Pith. sign in

REVIEW 4 major objections 5 minor 139 references

A Case for Specialisation in Non-Human Entities

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The paper argues that non-human entities should be specialised, not general, and that well-specified specialisation—with 'specified governance' for hard-to-specify tasks—is the key to safe, secure, governable AI.

desk verdict A serious position paper with a genuine inversion, but the security argument trips over its own definition of generality. read the letter →

arxiv 2503.04742 v2 pith:6NXBOEPY submitted 2025-02-05 cs.CY cs.AI

El-Mahdi El-Mhamdi , Lê-Nguyên Hoang , Mariame Tighanimine This is my paper · ORCID
classification cs.CYcs.AI
keywords AIspecialisationgeneralityAGIsafetyspecificationmodularsystemsspecifiedgovernanceadversarialmachinelearning
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

This paper argues that the prevailing goal of building general-purpose AI is the wrong objective for safety, security, and governance. For algorithms, models, and organisations, the key is well-specified specialisation: an entity should offer only a small, documented set of tasks, and its behaviour should be specified precisely enough to audit, verify, and govern. The authors review the classic objections to specialisation—worker harm, economic integration incentives, and the statistical advantage of pooled estimation—and argue that each applies differently, or not at all, to non-human entities. They build a positive case from machine-learning vulnerability, modular engineering, economics, and the sociology of work, and they close by proposing 'specified governance' for tasks that are too complex or contested to specify directly. If the case holds, AI development and regulation should shift from model scale and task breadth toward narrowness, documentation, and verifiable contracts.

What carries the argument

The load-bearing object is 'well-specified specialisation'. An entity is specialised, on the paper's definition, when the number of tasks it offers to external users is small; specification is the precise statement of what the entity does, what it cannot safely do, and how it is used. This pairing does the work throughout: specialisation makes modules small enough to audit, verify, and give least-privilege access, while specification makes their behaviour predictable enough to be tested, proven, or governed. For tasks that resist specification—moderation, recommendation, translation—the paper invokes a second-order mechanism, 'specified governance', which specifies the procedure for choosing specifications rather than the specifications themselves. This mirrors the general pattern of modular decomposition behind the classical thesis that any general computation is a composition of specialised elementary operations.

What would settle it

A controlled comparison between two models of equal parameter count and equal data cleanliness, one serving a single well-defined API and one serving many heterogeneous APIs, would test the claim: the paper predicts the many-API model is more vulnerable to poisoning, privacy leakage, and jailbreaking. If vulnerability stays flat as API breadth grows while size and data are held fixed, the security case for specialisation loses its force.

Watch

Extended reading notes

Core claim

The paper's central claim is that non-human information-processing entities—algorithms, AI models, and human organisations—should be specialised rather than general, provided the specialisation is well specified. Generality is defined as a large number of externally available tasks, an API-count view, and the paper argues that generality is bought at the price of scale and data heterogeneity that make systems harder to audit, more vulnerable to poisoning, privacy attack, and jailbreaking, and more concentrated in power. It then argues that modularisation, least privilege, comparative advantage, and organic solidarity all favour narrow, documented components, and that where tasks are too complex or contested to specify directly, the answer is 'specified governance': specifying the process by which specifications are chosen, much as a constitution governs law-making. If this is right, the dominant design objective of general-purpose AGI is the wrong target for safety and governance.

Load-bearing premise

The argument assumes that a system offering many external tasks must be trained on a huge, heterogeneous, hard-to-clean corpus, so that the security costs of generality are unavoidable; a small general system built on clean narrow data would escape the paper's strongest argument.

Editorial extensions

If this is right

  • If the paper is right, the stated goal of artificial general intelligence should be replaced by portfolios of narrow, documented systems whose external interfaces are deliberately small.
  • Regulators could classify AI systems by the number and breadth of externally available tasks rather than by self-reported purpose, making rules like the EU AI Act more enforceable.
  • Software and organisational design would favour modular components with least-privilege access, so that a compromised component cannot silently take over the rest of the system.
  • Hard-to-specify tasks would not be left to unconstrained models; they would be governed by specified procedures for choosing specifications, comparable to a constitution for a system.
  • Verifiable computing and typed interfaces could move machine learning toward the safety practices of classical engineering, but only for properties that can be stated in advance.

Reading between the lines

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

  • The paper leaves implicit that its API-count definition of generality gives regulators a measurable lever: count the externally exposed endpoints of an AI agent and cap that count, rather than debating the meaning of 'general purpose'.
  • A direct test of the paper's security claim would hold model weights fixed and vary only the number of tools or API endpoints an agent may call; the paper predicts attack surface grows with endpoint count even without changing the underlying model.
  • The argument transfers to human organisations: certification and licensing, long used to bound what doctors and engineers may do, are a form of specification that the paper's logic would extend to AI modules and AI agents.
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

4 major / 5 minor

Summary. El-Mhamdi, Hoang, and Tighanimine argue that for non-human entities -- algorithms, AI systems, and human organisations -- well-specified specialisation is preferable to generality, and that the current research and policy orientation toward general-purpose AI (or AGI) is therefore misguided. The paper has three parts: it reviews and rebuts common arguments against specialisation drawn from human labour, economics, and statistics; it presents four positive arguments for specialisation, based on adversarial machine learning and security, complex system engineering, economics, and the sociology of work; and it argues for specification as the necessary complement to specialisation, introducing 'specified governance' for tasks that cannot be fully specified. The intended conclusion is that specialisation with explicit specification should be the design objective for non-human information-processing systems. The paper is a position or argumentative essay rather than an empirical or formal contribution.

