Pith. sign in

REVIEW 2 major objections 5 minor 90 references

AI trustworthiness can be represented as an auditable, learnable mapping from measured profiles to discrete levels.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 21:14 UTC pith:ST42QA45

load-bearing objection A coherent, honest governance methodology whose synthetic proof-of-concept is internally circular; the framework is plausible and worth reviewing, but the empirical claims should be reframed. the 2 major comments →

arxiv 2607.16130 v1 pith:ST42QA45 submitted 2026-07-17 cs.CY cs.AIcs.SYeess.SY

A Methodology for Auditable Trustworthiness Levels in AI Lifecycle Governance

Andrea Ferrario This is my paper
classification cs.CY cs.AIcs.SYeess.SY
keywords AI governancetrustworthiness levelsAI lifecycleauditable rulesdecision treesboundary marginprofile driftexpert labeling
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper aims to make AI trustworthiness a governable object rather than an abstract ideal: it tries to establish that an institution can turn its expert judgments about what counts as acceptable, risky, or unacceptable functioning into a compact, readable rule that maps measured trustworthiness profiles to discrete levels, and then use that rule to monitor and reassess the system across its lifecycle. The method treats trustworthiness as relative to a documented protocol of measurable dimensions, collects expert-labeled anchor regions from pre-deployment review, and learns a shallow decision tree that reproduces those judgments. Two lightweight diagnostics—boundary margin and profile drift—are proposed to show when the system is near a level transition and how fast it is moving in trustworthiness space. If the central claim holds, AI governance gets an evidential middle layer between raw metrics and legal or ethical judgment, without claiming to replace either. The proof of concept runs on synthetic lifecycle traces and is explicitly framed as illustrative, not externally validated.

Core claim

Governance-relative trustworthiness can be formally represented by a protocol PA of measurable dimensions, aggregated into a profile QA(t) in [0,1]^n, and mapped by an interpretable rule bT_A,N to a level in {1,...,K}. The paper argues that expert-defined anchor regions, expressed as threshold constraints on selected dimensions, are enough to build a labeled dataset, and that a shallow decision tree trained on those labels gives an auditable operational rule for post-deployment monitoring. The learned rule may preserve, compress, or refine the expert scale; in stable settings it can be discarded in favor of the expert rule. Level transitions become explicit events, and two diagnostics, bound

What carries the argument

The load-bearing object is the trustworthiness profile QA(t): a vector of dimension scores in [0,1]^n assembled under a documented protocol. Around it sit three constructs: expert-defined anchor regions R1,...,RK, which supply coarse labels; a decision tree—an interpretable threshold rule partitioning the profile space into axis-aligned regions—learned from those labels; and the two diagnostics, boundary margin (distance from the current profile to the boundary of its assigned region) and profile drift (normalized change in profile over a fixed horizon). The tree carries the argument by turning costly, sparse expert judgment into a repeatable rule; the diagnostics give that rule a monitoring

Load-bearing premise

The framework stands or falls on whether the expert-assigned reference labels genuinely capture trustworthiness and whether a rule learned from them keeps working on new deployment data; the simulations make this easy by generating labels from the same threshold rules that serve as training inputs.

What would settle it

Give independent experts real or realistic profiles and have them assign trustworthiness levels without seeing the rule; train the decision tree on their labels and evaluate out-of-time. If agreement among experts is low, if test accuracy approaches chance on held-out segments, or if boundary-margin and drift alerts do not line up with expert-identified level changes, the central generalizability claim would be refuted.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Design-time expert labeling can be reused as a routine monitoring rule, reducing the need to convene full expert panels for every post-deployment check.
  • A level transition, a small boundary margin, or a large profile drift provides documented evidence for deciding whether a change is substantial enough to trigger reassessment.
  • The governance artifact (protocol, labels, learned rule, monitoring results) makes trustworthiness assignments versionable and contestable over time.
  • In stable environments with little regime variation, the methodology's honest output is to keep the expert rule and not learn; this is a governance decision, not a gap.
  • When rare low-trustworthiness states are weakly represented, the learned rule compresses the expert scale and exposes that insufficiency, prompting validation or relabeling.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If reference labels are noisy or contested across raters, the learned rule inherits that disagreement; a natural extension would be to quantify inter-rater reliability and carry it into confidence sets around level assignments—something the paper does not do.
  • The boundary margin could be used to set monitoring cadence or escalation thresholds dynamically, since small margins indicate fragility; this operationalization is left implicit.
  • The same profile-and-rule layer could serve as a common reporting format across multiple deployments of the same system, but the paper's own experiments show that differing institutional labeling granularity would still prevent direct comparability.
  • A probabilistic ordinal model over levels, rather than a point classifier, would give auditors an explicit uncertainty estimate and could be tested on the same synthetic traces.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper proposes a two-part methodology for AI lifecycle governance. The formal framework represents trustworthiness through a documented protocol of measurable dimensions, aggregates them into a profile Q(t), and learns an empirical trustworthiness-level rule from expert-labeled reference data using decision trees (Eq. (1)). The governance procedure embeds the learned rule in pre-deployment labeling, post-deployment monitoring, and reassessment, together with two diagnostics—boundary margin and profile drift—and a reporting artifact. A simulation study with three synthetic lifecycle experiments (two-dimensional baseline, asynchronous four-dimensional lending, and two hospital copies) illustrates the behavior of the framework. The paper is explicitly scoped as a proof of concept, not external validation.

