Pith. sign in

REVIEW 5 major objections 5 minor 52 references

Ethics through the Facets of Artificial Intelligence

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper argues that AI ethics failures stem from a blurred understanding of AI, and that assessing AI through three facets—science, technology, and sociotechnical imaginary—under a three-layer responsibility framework can steer AI…

desk verdict A readable synthesis of AI ethics with an original Borgmann analogy, undermined by a promissory first-order logic that clashes with its own dynamic-values claim. read the letter →

arxiv 2507.17020 v1 pith:Z46SG2JU submitted 2025-07-22 cs.CY

Flavio Soares Correa da Silva This is my paper · ORCID
classification cs.CY
keywords artificialintelligenceethicssociotechnicalimaginaryvirtuetechnologyassessmentAIgovernancethreefacetsofresponsibilityframeworkliteracy
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that most ethical failures around AI are not caused by any single technology but by a blurred understanding of what AI is: it is simultaneously a scientific endeavour to model intelligence, a family of problem-solving technologies, and a sociotechnical imaginary that organisations use to steer public decisions. From this diagnosis the author derives a framework for ethical assessment in three layers—transparency and literacy, shared ethical values, and regulations—assigned to the three stakeholder kinds that the imaginary facet brings into view: designers, users, and owners. The author claims that such a framework should be possible and helpful for ensuring AI serves society, and that it can be implemented as a first-order logical theory of virtue-driven ethical principles, with an executable specification currently under development.

What carries the argument

The load-bearing machinery is the three-facet characterisation of AI (scientific endeavour, technology, sociotechnical imaginary), which organises the entire ethical analysis. On top of it rests a three-layer responsibility framework—transparency and literacy at the baseline, ethical values in the middle, regulations at the top—that assigns obligations to designers, users, and owners. A further correspondence maps each facet to a mode of human-information interaction (information about reality, for reality, and as reality), which the paper uses to rank the difficulty of ethical assessment across facets. Finally, the promised implementability rests on a first-order logical theory meant to formalise virtue ethics; this theory is asserted but not yet presented.

What would settle it

A concrete counterexample would be an ethical judgment whose correct application depends on unstructured context or on who is judging (so that any first-order encoding either trivialises it or becomes inconsistent); showing that no such encoding exists would refute the implementability claim. Conversely, a small executable prototype that derives use-and-quality obligations for a specific AI tool from first-order virtue axioms would settle the question in the paper's favour.

Watch

Extended reading notes

Core claim

The paper's central claim is that the 'blurred understanding' of AI—conflating AI as science, AI as technology, and AI as a shared social imaginary—is a main source of ethical concern. By separating the three facets, the paper shows that each facet involves different stakeholders and different ethical problems: science strives toward behavioural alignment with biological intelligence; technology is judged by problem-solving effectiveness; the imaginary invites collective projection of possible futures and the values attached to them. The proposed response is a three-layer responsibility framework in which the baseline layer ensures transparency and literacy, the intermediate layer anchors shared ethical values through accredited institutions, and the top layer enforces alignment through regulations. The paper positions this framework as practical, not merely conceptual: it asserts that a first-order logical theory can express virtue-driven ethical principles and support formal verification of proper use and quality assessment of AI tools.

Load-bearing premise

The framework's promised implementability depends on the assumption that virtue-driven ethical principles can be captured in a first-order logical theory and used to verify the proper use and quality assessment of AI systems—an assumption stated without a formal sketch.