Significance. If the central claim were established, the paper would have substantial implications for AI governance, regulation, and research priorities: it would provide a normative and technical rationale for preferring modular, narrowly-interfaced systems over frontier 'general' models, and for investing in specification and verifiable computing. The paper is genuinely interdisciplinary, engages with classical and recent literatures, and is candid about the limits of specification; it also usefully highlights the external-interface definition of generality and the observation that a specialised system can still be internally massive. Its contribution is, however, conditional: the key link between interface-level specialisation and the internal properties (parameter count, data heterogeneity) that drive the security argument is asserted rather than demonstrated. The paper does not provide a mathematical model, empirical evaluation, or machine-checked proof; its value lies in the coherence and breadth of the argument, and in the clarity with which it frames the policy question.

major comments (4)
  1. [Section 2, 'Defining Generality' and Section 4, 'Large Models are More Vulnerable'] The stress-test concern lands. The paper defines generality at the external interface: 'We will regard a system as general, if the number of tasks that external users can ask the system to perform is large' (Section 2), and it explicitly concedes that 'a very specialised system in this sense may nevertheless be extremely complex' and may be composed of trillions of parameters (citing the Persia recommender). The security argument in Section 4, however, is about internal properties: 'The more parameters, the more vulnerabilities' and 'the more generality, the more data heterogeneity, the more vulnerability.' The paper never shows that reducing the external API count reduces parameter count or training-data heterogeneity, nor that internal complexity can be modularly isolated so that parameter-count vulnerabilities do not transfer. As written, a trillion-parameter, web-scale-trained recommender with a single recommendation API is specialised by Section 2's definition yet carries every vulnerability Section 4 attributes to generality. The policy conclusion 'specialise the interface' therefore does not follow from the security premises without an additional argument, e.g., that interface specialisation induces internal and training-data specialisation, or that modular isolation bounds the relevant vulnerabilities.
  2. [Section 3, 'Arguments from Statistics'] The presentation of Stein's paradox overstates the theorem. The text says that for each subset, for any specialised estimator, there is a general estimator whose expected mean square error is lower no matter what the ground truth is; this is not what Stein's result establishes. The classical result is that for estimating a multivariate normal mean in dimension at least three, the James-Stein estimator dominates the maximum-likelihood estimator, not that it dominates every specialised estimator. Since this is the paper's main statistical argument in favour of generality, the rebuttal in the following paragraphs is aimed at a stronger claim than the literature supports. The passage should be corrected to state the theorem accurately and to indicate the class of estimators to which it applies.
  3. [Section 1 (scope) and Section 3, 'The Case of Humans'] The paper's abstract and introduction explicitly include 'human organisations' among the non-human entities under discussion, but the dismissal of the human-labour arguments against specialisation applies only to algorithms. The text argues that algorithms are not moral patients and therefore psychological and meaning-based objections do not apply; it does not address the fact that human organisations contain human workers who can suffer the very harms (boredom, disengagement, loss of meaning) catalogued earlier. The later 'company towns' discussion in Section 4 actually acknowledges that organisational concentration affects people, which makes the omission more conspicuous. The paper should either restrict the claim to purely algorithmic agents or extend the discussion to show why the human welfare arguments do not apply to specialised human organisations.
  4. [Section 5, 'Specifying Unspecifiable Tasks Through Specified Governance'] The proposed 'specified governance' is the crux of the third contribution, but it is developed only by analogy. The constitutional-democracy analogy is suggestive, yet the paper does not explain how 'specifying how to specify' would be implemented, verified, or made accountable for algorithms, nor does it give criteria for when a governance specification is legitimate or stable. Without this, the claim that specified governance can address hard-to-specify tasks is a research program rather than a conclusion. I would like to see either a more explicit statement that this is a proposal, or at least one worked example showing the governance specification for an algorithmic task and the mechanism by which it would constrain the system.
minor comments (5)
  1. [Section 3, 'Arguments from Economics'] There is a duplicated word in 'an optimized optimized system'; one instance should be removed.
  2. [Section 5, 'Hypertelia and the Pitfalls of Automated Task Specification'] The name 'Kolmogorov' is misspelled as 'Kolomogorov' both in the heading and in the body text.
  3. [Section 2, 'Defining Generality'] The sentence 'This definition of generality-by-purpose can be also found in Model Cards' has awkward word order; 'can also be found' would be clearer.
  4. [Section 5, 'Two Limits of Specifications'] The informal definition of a formalized specification as 'a program that, given any program, returns whether the program verifies the specification' presupposes decidability, which is not generally available; this should be clarified or qualified.
  5. [Section 4, 'Large Models are More Vulnerable'] The slogan 'the more generality, the more data heterogeneity, the more vulnerability' should be made explicit and qualified, since the cited results concern specific attack models and do not automatically support an unqualified monotone claim.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the paper is an argumentative review whose central claim does not reduce to its inputs; minor self-citation is used as supporting evidence, not as a forced derivation.