Significance. The formal core is coherent: it cleanly separates trustworthiness protocol, profile, reference labels, and learned rule, and the decision-tree instantiation is a sensible choice for auditable threshold-based governance. The two diagnostics are simple and potentially useful for monitoring. The governance procedure is well-structured and connects to EU AI Act concepts such as conformity assessment, post-market monitoring, and substantial modification. The simulation code and data-generation equations are reported in detail, making the proof of concept reproducible. The paper is honest about the framework's dependence on label quality and protocol maturity (Section 7). If adopted by an institution with a well-documented labeling process, the framework offers a practical translation layer between high-level trustworthiness principles and measurable lifecycle monitoring.

major comments (2)
  1. [Section 5 and Appendix Tables 6, 9, 12] The synthetic labels are deterministic threshold functions of the same profile coordinates used as features. Consequently, the near-perfect results in Tables 7 and 10 largely recover the label-generating function. This does not test the claim in Section 3.3 that the learned rule 'extends costly expert judgment to routine post-deployment monitoring' when labels are noisy, delayed, contested, or depend on unmeasured variables. Section 7 acknowledges the general limitation, but the main text should explicitly state that the simulation uses feature-determined labels and therefore serves as an internal consistency check rather than evidence of generalization. I recommend either adding a simulation with label noise or adversarial labeling, or revising the wording in Section 3.3 and the contribution list to avoid overclaiming.
  2. [Section 3.3, dataset labeling procedure] The paper leaves open how the learned rule behaves in profile-space regions not covered by expert anchor regions R1,...,RK. Since the decision tree assigns a label to every profile, uncovered borderline regions are populated by interpolation/extrapolation from labeled anchor cases. The paper notes borderline cases may be escalated or excluded (footnote 5), but the operational rule will still make assignments there. In the proof of concept this issue is invisible because all profiles lie on trajectories and labels are total. Please either add a discussion of how uncovered regions should be validated (e.g., targeted expert review of tree-split boundaries) or state this as an explicit limitation of the framework.
minor comments (5)
  1. [Section 6.1] The text says 'the learned partition is simple.' Consider showing the actual tree rules in the main text or appendix to make this concrete and support reproducibility.
  2. [Table 7] For the clinical setting, balanced accuracy and macro-F1 on the training segment are lower because some trustworthiness levels are weakly represented. Please report per-class support counts so readers can calibrate these numbers.
  3. [Footnote 3] The reference trustworthiness-level rule T*_A is introduced but not used later. Either integrate it into the main text (e.g., when defining the learning target) or remove it to avoid a dangling concept.
  4. [Section 8.2, simulation equation] The noise terms are written as ε(1)(t) and ε(2)(t), while the text refers to 'noise_std' with separate standard deviations. Please make the mapping between the notation and the parameter table explicit at first use.
  5. [Figure 4] In the Hospital Alpha panel, the 'Operative/predicted TW level' is identical to the 'Local true TW level' by construction because no learned rule is used. The legend or caption should state this explicitly to avoid confusion.