Editorial extensions

If this is right

  • Ethics assessment becomes a stakeholder audit: designers and developers document capabilities and limits, users are trained in proper use and expectations, and owners are held to realistic claims about their tools.
  • Because the sociotechnical imaginary facet is where misinformation and profit-driven distortion enter, the framework implies that curbing that distortion requires institutional guardrails, not just better technology.
  • The facet-to-information-mode correspondence predicts that ethical evaluation becomes harder as AI moves from imitating behaviour (about reality), to prescribing actions (for reality), to substituting for reality in shared imaginaries.
  • If the first-order logical theory is realised in an executable specification, ethical claims about AI use could in principle be checked mechanically, turning moral alignment into a verification task.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A practical diagnostic follows that the paper does not spell out: when an AI ethics debate talks past itself, the disputants are often invoking different facets (one argues from technological capability, another from the imaginary), and naming the facet may resolve the impasse.
  • The framework's middle layer requires accredited institutions to maintain values, but the paper gives no account of how such institutions are themselves kept trustworthy; this is a natural next problem for the programme.
  • A testable prediction of the facet view is that ethical disagreements about AI should cluster by facet, so a content analysis of public AI debates could count how often clashes are cross-facet rather than within a single facet.
  • The implementability claim can be probed directly: a working formalisation of virtue ethics in first-order logic for a small case (e.g., an AI tool used in medical triage) would validate the programme, while a demonstration that virtuous judgement resists first-order encoding would force the framework back to a purely conceptual status.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes a conceptual framework for the ethical assessment of AI use. It argues that ethical concerns about AI stem from a blurred understanding of the concept and that AI is best viewed through three facets: AI as a scientific endeavour (rooted in the Dartmouth workshop and its seven research axes), AI as a technology (yielding four groups of tools: logical reasoning and deduction, NLP, adaptive and evolutionary optimisation, and machine learning), and AI as a sociotechnical imaginary (shared, value-laden visions of AI's social consequences and the stakeholders who promote them). After reviewing virtue, duty, and consequence ethics, the paper proposes a three-layer responsibility framework (transparency and literacy; ethical values; regulations) and connects the three facets to Borgmann's three modes of information (about reality, for reality, as reality). It concludes that implementation of the framework 'should be possible and helpful to ensure that AI tools are used for the benefit of society', resting this on a promised first-order logical theory for virtue-driven ethical principles; an executable specification is stated to be under development.

Significance. Strengths worth crediting: the manuscript is internally consistent and well referenced; it offers a genuinely useful historical synthesis (the seven Dartmouth axes mapped to four technology groups is clear and informative); the three-layer responsibility framework is stated simply enough to be actionable as a proposal; and the paper is transparent about its own limits, acknowledging that the executable specification 'is under development.' The Borgmann correspondence is thought-provoking and could open useful discussion. Limitations: this is a position essay with no formal results; the three-facet decomposition, the one-to-one Borgmann correspondence, and the feasibility of a first-order logical theory for virtue verification are all asserted rather than demonstrated. The framework's adequacy is argued by construction, so it offers organizing value rather than a falsifiable test. If the promised logical theory and executable specification are delivered, the framework could become a practical assessment instrument; at present its value is primarily pedagogical and organizational.