full rationale

The paper contains no equations, fitted parameters, or prediction-vs-fit structure, so the classic circularity mechanisms are absent. Its central claim, that well-specified specialisation is preferable for non-human entities, is advanced through argument and citation rather than derived from a definition. The main potential circularity would be if the Section 2 definition of generality (number of externally available tasks) were itself used to prove the Section 4 vulnerability claims; instead, Section 4 asserts as an empirical premise that generality is obtained by training large models on massive web data crawls, and supports the parameter-count and data-heterogeneity vulnerability claims with a mix of self-citations (El-Mhamdi et al. 2018, 2021, 2022; Hoang 2024) and independent references (Kattis and Nikolov 2017; Biggio et al. 2012; Suya et al. 2021; Oprea and Vassilev 2023). This is an internal-consistency weakness rather than circularity: the paper's own definition allows highly specialised systems with trillions of parameters, which weakens but does not invalidate the security argument as a matter of definition. Self-citations are numerous but not load-bearing in the sense of a uniqueness theorem or an ansatz smuggled in via citation; the same claims are corroborated by external peer-reviewed sources and the broader argument does not rest solely on the authors' prior work. Therefore no circular step is exhibited, and the score reflects only the presence of minor self-citation in a largely self-contained position paper.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

No free parameters or fitted values appear because the paper makes no quantitative derivation. The argument rests on five background assumptions: computability, moral status of machines, vulnerability scaling with scale and data heterogeneity, transfer of software-engineering modularity to ML, and the viability of governance-level specification. The last is the most paper-specific and the least independently established.

assumptions (5)
  • standard math The Church-Turing thesis: any computable function is a composition of elementary specialised operations.
    Invoked in Section 2 to frame generality as a composition of specialised components; a background result of computability theory that the paper relies on without proof.
  • domain assumption Non-human agents such as algorithms and organisations are not moral patients, so the psychological and physical harms of overspecialisation in human work do not apply to them.
    Section 3, 'The Case of Humans', cites Gibert and Martin (2022) to dismiss the human-welfare arguments against specialisation. If this moral-status premise fails, those arguments could re-enter the balance.
  • domain assumption The number of parameters in a learned model and the heterogeneity of its training data are the main drivers of its vulnerability to privacy breaches, poisoning, and jailbreaking.
    Section 4, 'Large Models are More Vulnerable', asserts this pattern and cites the authors' own prior work (El-Mhamdi et al. 2017, 2018, 2021; Hoang 2024) as support. It is load-bearing for the security case for specialisation.
  • domain assumption Principles of modularisation, abstraction, and least privilege from software engineering transfer to machine-learned components, so failures can be contained even when components are opaque.
    Section 4, 'Arguments from Complex System Engineering', applies Parnas (1972) and Saltzer and Schroeder (1975) to AI modules, assuming that the same decoupling and privilege-confinement properties hold.
  • ad hoc to paper Specified governance, specifying how to specify rather than specifying the task directly, can govern hard-to-specify systems, analogous to constitutional democracy.
    Section 5, 'Specifying Unspecifiable Tasks Through Specified Governance', introduces this as the paper's proposed solution to the two limits of specification. It is an analogy, not a demonstrated mechanism, and is specific to this paper's argument.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Case for Specialisation in Non-Human Entities." pith.science (2026). https://pith.science/paper/6NXBOEPY

@misc{pith2026250304742,
  author       = {Pith},
  title        = {Pith review of: A Case for Specialisation in Non-Human Entities},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6NXBOEPY}},
  note         = {Machine review of arXiv:2503.04742}
}
read the original abstract

With the rise of large multi-modal AI models, fuelled by recent interest in large language models (LLMs), the notion of artificial general intelligence (AGI) went from being restricted to a fringe community, to dominate mainstream large AI development programs. In contrast, in this paper, we make a case for specialisation, by reviewing the pitfalls of generality and stressing the industrial value of specialised systems. Our contribution is threefold. First, we review the most widely accepted arguments against specialisation, and discuss how their relevance in the context of human labour is actually an argument for specialisation in the case of non human agents, be they algorithms or human organisations. Second, we propose four arguments in favor of specialisation, ranging from machine learning robustness, to computer security, social sciences and cultural evolution. Third, we finally make a case for specification, discuss how the machine learning approach to AI has so far failed to catch up with good practices from safety-engineering and formal verification of software, and discuss how some emerging good practices in machine learning help reduce this gap. In particular, we justify the need for specified governance for hard-to-specify systems.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

139 extracted references · 69 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    B.; Kuppan, K.; and Divya, B

    Acharya, D. B.; Kuppan, K.; and Divya, B. 2025. Agentic AI: Autonomous Intelligence for Complex Goals--A Comprehensive Survey. IEEE Access

    2025

  4. [4]

    Ahmad, H.; Wang, L.; Hong, H.; Li, J.; Dawood, H.; Ahmed, M.; and Yang, Y. 2018. Primitives towards verifiable computation: a survey. Frontiers of Computer Science, 12: 451--478

    2018

  5. [5]

    American Medical Association . 1871. Code of Ethics of the American Medical Association Adopted May, 1847. Turner, Hamilton

  6. [6]

    J.; et al

    Anil, C.; Durmus, E.; Rimsky, N.; Sharma, M.; Benton, J.; Kundu, S.; Batson, J.; Tong, M.; Mu, J.; Ford, D. J.; et al. 2024. Many-shot jailbreaking. In The Thirty-eighth Annual Conference on Neural Information Processing Systems

    2024

  7. [7]

    Arrow, K. J. 1975. Vertical integration and communication. The Bell Journal of Economics, 173--183

    1975

  8. [8]

    Autor, D. 2014. Polanyi's paradox and the shape of employment growth. Technical report, National Bureau of Economic Research