Circularity Check

0 steps flagged

No significant circularity: the formal framework is a supervised-learning/governance translation layer, and the synthetic proof of concept explicitly disclaims external validation.

full rationale

The paper's central derivation is not circular. It defines trustworthiness protocols, profiles, and reference labels as distinct objects, then poses a standard supervised-learning problem: learn an interpretable rule h from profiles QA(t) to expert-assigned labels Yi (Eq. 1). The learned rule is an empirical approximation of the expert labeling procedure, not a redefinition of the target in terms of the model output. In the simulations, reference labels are generated by hand-set threshold rules over the same profile coordinates (Appendix Tables 6, 9, 12), so a decision tree can recover those thresholds; however, this is a self-consistency check, not a hidden reduction. The paper explicitly states: 'The aim of this study is not to establish external validity across domains, but to illustrate how the proposed governance objects behave under controlled conditions' (Section 5), and later notes 'the framework is only as informative as the trustworthiness object that an institution can actually construct and maintain' (Section 7). Experiment 3 also demonstrates that the learned rule can fail to recover the expert scale, showing the mapping is not forced by construction. Self-citations to Ferrario 2024, 2025, and 2026 are contextual and not load-bearing; the formal definitions and learning objective are stated in the paper itself. Therefore no step reduces to its own inputs by definition or by fitted-parameter renaming.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The framework intentionally rests on governance choices (protocol, scale, labeling) rather than deriving them; the simulation adds numerous hand-set coefficients. The main audit finding is that the synthetic labels are generated from the same profiles used as inputs, which weakens the simulation as independent evidence but does not make the formal framework itself circular. No physical or explanatory entities are introduced beyond bookkeeping constructs such as profiles, levels, and diagnostics.

free parameters (5)
  • Expert-defined trustworthiness-level thresholds (simulation labels) = e.g., recommender: q_acc<0.60 → level 1; q_acc≥0.60 & q_rob<0.60 → level 2; else level 3
    Used to generate reference labels Yi; the same labels are used to train and evaluate learned rules, making simulation results partly self-referential.
  • Lifecycle simulation coefficients (baseline b, drift d, shock/update magnitudes, event times, noise std, cadence) = Tables 5, 8, 11
    Hand-set to create synthetic profiles; no real deployment data.
  • Decision-tree hyperparameter grid = Table 4: criterion {gini,entropy}, max_leaf_nodes {3..12,None}, max_depth {2..8,None}, ccp_alpha {0,1e-4,...,1e-1}
    Selected by temporal CV; final trees all gini/max_depth=2/max_leaf_nodes=3/ccp_alpha=0.
  • Trustworthiness-level scale K and protocol dimensions = K=3 or 4; dimensions and metrics per experiment
    Chosen by authors as governance parameters; not derived from data.
  • Boundary-margin distance and profile-drift horizon h = dist unspecified; h=7 (dense) or 1 (sparse)
    Diagnostic outputs depend on these choices; paper says they must be documented.
axioms (5)
  • domain assumption Trustworthiness can be treated as objective relative to a documented governance standard and summarized by a finite profile of dimensions normalized to [0,1].
    Section 3.1 commitments; if trustworthiness is irreducibly contested or non-measurable, the framework cannot be applied.
  • domain assumption Expert reference labels Yi from a documented adjudication procedure are reliable ground truth for learning.
    Section 3.3 dataset labeling procedure; simulation labels are actually generated by the authors' threshold rules.
  • ad hoc to paper Axis-parallel threshold rules (decision trees) are an adequate model class for governance trustworthiness levels.
    Section 3.3; defended as a reasonable first model, not derived from governance requirements.
  • ad hoc to paper The synthetic generation equation z(t)=clip[0,1](b-dτ+ε1+Σs_j 1 + Σu_k 1 + ε2) captures lifecycle dynamics relevant to trustworthiness.
    Appendix 8.1; no real data, so external validity is not established.
  • standard math Temporal validation and standard decision-tree learning assumptions (stable target, informative labels) hold.
    Section 3.3 safeguards and Appendix 8.1; noise and label drift are acknowledged but not modeled.

pith-pipeline@v1.3.0-alltime-deepseek · 22968 in / 13508 out tokens · 103286 ms · 2026-08-01T21:14:41.039928+00:00 · methodology