major comments (5)
  1. [Section 5, final paragraph, and Section 6, final paragraph] The paper's central feasibility claim — that the 'actual implementation of the proposed framework... should be possible and helpful to ensure that AI tools are used for the benefit of society' — is stated to rest on 'a first-order logical theory' for verifying virtue-driven ethical principles. No signature, axioms, semantics, or proof sketch of this theory is given anywhere in the manuscript, and the paper itself acknowledges that 'an executable specification of the framework is under development.' Since the implementability claim is the paper's principal conclusion, it currently rests on a bare promissory note; the author should provide at least an informal sketch of the theory (vocabulary, intended semantics, representation of ethical principles, and what 'verification' would establish), or explicitly re-scope the conclusion as a research agenda.
  2. [Section 4 vs. Section 5, final paragraph] Section 4 asserts that 'values and value scales are individually and collectively shape shifting' and that 'the values ascribed to specific scenarios are also dynamic and unpredictable,' while the final paragraph of Section 5 proposes a first-order logical theory in which virtue-driven ethical principles are verified. A static first-order theory cannot represent the dynamics and unpredictability the paper emphasizes without additional temporal, modal, or nonmonotonic machinery, and if the theory is revised as values shift, verification becomes relative to a moving target. This tension is load-bearing for the implementability claim and is not addressed; the revised version should explain how the logical theory accommodates value dynamics, or explicitly limit the scope of what can be verified.
  3. [Section 6, Borgmann correspondence] The claimed one-to-one correspondence between the three AI facets and Borgmann's three modes of information, and the resulting claim that the sequence 'features increasing difficulty in exerting virtuous behaviour,' are asserted rather than argued. Because the framework's ordering of ethical difficulty follows directly from this correspondence, the mapping needs justification (or a caveat that it is a working hypothesis); otherwise the ordering — and with it the framework's practical guidance — is unsupported.
  4. [Abstract and Section 1] The thesis that ethical concerns 'stem from a blurred understanding of AI' is stated as a causal claim without considering alternative sources of ethical concern — for example the distributional harms and conflicting interests that the paper itself documents in Section 4 (misinformation, profit-driven distortion). As written, the exclusivity of the cause is not established; the claim should be qualified to position conceptual blur as one important source among several.
  5. [Section 5, three-layer framework] The proposed three-layer responsibility framework is not compared with existing AI ethics and governance structures (e.g., the risk-based tiers of the EU AI Act, the OECD AI Principles, or the global survey of AI ethics principles by Jobin, Ienca, and Vayena, 2019). Without such a comparison, the manuscript does not clearly delimit what the three-layer framework adds; the author should state the intended relationship (overlap, refinement, or alternative) to at least one existing framework.
minor comments (5)
  1. [Abstract, Sections 3 and 4] The causal diagnoses of ethical concern are stated in parallel without being unified: the Abstract says concerns 'stem from a blurred understanding,' Section 3 says they 'stem from poor assessment and inadequate utilisation of tools,' and Section 4 attributes them to 'misinformed interpretations' and 'purposeful distortion.' The author should state whether these are facets of a single diagnosis or three distinct claims, and how they relate to the three-facet structure.
  2. [Throughout] There are typographical artifacts (e.g., 'F APESP' in the footnote on page 1, 'd'ˆetre' in Section 2, and irregular spacing in the reference '[DL VRC+23]'); a careful proofreading pass is needed.
  3. [Section 4] The claim that misinformation about AI 'has been amplified by an oligopoly of organisations interested in profit making' is stronger than the cited literature (e.g., [GT24, IDZB24, MG24]) seems to warrant; a more precise characterization (e.g., 'a small number of large firms with commercial interests') would keep the claim aligned with its evidence base.
  4. [Section 5] The three ethical approaches are said to be 'complementary,' but the paper does not say whether the three-layer framework presupposes one approach (virtue ethics is named only as 'a possibility') or is intended to be approach-neutral; one sentence clarifying this would make the framework's domain of application clearer.
  5. [Section 6] Since the executable specification is presented as future work, state this scope explicitly in the introduction, so that the paper's contributions (analysis and framework proposal, not implementation) are set for the reader from the outset.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is a self-contained conceptual synthesis with no fitted inputs, no self-citations, and no prediction that reduces to its definitions.

full rationale

The paper is an essay that proposes a three-facet descriptive view of AI (scientific endeavour, technology, sociotechnical imaginary) and a three-layer ethical assessment framework (transparency and literacy, ethical values, regulations). There are no equations, no fitted parameters, and no data-driven predictions whose success could be forced by construction. The facets are introduced as historical and conceptual lenses, supported by external literature (Dartmouth proposal [MMRS55], Moor 2006, Jasanoff and Kim [JK09, JK13], Borgmann [Bor99]), none of which are self-citations of the author. The mapping to Borgmann's three modes of information is presented as a correspondence, not as a derivation that assumes its conclusion. The only formally load-bearing promissory element is the claim in Section 5 that 'a first-order logical theory can be formulated' for virtue-driven principles, and Section 6 states that an executable specification 'is under development.' This is an unsupported feasibility assertion and a correctness risk, but it is not circular: the theory is not used as an input to derive the framework, and no theorem is imported from prior authorial work. The framework's central conclusion ('should be possible and helpful') is an interpretive judgment based on the preceding descriptive review, not a prediction that collapses into its definitions. Accordingly, no concrete circular step can be exhibited, and the appropriate finding is no significant circularity.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The central claim relies on a tripartite classification of AI and a promissory logical theory. No free parameters are involved, but several domain assumptions are asserted without proof.