    2014

Show all 139 references
  1. [9]

    Baack, S. 2024. A Critical Analysis of the Largest Source for Generative AI Training Data: Common Crawl. In The 2024 ACM Conference on Fairness, Accountability, and Transparency, 2199--2208

    2024

  2. [10]

    everything app

    Belanger, A. 2024. Elon Musk’s improbable path to making X an “everything app”. Ars Technica

    2024

  3. [11]

    Bellamy, R. 2017. The rule of law and the separation of powers. Routledge

    2017

  4. [12]

    Ben - Porat, O.; Mansour, Y.; Moshkovitz, M.; and Taitler, B. 2024. Principal-Agent Reward Shaping in MDPs. In Wooldridge, M. J.; Dy, J. G.; and Natarajan, S., eds., Thirty-Eighth AAAI Conference on Artificial Intelligence, AAAI 2024, Thirty-Sixth Conference on Innovative Appl...

    2024

  5. [13]

    Benkler, Y. 2006. The Wealth of Networks: How Social Production Transforms Markets and Freedom. Yale University Press

    2006

  6. [14]

    Bianchi, T. 2024. Google: annual advertising revenue 2001-2023. Statista

    2024

  7. [15]

    Biggio, B.; Nelson, B.; and Laskov, P. 2012. Poisoning Attacks against Support Vector Machines. In Proceedings of the 29th International Conference on Machine Learning, ICML 2012, Edinburgh, Scotland, UK, June 26 - July 1, 2012 . icml.cc / Omnipress

    2012

  8. [16]

    A.; Metcalf, J.; Murai, F.; Salvaggio, E.; Smart, A

    Blili-Hamelin, B.; Graziul, C.; Hancox-Li, L.; Hazan, H.; El-Mhamdi, E.-M.; Ghosh, A.; Heller, K. A.; Metcalf, J.; Murai, F.; Salvaggio, E.; Smart, A. J.; Snider, T.; Tighanimine, M.; Ringer, T.; Mitchell, M.; and Dori-Hacohen, S. 2025. Position: Stop treating ` AGI ' as the n...

    2025

  9. [17]

    Bookwalter Drury, H. 1918. Scientific management: A history and criticism. Columbia University Press

    1918

  10. [18]

    M.; Bottoni, P.; and Pareschi, R

    Borghoff, U. M.; Bottoni, P.; and Pareschi, R. 2025. Human-artificial interaction in the age of agentic AI: a system-theoretical approach. Frontiers in Human Dynamics, 7: 1579166

    2025

  11. [19]

    Boullier, D.; and El Mhamdi, E. M. 2020. Le machine learning et les sciences sociales \`a l’ \'e preuve des \'e chelles de complexit \'e algorithmique. Revue d’anthropologie des connaissances, 14(14-1)

    2020

  12. [20]

    Bradford, A. 2020. The Brussels Effect: How the European Union Rules the World. Oxford University Press

    2020

  13. [21]

    The Brussels Effect

    Bradford, A. 2024. “The Brussels Effect” and European Sovereignty. In European Sovereignty: The Legal Dimension--A Union in Control of its own Destiny, 191--200. Springer

    2024

  14. [22]

    Brennan, T. J. 2025. US government antitrust Google and Facebook cases: three neglected questions. In Research Handbook On Competition And Technology, 212--231. Edward Elgar Publishing

    2025

  15. [23]

    F.; and Levin, J

    Bresnahan, T. F.; and Levin, J. D. 2012. Vertical integration and market structure. National Bureau of Economic Research Cambridge (MA)

    2012

  16. [24]

    Brown, T. B.; Mann, B.; Ryder, N.; Subbiah, M.; Kaplan, J.; Dhariwal, P.; Neelakantan, A.; Shyam, P.; Sastry, G.; Askell, A.; Agarwal, S.; Herbert - Voss, A.; Krueger, G.; Henighan, T.; Child, R.; Ramesh, A.; Ziegler, D. M.; Wu, J.; Winter, C.; Hesse, C.; Chen, M.; Sigler, E.;...

    2020

  17. [25]

    Brunner Von Wattenwyl, K. 1874. Ueber die Hypertelie in der Natur. Wien

  18. [26]

    Brynjolfsson, E.; and Mitchell, T. 2017. What can machine learning do? Workforce implications. Science, 358(6370): 1530--1534

    2017

  19. [27]

    Cardelli, L. 1996. Type systems. ACM Computing Surveys (CSUR), 28(1): 263--264

    1996

  20. [28]

    Carlton, D. W. 1979. Vertical integration in competitive markets under uncertainty. The Journal of Industrial Economics, 189--209

    1979

  21. [29]

    Casilli, A. A. 2020. From the virtual class to the click workers: the transformation of work into service in the era of digital platforms. MATRIZes, 14(1): 13--21

    2020

  22. [30]

    Church, A. 1936. A note on the Entscheidungsproblem. The journal of symbolic logic, 1(1): 40--41

    1936

  23. [31]

    Coase, R. H. 1937. The Nature of the Firm. Economica

    1937

  24. [32]

    Cohen, S.; Bitton, R.; and Nassi, B. 2024. Here Comes The AI Worm: Unleashing Zero-click Worms that Target GenAI-Powered Applications. CoRR, abs/2403.02817

    2024 arXiv

  25. [33]

    Davies, T.; and Georgieva, Z. 2024. Google AdTech: Break Up or Break Out? Utrecht Law Journal Special Issue on Modern Bigness