0 comments
read the original abstract

AI governance increasingly requires judgments about whether an AI system remains adequately trustworthy over time, whether observed changes are tolerable, and how such judgments should be documented in a transparent and contestable way. Yet existing work on AI trustworthiness remains either too high-level to support lifecycle monitoring and reassessment or too narrowly metric-driven to connect with governance needs. We therefore propose a lightweight methodology for auditable trustworthiness levels in AI governance. The methodology has two components: a formal framework for representing and learning trustworthiness levels, and a lightweight AI lifecycle governance procedure for documenting, monitoring, and reassessing them over time. The formal framework models governance-relative trustworthiness through a context-sensitive protocol of measurable dimensions and learns trustworthiness levels as interpretable rules over trustworthiness profiles. Using decision trees as an interpretable proof-of-concept model class, the methodology yields explicit trustworthiness plateaus, readable level transitions, and two simple lifecycle diagnostics: boundary margins and profile drift. The governance procedure embeds these formal objects in a conformity-oriented workflow for design-time labeling, post-deployment monitoring, reassessment, and reporting. It also assigns human responsibilities and control gates for protocol design, validation, monitoring, and reassessment. We illustrate the methodology on synthetic AI lifecycle traces involving degradation, shocks, updates, heterogeneous monitoring cadences, and system comparison. Our methodology does not replace legal or other expert judgment: it supports conformity documentation and lifecycle monitoring by providing an evidential basis for documenting and tracking AI governance-relevant changes over time.

Figures

Figures reproduced from arXiv: 2607.16130 by Andrea Ferrario.

Figure 1
Figure 1. Figure 1: AI Lifecycle Governance Procedure. Phase 1 specifies the AI system, trustworthiness protocol, measures, and labeling procedure. Phase 2 constructs the initial labeled dataset and learns a candidate trustworthiness-level rule. Phase 3 makes explicit the choice of operative governance rule—expert-defined or learned—and then supports post-deployment monitoring, reporting, validation, and possible dataset or p… view at source ↗
Figure 2
Figure 2. Figure 2: Experiment 1: lifecycle baselines. Trustworthiness-profile and level trajectories for a densely monitored recommender system and a sparsely monitored clinical decision-support system. The two systems use the same two￾dimensional profile structure but differ in monitoring cadence, lifecycle dynamics, and expert-defined trustworthiness scales. 6.1 Experiment 1: Lifecycle Baseline Experiment 1 shows that the … view at source ↗
Figure 3
Figure 3. Figure 3: Experiment 2: single-system lifecycle with asynchronous monitoring. The upper panel reports the raw trustworthiness metrics observed at different monitoring cadences. The lower panel reports the resulting latest-available trustworthiness profiles and their reference and predicted trustworthiness levels in the lending use case. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Experiment 3: AI Deployment and Governance-Context Comparison. The two panels compare simulated copies of the same clinical decision-support system. Hospital Alpha remains in the highest trustworthiness regime under a coarse local expert rule, whereas Hospital Beta moves across multiple regimes under a more refined expert-defined rule and lifecycle events involving adverse shocks and corrective updates. 7 … view at source ↗
Figure 5
Figure 5. Figure 5: Experiment 1. Reporting boundary margin and profile drift over time. [PITH_FULL_IMAGE:figures/full_fig_p030_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Experiment 2. Reporting boundary margin and profile drift over time. [PITH_FULL_IMAGE:figures/full_fig_p031_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Experiment 3. Reporting boundary margin and profile drift over time for Hospital Beta. [PITH_FULL_IMAGE:figures/full_fig_p032_7.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

90 extracted references · 3 linked inside Pith

  1. [1]

    2016 , howpublished =

    2016

  2. [2]

    2019 , month = apr, institution =

    2019

  3. [3]

    Recommendation on the Ethics of Artificial Intelligence , year =

  4. [4]

    Conformity assessment under the

    Thelisson, Eva and Verma, Himanshu , journal=. Conformity assessment under the. 2024 , publisher=

    2024

  5. [5]

    Conformity assessments and post-market monitoring:

    M. Conformity assessments and post-market monitoring:. Minds and Machines , volume=. 2022 , publisher=

    2022

  6. [6]

    Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency , pages=

    Nolte, Henrik and Rateike, Miriam and Finck, Mich. Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency , pages=

    2025

  7. [7]

    2019 , organization=

    Towards an ethical code for data-based business , author=. 2019 , organization=

    2019

  8. [8]

    The law of recitals in European Community legislation , author=. ILSA J. Int'l & Comp. L. , volume=. 2008 , publisher=

    2008

  9. [9]