assumptions (4)
  • ad hoc to paper AI can be meaningfully decomposed into the three facets: scientific endeavour, technology, and sociotechnical imaginary.
    The entire framework rests on this tripartite division (Sections 1,2,3,4). It is presented as a description of the field, not derived from prior literature.
  • domain assumption The three fundamental approaches to ethics (virtue, duty, consequence) are exhaustive and complementary.
    Section 5 lists these three as standard; the framework aligns with them, but no argument is given that these are the only relevant ones.
  • ad hoc to paper A first-order logical theory can capture virtue-driven ethical principles sufficiently for verification.
    Section 5 last paragraph asserts this is possible; no proof or sketch is given.
  • ad hoc to paper Borgmann's three modes of information correspond to the three AI facets in a one-to-one manner.
    Section 6 draws the correspondence; this analogy is presented without argument.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Ethics through the Facets of Artificial Intelligence." pith.science (2026). https://pith.science/paper/Z46SG2JU

@misc{pith2026250717020,
  author       = {Pith},
  title        = {Pith review of: Ethics through the Facets of Artificial Intelligence},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z46SG2JU}},
  note         = {Machine review of arXiv:2507.17020}
}
read the original abstract

Artificial Intelligence (AI) has received unprecedented attention in recent years, raising ethical concerns about the development and use of AI technology. In the present article, we advocate that these concerns stem from a blurred understanding of AI, how it can be used, and how it has been interpreted in society. We explore the concept of AI based on three descriptive facets and consider ethical issues related to each facet. Finally, we propose a framework for the ethical assessment of the use of AI.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 50 canonical work pages

  1. [1]

    K. R. Apt et al. From logic programming to Prolog , volume 362. Prentice Hall London, 1997

    work page 1997

  2. [2]

    K. R. Apt. Logic programming. In Handbook of theoretical computer science (vol. B) formal models and semantics , pages 493--574. North Holland, 1991

    work page 1991

  3. [3]

    T Augustine

    M. T Augustine. A survey on universal approximation theorems. arXiv preprint arXiv:2407.12895 , 2024

    arXiv 2024

  4. [4]

    E. M. Bender, T. Gebru, A. McMillan-Major, and S. Shmitchell. On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM conference on fairness, accountability, and transparency , pages 610--623, 2021

    work page 2021

  5. [5]

    A. Blum, N. Haghtalab, A. D Procaccia, and M. Qiao. Collaborative pac learning. Advances in Neural Information Processing Systems , 30, 2017

    work page 2017

  6. [6]

    Borgmann

    A. Borgmann. Holding on to Reality: The Nature of Information at the Turn of the Millennium . University of Chicago press, 1999

    work page 1999

  7. [7]

    Cullina, A

    D. Cullina, A. N. Bhagoji, and P. Mittal. Pac-learning in the presence of adversaries. Advances in Neural Information Processing Systems , 31, 2018

    work page 2018

  8. [8]

    De La Vega, N

    N. De La Vega, N. Razin, N. Cohen, et al. What makes data suitable for a locally connected neural network? a necessary and sufficient condition based on quantum entanglement. Advances in Neural Information Processing Systems , 36:40994--41033, 2023

    work page 2023

Show all 52 references
  1. [9]

    A. Ervik. Generative ai and the collective imaginary: The technology-guided social imagination in ai-imagenesis. Image , 2023

    2023

  2. [10]

    Een and N

    N. Een and N. Sorensson. An extensible sat-solver. In International conference on theory and applications of satisfiability testing , pages 502--518. Springer, 2003

    2003

  3. [11]

    E Eiben and J

    A. E Eiben and J. E. Smith. Introduction to evolutionary computing . Springer, 2015

    2015

  4. [12]

    D. Fensel. Ontologies . Springer, 2001

    2001

  5. [13]