    2024

  26. [34]

    Diaz, F.; and Madaio, M. 2024. Scaling laws do not scale. In Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, volume 7, 341--357

    2024

  27. [35]

    u rgens, U.; and Nialsch, T. 1985. From

    Dohse, K.; J \"u rgens, U.; and Nialsch, T. 1985. From" Fordism" to" Toyotism"? The social organization of the labor process in the Japanese automobile industry. Politics & Society, 14(2): 115--146

    1985

  28. [36]

    F.; and Wu, D

    Dou, Y.; Niculescu, M. F.; and Wu, D. 2013. Engineering optimal network effects via social media features and seeding in markets for digital goods and services. Information Systems Research, 24(1): 164--185

    2013

  29. [37]

    Dubey, A.; Jauhri, A.; Pandey, A.; Kadian, A.; Al - Dahle, A.; Letman, A.; Mathur, A.; Schelten, A.; Yang, A.; Fan, A.; Goyal, A.; Hartshorn, A.; Yang, A.; Mitra, A.; Sravankumar, A.; Korenev, A.; Hinsvark, A.; Rao, A.; Zhang, A.; Rodriguez, A.; Gregerson, A.; Spataru, A.; Roz...

    2024 arXiv

  30. [38]

    Durkheim, \'E . 1893. De la division du travail. FB Editions

  31. [39]

    Dwork, C.; McSherry, F.; Nissim, K.; and Smith, A. D. 2006. Calibrating Noise to Sensitivity in Private Data Analysis. In Theory of Cryptography, Third Theory of Cryptography Conference, TCC

    2006

  32. [40]

    Economides, N. 2001. The Microsoft antitrust case. Journal of Industry, Competition and Trade, 1: 7--39

    2001

  33. [41]

    Efron, B.; and Morris, C. 1977. Stein's paradox in statistics. Scientific American, 236(5): 119--127

    1977

  34. [42]

    D.; Riedl, J

    Ekstrand, M. D.; Riedl, J. T.; Konstan, J. A.; et al. 2011. Collaborative filtering recommender systems. Foundations and Trends in Human--Computer Interaction , 4(2): 81--173

    2011

  35. [43]

    M.; Farhadkhani, S.; Guerraoui, R.; Guirguis, A.; Hoang, L.-N.; and Rouault, S

    El-Mhamdi, E. M.; Farhadkhani, S.; Guerraoui, R.; Guirguis, A.; Hoang, L.-N.; and Rouault, S. 2021. Collaborative learning in the jungle (decentralized, byzantine, heterogeneous, asynchronous and nonconvex learning). Advances in neural information processing systems, 34: 25044--25057

    2021

  36. [44]

    El-Mhamdi, E.-M.; Farhadkhani, S.; Guerraoui, R.; Gupta, N.; Hoang, L.-N.; Pinot, R.; Rouault, S.; and Stephan, J. 2022. On the impossible safety of large AI models. arXiv preprint arXiv:2209.15259

    2022 arXiv

  37. [45]

    El-Mhamdi, E.-M.; Guerraoui, R.; Rouault, S.; et al. 2017. On The Robustness of a Neural Network. In 2017 IEEE 36th Symposium on Reliable Distributed Systems (SRDS), 84--93. IEEE

    2017

  38. [46]

    El-Mhamdi, E.-M.; Guerraoui, R.; Rouault, S.; et al. 2018. The Hidden Vulnerability of Distributed Learning in Byzantium. In Proceedings of the 35th International Conference on Machine Learning (ICML)

    2018

  39. [47]

    El-Mhamdi, E.-M.; and Hoang, L.-N. 2024. On Goodhart's law, with an application to value alignment. arXiv preprint arXiv:2410.09638

    2024 arXiv

  40. [48]

    Evans, M.; and Zimmermann, A. 2014. The global relevance of subsidiarity: An overview. Global perspectives on subsidiarity, 1--7

    2014

  41. [49]

    Fan, Y.; Ma, K.; Zhang, L.; Lei, X.; Xu, G.; and Tan, G. 2024 a . ValidCNN: A large-scale CNN predictive integrity verification scheme based on zk-SNARK. IEEE Transactions on Dependable and Secure Computing

    2024

  42. [50]

    Fan, Y.; Ma, K.; Zhang, L.; Liu, J.; Xiong, N.; and Yu, S. 2024 b . VeriCNN: Integrity verification of large-scale CNN training process based on zk-SNARK. Expert Systems with Applications, 124531

    2024

  43. [51]

    N.; and Villemaud, O

    Farhadkhani, S.; Guerraoui, R.; Hoang, L. N.; and Villemaud, O. 2022. An Equivalence Between Data Poisoning and Byzantine Gradient Attacks. In International Conference on Machine Learning, ICML

    2022

  44. [52]

    Fetz, A.; and Filippini, M. 2010. Economies of vertical integration in the Swiss electricity sector. Energy economics, 32(6): 1325--1330

    2010

  45. [53]

    S.; Sinnott-Armstrong, W.; Dickerson, J

    Freedman, R.; Borg, J. S.; Sinnott-Armstrong, W.; Dickerson, J. P.; and Conitzer, V. 2020. Adapting a kidney exchange algorithm to align with human values. Artificial Intelligence, 283: 103261

    2020

  46. [54]

    Friedman, G. 1961. The anatomy of work: labor, leisure and the implications of automation. Transaction Publishers