    Trustworthy artificial intelligence and the

    Laux, Johann and Wachter, Sandra and Mittelstadt, Brent , journal=. Trustworthy artificial intelligence and the. 2024 , publisher=

    2024

  10. [10]

    Proceedings of the Conference on Fairness, Accountability, and Transparency (FAT* '19) , year =

    Margaret Mitchell and Simone Wu and Andrew Zaldivar and Parker Barnes and Lucy Vasserman and Ben Hutchinson and Elena Spitzer and Inioluwa Deborah Raji and Timnit Gebru , title =. Proceedings of the Conference on Fairness, Accountability, and Transparency (FAT* '19) , year =

  11. [11]

    Regulating the undefined: Addressing systemic risks in the Digital Services Act (with an appendix on the

    Loi, Michele and Fabbri, Matteo and Ferrario, Andrea , journal=. Regulating the undefined: Addressing systemic risks in the Digital Services Act (with an appendix on the. 2025 , publisher=

    2025

  12. [12]

    Nature , volume=

    Role play with large language models , author=. Nature , volume=. 2023 , publisher=

    2023

  13. [13]

    Communications of the ACM , volume=

    Talking about large language models , author=. Communications of the ACM , volume=. 2024 , publisher=

    2024

  14. [14]

    2024 , month = jun, institution =

    2024

  15. [15]

    2022 , url =

    European declaration on. 2022 , url =

    2022

  16. [16]

    2001 , month = nov, howpublished =

    2001

  17. [17]

    Mitigating the adverse effects of

    W. Mitigating the adverse effects of. Global Business and Organizational Excellence , volume=. 2024 , publisher=

    2024

  18. [18]

    2020 , month = feb, url =

    2020

  19. [19]

    2017 , month = apr, url =

    2017

  20. [20]

    2024 , publisher=

    Anthropomorphising machines and computerising minds: The crosswiring of languages between Artificial Intelligence and Brain & Cognitive Sciences , author=. 2024 , publisher=

    2024

  21. [21]

    Quine, W. V. O. , title =. 1969 , address =

    1969

  22. [22]

    2001 , address =

    Wiggins, David , title =. 2001 , address =

    2001

  23. [23]

    Lowe, E. J. , title =. Artefact Kinds: Ontology and the Human-Made World , editor =. 2014 , pages =

    2014

  24. [24]

    , title =

    Pinter, Charles C. , title =. 2014 , publisher =

    2014

  25. [25]

    , title =

    Olson, Eric T. , title =. 2024 , publisher =

    2024

  26. [26]

    2020 , publisher=

    In AI we trust: Ethics, artificial intelligence, and reliability , author=. 2020 , publisher=

    2020

  27. [27]

    Akbulut, Canfer and Weidinger, Laura and Manzini, Arianna and Gabriel, Iason and Rieser, Verena , booktitle=

  28. [28]

    Formalizing trust in artificial intelligence:

    Jacovi, Alon and Marasovi. Formalizing trust in artificial intelligence:

  29. [29]

    2020 , address =

    Bradford, Anu , title =. 2020 , address =

    2020

  30. [30]

    2020 , address =

    Petit, Nicolas , title =. 2020 , address =

    2020

  31. [31]

    Proposed

    Kazim, Emre and G. Proposed. AI and Ethics , volume=. 2023 , publisher=

    2023

  32. [32]

    Floridi, Luciano , journal=. The. 2021 , publisher=

    2021

  33. [33]

    Ho, Calvin Wai-Loon and Caals, Karel , journal=. How the. 2024 , publisher=

    2024

  34. [34]

    W. The. AI and Ethics , volume=. 2025 , publisher=

    2025

  35. [35]

    Institutionalised distrust and human oversight of artificial intelligence: towards a democratic design of

    Laux, Johann , journal=. Institutionalised distrust and human oversight of artificial intelligence: towards a democratic design of. 2024 , publisher=

    2024

  36. [36]

    A blueprint for auditing generative

    M. A blueprint for auditing generative. 2025 , publisher=

    2025

  37. [37]

    Auditing large language models:

    M. Auditing large language models:. 2024 , publisher=

    2024

  38. [38]

    Smuha, Nathalie A and Rengers, Emma and Harkens, Adam and Li, Wenlong and MacLaren, James and Piselli, Riccardo and Yeung, Karen , journal =. How the. 2021 , month = aug, doi =

    2021

  39. [39]