    Fagin, J

    R. Fagin, J. Y. Halpern, Y. Moses, and M. Vardi. Reasoning about knowledge . MIT press, 2004

    2004

  6. [14]

    Gavranovic

    B. Gavranovic. Fundamental components of deep learning : a category-theoretic approach . PhD thesis, University of Strathclyde, 2024

    2024

  7. [15]

    R. Gordon. Learning from fiction: applications in emerging technologies. Bulletin of Science, Technology & Society , 29(6):470--475, 2009

    2009

  8. [16]

    Gebru and E

    T. Gebru and E. P. Torres. The tescreal bundle: Eugenics and the promise of utopia through artificial general intelligence. First Monday , 2024

    2024

  9. [17]

    J. Y. Halpern and Y. Moses. Knowledge and common knowledge in a distributed environment. Journal of the ACM (JACM) , 37(3):549--587, 1990

    1990

  10. [18]

    J. Y. Halpern and Y. Moses. A guide to completeness and complexity for modal logics of knowledge and belief. Artificial intelligence , 54(3):319--379, 1992

    1992

  11. [19]

    N. Inie, S. Druga, P. Zukerman, and E. M. Bender. From ai to probabilistic automation: How does anthropomorphization of technical systems descriptions influence trust? In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency , pages 2322--2347, 2024

    2024

  12. [20]

    Isaacson

    W. Isaacson. Leonardo da vinci--the scientist. Substantia , 3(2):59--64, 2019

    2019

  13. [21]

    Jasanoff and S-H

    S. Jasanoff and S-H. Kim. Containing the atom: Sociotechnical imaginaries and nuclear power in the united states and south korea. Minerva , 47:119--146, 2009

    2009

  14. [22]

    Jasanoff and S-H

    S. Jasanoff and S-H. Kim. Sociotechnical imaginaries and national energy policies. Science as culture , 22(2):189--196, 2013

    2013

  15. [23]

    Johri, S

    P. Johri, S. K. Khatri, A. T. Al-Taani, M. Sabharwal, S. Suvanov, and A. Kumar. Natural language processing: History, evolution, application, and future work. In Proceedings of 3rd International Conference on Computing Informatics and Networks: ICCIN 2020 , pages 365--375. Spr...

    2020

  16. [24]

    Y. Jia, G. Peng, Z. Yang, and T. Chen. Category-theoretical and topos-theoretical frameworks in machine learning: A survey. arXiv preprint arXiv:2408.14014 , 2024

    2024 arXiv

  17. [25]

    Kazantzakis

    N. Kazantzakis. Report to Greco . Simon and Schuster, 2012

    2012

  18. [26]

    Krizhevsky, I

    A. Krizhevsky, I. Sutskever, and G. E Hinton. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems , 25, 2012

    2012

  19. [27]

    Vretos, and P

    Thomas Kopalidis, Vassilios Solachidis, N. Vretos, and P. Daras. Advances in facial expression recognition: A survey of methods, benchmarks, models, and datasets. Information , 15(3):135, 2024

    2024

  20. [28]

    Kotseruba and J

    I. Kotseruba and J. K. Tsotsos. 40 years of cognitive architectures: core cognitive abilities and practical applications. Artificial Intelligence Review , 53(1):17--94, 2020

    2020

  21. [29]

    J. E. Laird. The Soar cognitive architecture . MIT press, 2019

    2019

  22. [30]

    Lemke and D

    C. Lemke and D. Monett. Ai-based service systems: digital-ethical issues and their impact on value co-creation. In Handbook of Services and Artificial Intelligence , pages 228--249. Edward Elgar Publishing, 2024

    2024

  23. [31]

    Levine, D

    Y. Levine, D. Yakira, N. Cohen, and A. Shashua. Deep learning and quantum entanglement: fundamental connections with implications to network design 6th int. conf. on learning representations, iclr 2018 (vancouver, bc, canada, april 30--may 3, 2018) conf. In Track Proc.(OpenRev...