    1961

  47. [55]

    Friedmann, G. 1955. Le Travail en miettes. Esprit (1940-), 1725--1747

    1955

  48. [56]

    G \"a her, L.; Sammler, M.; Jung, R.; Krebbers, R.; and Dreyer, D. 2024. Refinedrust: A type system for high-assurance verification of Rust programs. Proceedings of the ACM on Programming Languages, 8(PLDI): 1115--1139

    2024

  49. [57]

    Garg, S.; and Srinivasan, A. 2022. Two-round multiparty secure computation from minimal assumptions. Journal of the ACM, 69(5): 1--30

    2022

  50. [58]

    Garner, J. 1992. The company town: architecture and society in the early industrial age. Oxford University Press

    1992

  51. [59]

    W.; Wallach, H.; Iii, H

    Gebru, T.; Morgenstern, J.; Vecchione, B.; Vaughan, J. W.; Wallach, H.; Iii, H. D.; and Crawford, K. 2021. Datasheets for datasets. Communications of the ACM, 64(12): 86--92

    2021

  52. [60]

    Gibert, M.; and Martin, D. 2022. In search of the moral status of AI: why sentience is a strong argument. AI & SOCIETY, 37(1): 319--330

    2022

  53. [61]

    J.; and Hart, O

    Grossman, S. J.; and Hart, O. D. 1986. The costs and benefits of ownership: A theory of vertical and lateral integration. Journal of political economy, 94(4): 691--719

    1986

  54. [62]

    Gunst, S.; and De Ville, F. 2021. The Brussels effect: how the GDPR conquered Silicon Valley. European Foreign Affairs Review, 26(3)

    2021

  55. [63]

    Guo, X.; Yu, F.; Zhang, H.; Qin, L.; and Hu, B. 2024. COLD-Attack: Jailbreaking LLMs with Stealthiness and Controllability. In Forty-first International Conference on Machine Learning, ICML 2024, Vienna, Austria, July 21-27, 2024 . OpenReview.net

    2024

  56. [64]

    Hackman, J. R. 1969. NATURE OF THE TASK AS A DETERMINER OF JOB BEHAVIOR. Personnel Psychology, 22(4)

    1969

  57. [65]

    Hoang, L.-N. 2024. The poison of dimensionality. arXiv preprint arXiv:2409.17328

    2024 arXiv

  58. [66]

    N.; Beylerian, R.; Colbois, B.; Fageot, J.; Faucon, L.; Jungo, A.; Noac'h, A

    Hoang, L. N.; Beylerian, R.; Colbois, B.; Fageot, J.; Faucon, L.; Jungo, A.; Noac'h, A. L.; Matissart, A.; and Villemaud, O. 2024. Solidago: A Modular Collaborative Scoring Pipeline. CoRR, abs/2211.01179

    2024 arXiv

  59. [67]

    N.; and El Mhamdi, E

    Hoang, L. N.; and El Mhamdi, E. M. 2019. Le fabuleux chantier: Rendre l’intelligence artificielle robustement b \'e n \'e fique . edp Sciences

    2019

  60. [68]

    Hoang, L.-N.; Faucon, L.; Jungo, A.; Volodin, S.; Papuc, D.; Liossatos, O.; Crulis, B.; Tighanimine, M.; Constantin, I.; Kucherenko, A.; et al. 2021. Tournesol: A quest for a large, secure and trustworthy database of reliable human judgments. arXiv preprint arXiv:2107.07334

    2021 arXiv

  61. [69]

    Hou, X.; Zhao, Y.; Wang, S.; and Wang, H. 2025. Model context protocol (mcp): Landscape, security threats, and future research directions. arXiv preprint arXiv:2503.23278

    2025 arXiv

  62. [70]

    Hughes, E. C. 1951. Mistakes at work. Canadian Journal of Economics and Political Science/Revue canadienne de economiques et science politique, 17(3): 320--327

    1951

  63. [71]

    Hughes, E. C. 1956. Social role and the division of labor. The Midwest Sociologist, 18(2): 3--7

    1956

  64. [72]

    Hutter, M. 2003. A gentle introduction to the universal algorithmic agent AIXI

    2003

  65. [73]

    Jones, R. 1993. Heckscher-Ohlin trade theory: Harry Flam and M. June Flanders, eds., (The MIT Press, Cambridge, MA, 1991) pp. x + 222. Journal of International Economics, 35(1-2): 197--199

    1993

  66. [74]

    Kakabadse, A.; and Kakabadse, N. 2005. Outsourcing: current and future trends. Thunderbird international business review, 47(2): 183--204

    2005

  67. [76]

    B.; Chess, B.; Child, R.; Gray, S.; Radford, A.; Wu, J.; and Amodei, D

    Kaplan, J.; McCandlish, S.; Henighan, T.; Brown, T. B.; Chess, B.; Child, R.; Gray, S.; Radford, A.; Wu, J.; and Amodei, D. 2020 b . Scaling Laws for Neural Language Models. CoRR, abs/2001.08361

    2020 arXiv

  68. [77]

    Kattis, A.; and Nikolov, A. 2017. Lower Bounds for Differential Privacy from Gaussian Width. In 33rd International Symposium on Computational Geometry, SoCG 2017, July 4-7, 2017, Brisbane, Australia, volume 77 of LIPIcs, 45:1--45:16. Schloss Dagstuhl - Leibniz-Zentrum f \" u r...