    Twenty-four years of empirical research on trust in

    Benk, Michaela and Kerstan, Sophie and von Wangenheim, Florian and Ferrario, Andrea , journal=. Twenty-four years of empirical research on trust in. 2025 , publisher=

    2025

  40. [40]

    An overview of key trustworthiness attributes and

    Mattioli, Juliette and Sohier, Henri and Delaborde, Agn. An overview of key trustworthiness attributes and. AI and Ethics , volume=. 2024 , publisher=

    2024

  41. [41]

    Trustworthy artificial intelligence:

    Kaur, Davinder and Uslu, Suleyman and Rittichier, Kaley J and Durresi, Arjan , journal=. Trustworthy artificial intelligence:. 2022 , publisher=

    2022

  42. [42]

    Ala-Pietil. The. 2020 , publisher=

    2020

  43. [43]

    Z-Inspection

    Zicari, Roberto V and Brodersen, John and Brusseau, James and D. Z-Inspection. IEEE Transactions on Technology and Society , volume=. 2021 , publisher=

    2021

  44. [44]

    Justifying our credences in the trustworthiness of

    Ferrario, Andrea , journal=. Justifying our credences in the trustworthiness of. 2024 , publisher=

    2024

  45. [45]

    Trustworthy

    Li, Bo and Qi, Peng and Liu, Bo and Di, Shuai and Liu, Jingen and Pei, Jiquan and Yi, Jinfeng and Zhou, Bowen , journal=. Trustworthy. 2023 , publisher=

    2023

  46. [46]

    2019 , howpublished =

    Ethics guidelines for. 2019 , howpublished =

    2019

  47. [47]

    and Davis, James H

    Mayer, Roger C. and Davis, James H. and Schoorman, F. David , title =. Academy of Management Review , volume =. 1995 , doi =

    1995

  48. [48]

    Ethics-based auditing to develop trustworthy

    M. Ethics-based auditing to develop trustworthy. Minds and Machines , volume=. 2021 , publisher=

    2021

  49. [49]

    Trust, distrust and commitment , author=. No. 2014 , publisher=

    2014

  50. [50]

    Synthese , volume=

    The fine-grained metaphysics of artifactual and biological functional kinds , author=. Synthese , volume=. 2009 , publisher=

    2009

  51. [51]

    A Trustworthiness-based metaphysics of

    Ferrario, Andrea , booktitle=. A Trustworthiness-based metaphysics of

  52. [52]

    Philosophical Explorations , volume=

    The ontology of artifacts , author=. Philosophical Explorations , volume=. 2004 , publisher=

    2004

  53. [53]

    , year =

    Gallois, A. , year =

  54. [54]

    Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

    Social misattributions in conversations with large language models , author=. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

  55. [55]

    Ascribing functions to technical artefacts:

    Vermaas, Pieter E and Houkes, Wybo , journal=. Ascribing functions to technical artefacts:. 2003 , publisher=

    2003

  56. [56]

    Towards a science of

    Rabanser, Stephan and Kapoor, Sayash and Kirgis, Peter and Liu, Kangheng and Utpala, Saiteja and Narayanan, Arvind , journal=. Towards a science of

  57. [57]

    2020 , note =

    Assessment List for Trustworthy Artificial Intelligence (. 2020 , note =

    2020

  58. [58]

    Frontiers in Artificial Intelligence , year =

    Radclyffe, Catherine and others , title =. Frontiers in Artificial Intelligence , year =

  59. [59]

    2023 , number =

    Artificial Intelligence Risk Management Framework (. 2023 , number =

    2023

  60. [60]

    Proceedings of the Conference on Fairness, Accountability, and Transparency , year =

    Mitchell, Margaret and Wu, Simone and Zaldivar, Andrew and Barnes, Parker and Vasserman, Lucy and Hutchinson, Ben and Spitzer, Elena and Raji, Inioluwa Deborah and Gebru, Timnit , title =. Proceedings of the Conference on Fairness, Accountability, and Transparency , year =

  61. [61]

    Datasheets for Datasets , journal =

    Gebru, Timnit and Morgenstern, Jamie and Vecchione, Briana and Vaughan, Jennifer Wortman and Wallach, Hanna and Daum. Datasheets for Datasets , journal =. 2021 , volume =

    2021

  62. [62]

