulting risks and judgement outcomes.Transformations select, summarise and recombine representations; attrition is the loss, attenuation or obscuring of consequential qualifications through those transformations.Source: §6.1 · p. 14
  • Boundary setting, interpretive challenge, reliance calibration, and authorisation with answerability are proposed to shape qualification preservation and the grounds usable at commitment.Their adequacy is proposed to affect epistemic and accountability risk and judgement quality.Source: §6.1 · p. 14

6.2 Propositions

  • Greater qualification attrition across successive transformations increases the formation–authorisation gap, even when individual contributions are locally competent.The resulting commitment may rest on thinner grounds than the process originally contained.Source: §6.2 · pp. 14–15
  • A larger formation–authorisation gap is proposed to increase epistemic risk even when task difficulty, analytical performance and local reliance appropriateness are held constant.P2 tests whether the gap adds explanatory value beyond component performance and local reliance.Source: §6.2 · pp. 15–16
  • Under significant ambiguity, independently grounded interpretive challenge is proposed to reduce attrition and epistemic risk more than adding outputs from shared premises.The recovered qualifications must remain visible and evaluable at authorisation.Source: §6.2 · p. 16
  • Reducing the gap while preserving effective intervention rights is proposed to reduce accountability risk more than nominal human approval without changes to access or control.The proposition connects epistemic access at commitment to qualification attrition and the formation–authorisation gap, rather than proposing a new accountability theory.Source: §6.2 · p. 17
  • The propositions distinguish process, epistemic and accountability outcomes, while treating preservation of consequential qualifications—not every detail—as the challenge.The proposed causal relationships are not empirically estimated; Figure 1 separates transformations from attrition and depicts feedback across later episodes.Source: §6.1 · p. 14, Figure 1 · p. 16, §6.2 · p. 17

7 Domain illustrations and evidential limits

The domain illustrations apply the process model to strategic evaluation, clinical advice, credit decisions and agentic workflows. The cited evidence motivates these applications but does not validate the executive construct or establish that the governance dimensions improve outcomes.

7.1 Strategic evaluation

  • In strategic evaluation, stable aggregated AI rankings may resemble expert rankings, but agreement with experts is not realised strategic success or proof of more defensible commitments.The committee example asks whether criteria fit the strategic issue and whether assumptions omitted from the ranking remain visible; these implications are interpretations, not observed executive behaviours.Source: §7.1 · p. 17

7.2 Clinical advice and organisational deployment

  • Clinical advice experiments document susceptibility to incorrect advice across expertise levels, while hospital-dataset studies address variable model generalisation rather than clinician reliance.Gaube et al. (2021) study responses to diagnostic advice; Zech et al. (2018) examine model performance and dataset characteristics. Neither validates the proposed executive construct or determines legal liability.Source: §7.2 · p. 18

7.3 Credit decisions

  • In credit decisions, predictive performance and distributional outcomes are distinct: better prediction does not establish unqualified improvement for every borrower group.The illustration distinguishes evaluating a predictor from authorising a policy that uses its predictions to allocate credit; the cited empirical work does not establish that the four governance dimensions improve credit decisions.Source: §7.3 · p. 18

7.4 Agentic workflows

  • For agentic workflows, the focal commitment may be approval of a policy governing a class of actions, with monitoring and exceptions shaping challenge and calibration.As intervention time decreases, the model implies that permissible-action limits and reversibility may need to be set before execution; it does not claim continuous executive supervision is possible.Source: §7.4 · p. 18

8 Boundary conditions and empirical development

The argument is most applicable when AI influences consequential premises, evidence is contested, and authority is identifiable; empirical work should test its constructs and boundary conditions. The authors specify operational measures and identify rival explanations and findings that would weaken the model.

8.1 Where the argument applies

  • The model fits consequential AI-influenced decisions with contested evidence and identifiable authority less well when knowledge and authority are co-located, decisions are decentralized, or organisational control over AI use is absent.Power may also let executives use governance routines to legitimise predetermined commitments rather than improve their grounds.Source: §8.1 · p. 19

8.2 Operationalisation and research design

  • Researchers can measure qualification attrition by tracing whether consequential qualifications survive successive representations, then compare authorisers’ accounts of grounds with the documented formation process.The formation–authorisation gap should be assessed through accessibility, intelligibility and challengeability, not inferred from outcomes.Source: §8.2 · pp. 19–20

8.3 Falsification and rival explanations

  • The model is weakened if its dimensions add no explanatory value beyond ordinary decision quality, expertise, resources and existing AI governance, or if its propositions fail under controlled comparisons.Reverse causality and organisational resources may confound governance–outcome associations; adjacent theories may also explain the same variation.Source: §8.3 · p. 20

9 Conclusion

The paper narrows the problem of human–AI decision-making to whether warrant-relevant qualifications survive transformation into authorised commitments. It proposes a process theory and governance dimensions, while presenting the account as bounded and still requiring empirical tests.

Conclusion

  • The paper locates the central problem in compositional epistemic failure between distributed judgement formation and authorised commitment, as consequential qualifications may be lost across transformations.The resulting gap can leave authorisers with apparently sufficient grounds that omit qualifications needed for warranted commitment.Source: §9 · p. 21
  • Executive judgement governance addresses this process through boundary setting, interpretive challenge, reliance calibration, and authorisation with answerability.The argument reframes executive capability around preserving the epistemic adequacy of commitments across transformations.Source: §9 · p. 21
  • The practical implication is to examine the transformation chain rather than infer sound governance from adoption, fluency, accuracy or nominal oversight.The article offers a bounded process theory and research agenda, not a demonstrated solution; future work should test whether the constructs explain epistemic risk beyond existing factors.Source: §9 · p. 21
Loading 2609.39442v1…