    2017

  69. [78]

    Kierans, A.; Ghosh, A.; Hazan, H.; and Dori - Hacohen, S. 2025. Quantifying Misalignment Between Agents: Towards a Sociotechnical Understanding of Alignment. In Walsh, T.; Shah, J.; and Kolter, Z., eds., AAAI-25, Sponsored by the Association for the Advancement of Artificial I...

    2025

  70. [79]

    Kifer, D.; and Machanavajjhala, A. 2011. No free lunch in data privacy. In Proceedings of the ACM SIGMOD International Conference on Management of Data, SIGMOD 2011, Athens, Greece, June 12-16, 2011 , 193--204. ACM

    2011

  71. [80]

    Kleppmann, M.; Frazee, P.; Gold, J.; Graber, J.; Holmgren, D.; Ivy, D.; Johnson, J.; Newbold, B.; and Volpert, J. 2024. Bluesky and the at protocol: Usable decentralized social media. In Proceedings of the ACM Conext-2024 Workshop on the Decentralization of the Internet, 1--7

    2024

  72. [81]

    Kolmogorov, A. N. 1963. On tables of random numbers. Sankhy \=a : The Indian Journal of Statistics, Series A , 369--376

    1963

  73. [82]

    M.; and Posner, R

    Landes, W. M.; and Posner, R. A. 1997. Market power in antitrust cases. J. Reprints Antitrust L. & Econ., 27: 493

    1997

  74. [83]

    Laux, J.; Wachter, S.; and Mittelstadt, B. 2024. Trustworthy artificial intelligence and the European Union AI act: On the conflation of trustworthiness and acceptability of risk. Regulation & Governance, 18(1): 3--32

    2024

  75. [84]

    K.; Kusbit, D.; Kahng, A.; Kim, J

    Lee, M. K.; Kusbit, D.; Kahng, A.; Kim, J. T.; Yuan, X.; Chan, A.; See, D.; Noothigattu, R.; Lee, S.; Psomas, A.; et al. 2019. WeBuildAI: Participatory framework for algorithmic governance. Proceedings of the ACM on human-computer interaction, 3(CSCW): 1--35

    2019

  76. [85]

    Li, Z.; and Agarwal, A. 2017. Platform integration and demand spillovers in complementary markets: Evidence from Facebook’s integration of Instagram. Management Science, 63(10): 3438--3458

    2017

  77. [86]

    Lian, X.; Yuan, B.; Zhu, X.; Wang, Y.; He, Y.; Wu, H.; Sun, L.; Lyu, H.; Liu, C.; Dong, X.; Liao, Y.; Luo, M.; Zhang, C.; Xie, J.; Li, H.; Chen, L.; Huang, R.; Lin, J.; Shu, C.; Qiu, X.; Liu, Z.; Kong, D.; Yuan, L.; Yu, H.; Yang, S.; Zhang, C.; and Liu, J. 2022. Persia: An Ope...

    2022

  78. [87]

    Lieberman, M. B. 1991. Determinants of vertical integration: An empirical test. In Academy of Management Proceedings, 31--35. Academy of Management Briarcliff Manor, NY 10510

    1991

  79. [88]

    Littler, C. R. 1978. Understanding taylorism. British Journal of Sociology, 185--202

    1978

  80. [89]

    Locke, J. 1690. Two Treatises of Government. Awnsham Churchill

  81. [90]

    Loukidou, L.; Loan-Clarke, J.; and Daniels, K. 2009. Boredom in the workplace: More than monotonous tasks. International Journal of Management Reviews

    2009

  82. [91]

    Majka, A.; and El-Mhamdi, E.-M. 2025. The Strong, Weak and Benign Goodhart's law. An independence-free and paradigm-agnostic formalisation. arXiv preprint arXiv:2505.23445

    2025 arXiv

  83. [92]

    Marx, K. 1847. Mis \`e re de la philosophie . C.G. Vogler, Bruxelles

  84. [93]

    Marx, K. 1867. Das Kapital. Hamburg: Otto Meissner, 1867; New-York: L. W. Schmidt

  85. [94]

    D.; and Klock, F

    Matsakis, N. D.; and Klock, F. S. 2014. The rust language. ACM SIGAda Ada Letters, 34(3): 103--104

    2014

  86. [95]

    Maxwell, T. 2024. Leaked Documents Show OpenAI Has a Very Clear Definition of ‘AGI’. Gizmodo

    2024

  87. [96]

    D.; Ruderman, M

    McCauley, C. D.; Ruderman, M. N.; Ohlott, P. J.; and Morrow, J. E. 1994. Assessing the developmental components of managerial jobs. Journal of applied psychology, 79(4): 544

    1994

  88. [97]

    2004 (initial version 1953)

    Ministère de la santé . 2004 (initial version 1953) . Article R4127-70. Code de la santé publique

    2004

  89. [98]

    D.; and Gebru, T

    Mitchell, M.; Wu, S.; Zaldivar, A.; Barnes, P.; Vasserman, L.; Hutchinson, B.; Spitzer, E.; Raji, I. D.; and Gebru, T. 2019. Model cards for model reporting. In Proceedings of the conference on fairness, accountability, and transparency, 220--229

    2019

  90. [99]

    Montesquieu. 1748. De l'esprit des lois. Barrillot & Fils

  91. [100]

    Munir, S.; Kollnig, K.; Shuba, A.; and Shafiq, Z. 2024. Google's Chrome Antitrust Paradox. arXiv preprint arXiv:2406.11856