    Bellamy, Rachel K. E. and Dey, Kuntal and Hind, Michael and Hoffman, Samuel C. and Houde, Stephanie and Kannan, Kalapriya and Lohia, Praveen and Martino, Jacqueline and Mehta, Sameep and Mojsilovi. IEEE International Conference on Data Mining Workshops , year =

  63. [63]

    Fairlearn: A Toolkit for Assessing and Improving Fairness in

    Bird, Sarah and Dud. Fairlearn: A Toolkit for Assessing and Improving Fairness in. Microsoft Research White Paper , year =

  64. [64]

    Arya, Vijay and Bellamy, Rachel K. E. and Chen, Pin-Yu and Dhurandhar, Amit and Hind, Michael and Hoffman, Samuel C. and Houde, Stephanie and Liao, Q. Vera and Luss, Ronny and Mojsilovi. One Explanation Does Not Fit All: A Toolkit and Taxonomy of. Proceedings of the IJCAI 2019 Workshop on Explainable Artificial Intelligence , year =

    2019

  65. [65]

    and others , title =

    Zicari, Roberto V. and others , title =. arXiv preprint arXiv:2206.09887 , year =

    Pith/arXiv arXiv

  66. [66]

    and others , title =

    Zicari, Roberto V. and others , title =. IEEE International Conference on Technology and Society , year =

  67. [67]

    2021 , number =

    Tools for Trustworthy AI: A Framework to Compare Implementation Tools for Trustworthy AI Systems , institution =. 2021 , number =

    2021

  68. [68]

    Proceedings of the Workshop on AI Governance , year =

    McCormack, Laura and others , title =. Proceedings of the Workshop on AI Governance , year =

  69. [69]

    and others , title =

    Schlicker, N. and others , title =. Computers in Human Behavior , year =

  70. [70]

    arXiv preprint arXiv:2602.16666 , year =

    Rabanser, Stephan and Kapoor, Sayash and Kirgis, Peter and Liu, Kangheng and Utpala, Saiteja and Narayanan, Arvind , title =. arXiv preprint arXiv:2602.16666 , year =

    Pith/arXiv arXiv

  71. [71]

    arXiv preprint arXiv:2603.16586 , year =

    Kaptein, Maurits and Khan, Vassilis-Javed and Podstavnychy, Andriy , title =. arXiv preprint arXiv:2603.16586 , year =

    arXiv

  72. [72]

    Pinar and Casper, Stephen and Kolt, Noam , title =

    Staufer, Leon and Feng, Kevin and Wei, Kevin and Bailey, Luke and Duan, Yawen and Yang, Mick and Ozisik, A. Pinar and Casper, Stephen and Kolt, Noam , title =. arXiv preprint arXiv:2602.17753 , year =

    Pith/arXiv arXiv

  73. [73]

    Machine learning operations (

    Kreuzberger, Dominik and K. Machine learning operations (. IEEE Access , volume=. 2023 , publisher=

    2023

  74. [74]

    Journal of Systems and Software , volume=

    Adaptive data quality scoring operations framework using drift-aware mechanism for industrial applications , author=. Journal of Systems and Software , volume=. 2024 , publisher=

    2024

  75. [75]

    A multivocal review of

    Eken, Beyza and Pallewatta, Samodha and Tran, Nguyen and Tosun, Ayse and Babar, Muhammad Ali , journal=. A multivocal review of. 2025 , publisher=

    2025

  76. [76]

    Towards a better understanding of evaluating trustworthiness in

    Kemmerzell, Nils and Schreiner, Annika and Khalid, Haroon and Schalk, Michael and Bordoli, Letizia , journal=. Towards a better understanding of evaluating trustworthiness in. 2025 , publisher=

    2025

  77. [77]

    Creating Characteristically Auditable Agentic

    Phiri, Charles Chimwemwe , booktitle=. Creating Characteristically Auditable Agentic

  78. [78]

    Understanding

    Mehrotra, Siddharth and Huang, Jin and Fu, Xuelong and Dobbe, Roel and S. Understanding. Journal of Artificial Intelligence Research , volume=

  79. [79]

    Nature Communications , volume=

    Active label cleaning for improved dataset quality under resource constraints , author=. Nature Communications , volume=. 2022 , publisher=

    2022

  80. [80]

    2019 , month=jun # " 11", publisher=

    Structured labeling to facilitate concept evolution in machine learning , author=. 2019 , month=jun # " 11", publisher=

    2019

Showing first 80 references.