    2018

  24. [32]

    Monett and B

    D. Monett and B. Grigorescu. Deconstructing the ai myth: Fallacies and harms of algorithmification. In European Conference on e-Learning , volume 23, pages 242--248. Academic Conferences International Limited, 2024

    2024

  25. [33]

    McCarthy, M

    J. McCarthy, M. L. Minsky, N. Rochester, and C. E. Shannon. A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence . Aug. 1955

    1955

  26. [34]

    M. W. Moskewicz, C. F. Madigan, Y. Zhao, L. Zhang, and S. Malik. Chaff: Engineering an efficient sat solver. In Proceedings of the 38th annual Design Automation Conference , pages 530--535, 2001

    2001

  27. [35]

    J. Moor. The dartmouth college artificial intelligence conference: The next fifty years. AI magazine , 27(4):87--87, 2006

    2006

  28. [36]

    D. A. Norman. Design for a better world: Meaningful, sustainable, humanity centered . MIT Press, 2023

    2023

  29. [37]

    G. ORegan. Ethical and Legal Aspects of Computing . Springer, 2024

    2024

  30. [38]

    Prata, G

    M. Prata, G. Masi, L. Berti, V. Arrigoni, A. Coletta, I. Cannistraci, S. Vyetrenko, P. Velardi, and N. Bartolini. Lob-based deep learning models for stock price trend prediction: a benchmark study. Artificial Intelligence Review , 57(5):116, 2024

    2024

  31. [39]

    L. Poirier. Neat vs. scruffy: How early ai researchers classified epistemic cultures of knowledge representation. IEEE Annals of the History of Computing , 2024

    2024

  32. [40]

    Perconti and A

    P. Perconti and A. Plebe. Deep learning and cognitive science. Cognition , 203:104365, 2020

    2020

  33. [41]

    Paullada, I

    A. Paullada, I. D. Raji, E. M. Bender, E. Denton, and A. Hanna. Data and its (dis) contents: A survey of dataset development and use in machine learning research. Patterns , 2(11), 2021

    2021

  34. [42]

    Russell and L

    P. Russell and L. Buck. Humanity-centred design--defining the emerging paradigm in design education and practice. In VIA Design . The Design Society, 2020

    2020

  35. [43]

    Rubachev, N

    I. Rubachev, N. Kartashev, Y. Gorishniy, and A. Babenko. Tabred: Analyzing pitfalls and filling the gaps in tabular deep learning benchmarks. arXiv preprint arXiv:2406.19380 , 2024

    2024 arXiv

  36. [44]

    C. Taylor. On social imaginary. C. Taylor, Modern social imaginaries , pages 23--30, 2004

    2004

  37. [45]

    Uschold and M

    M. Uschold and M. Gruninger. Ontologies: Principles, methods and applications. The knowledge engineering review , 11(2):93--136, 1996

    1996

  38. [46]

    G Valiant

    L. G Valiant. A theory of the learnable. Communications of the ACM , 27(11):1134--1142, 1984

    1984

  39. [47]

    G Valiant

    L. G Valiant. Robust logics. In Proceedings of the thirty-first annual ACM symposium on Theory of Computing , pages 642--651, 1999

    1999

  40. [48]

    S. Vallor. The AI mirror: How to reclaim our humanity in an age of machine thinking . Oxford University Press, 2024

    2024

  41. [49]

    Van Harmelen, V

    F. Van Harmelen, V. Lifschitz, and B. Porter. Handbook of knowledge representation , volume 1. Elsevier, 2008

    2008

  42. [50]

    Vallor and W

    S. Vallor and W. J. Rewark. An introduction to data ethics. Course module.) Santa Clara, CA: Markkula Center for Applied Ethics , 2018

    2018

  43. [51]

    Y. Wan, Z. Bi, Y. He, J. Zhang, H. Zhang, Y. Sui, G. Xu, H. Jin, and P. Yu. Deep learning for code intelligence: Survey, benchmark and toolkit. ACM Computing Surveys , 56(12):1--41, 2024

    2024

  44. [52]

    Y. Wilks. The history of natural language processing and machine translation. Encyclopedia of language and linguistics , 9, 2005

    2005

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.