    2024 arXiv

  92. [101]

    Murugesan, S. 2025. The Rise of Agentic AI: Implications, Concerns, and the Path Forward. IEEE Intelligent Systems, 40(2): 8--14

    2025

  93. [102]

    patronage

    Noiriel, G. 1988. Du" patronage" au" paternalisme": la restructuration des formes de domination de la main-d' uvre ouvri \`e re dans l'industrie m \'e tallurgique fran c aise. Le mouvement social, 17--35

    1988

  94. [103]

    Noothigattu, R.; Gaikwad, S. N. S.; Awad, E.; Dsouza, S.; Rahwan, I.; Ravikumar, P.; and Procaccia, A. D. 2018. A Voting-Based System for Ethical Decision Making. In Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence, (AAAI-18), the 30th innovative App...

    2018

  95. [104]

    Oprea, A.; and Vassilev, A. 2023. Adversarial machine learning: A taxonomy and terminology of attacks and mitigations. Technical report, National Institute of Standards and Technology

    2023

  96. [105]

    \"O zdemir, G.; and Tuna, G. 2024. Privilege Escalation: Threats, Prevention, and a Case Study. In Cases on Forensic and Criminological Science for Criminal Detection and Avoidance, 151--187. IGI Global

    2024

  97. [106]

    Parnas, D. L. 1972. On the criteria to be used in decomposing systems into modules. Communications of the ACM, 15(12): 1053--1058

    1972

  98. [107]

    Penard, W.; and Van Werkhoven, T. 2008. On the secure hash algorithm family. Cryptography in context, 1--18

    2008

  99. [108]

    Perrigo, B. 2023. Exclusive: OpenAI Lobbied the EU to Water Down AI Regulation. Time

    2023

  100. [109]

    Polanyi, M. 2009. The tacit dimension. In Knowledge in organisations, 135--146. Routledge

    2009

  101. [110]

    Radosevich, B.; and Halloran, J. 2025. MCP Safety Audit: LLMs with the Model Context Protocol Allow Major Security Exploits. arXiv preprint arXiv:2504.03767

    2025 arXiv

  102. [111]

    Ray, P. P. 2025. A survey on model context protocol: Architecture, state-of-the-art, challenges and future directions. Authorea Preprints

    2025

  103. [112]

    Ricardo, D. 1817. On the Principles of Political Economy and Taxation. McMaster University Archive for the History of Economic Thought, 3 edition

  104. [113]

    Righes, L.; Saeed, M.; Demartini, G.; and Papotti, P. 2023. The Community Notes Observatory: Can Crowdsourced Fact-Checking be Trusted in Practice? In Companion Proceedings of the ACM Web Conference 2023, 172--175

    2023

  105. [114]

    H.; and Schroeder, M

    Saltzer, J. H.; and Schroeder, M. D. 1975. The protection of information in computer systems. Proceedings of the IEEE, 63(9): 1278--1308

    1975

  106. [115]

    Samuelson, P. A. 1948. International trade and the equalisation of factor prices. The Economic Journal, 58(230): 163--184

    1948

  107. [116]

    Y.; and Tiku, N

    Schaul, K.; Chen, S. Y.; and Tiku, N. 2023. Inside the secret list of websites that make AI like ChatGPT sound smart. Washington Post, 19

    2023

  108. [117]

    Shapiro, C. 1999. Information rules: A strategic guide to the network economy. Harvard Business School Press

    1999

  109. [118]

    Shaw, M. 1989. Larger scale systems require higher-level abstractions. In Proceedings of the 5th International Workshop on Software Specification and Design, IWSSD 1989, Pittsburgh, Pennsylvania, USA, 1989 , 143--146. ACM

    1989

  110. [119]

    Siegmann, C.; and Anderljung, M. 2022. The Brussels effect and artificial intelligence: How EU regulation will impact the global AI market. arXiv preprint arXiv:2208.12645

    2022 arXiv

  111. [120]

    Silvers, R.; and Alperovitch, D. 2024. Review of the Summer 2023 Microsoft Exchange Online Intrusion. Technical report, Cybersecurity and Infrastructure Security Agency

    2024

  112. [121]

    Simondon, G. 1958. Du mode d'existence des objets techniques. Éditions Aubier-Montaigne

    1958

  113. [122]

    Simondon, G.; Mellamphy, N.; and Hart, J. 1980. On the mode of existence of technical objects. University of Western Ontario London

    1980

  114. [123]

    Skendzic, A.; and Kovacic, B. 2012. Microsoft office 365-cloud in business environment. In 2012 Proceedings of the 35th International Convention MIPRO, 1434--1439. IEEE

    2012

  115. [124]

    Small, C.; Bjorkegren, M.; Erkkil \"a , T.; Shaw, L.; and Megill, C. 2021. Polis: Scaling deliberation by mapping high dimensional opinion spaces. Recerca: revista de pensament i an \`a lisi , 26(2)

    2021

  116. [125]

    Smith, A. 1776. An Inquiry into the Nature and Causes of the Wealth of Nations. McMaster University Archive for the History of Economic Thought

  117. [126]

    Solomonoff, R. J. 1960. A preliminary report on a general theory of inductive inference

    1960

  118. [127]

    Suya, F.; Mahloujifar, S.; Suri, A.; Evans, D.; and Tian, Y. 2021. Model-Targeted Poisoning Attacks with Provable Convergence. In Proceedings of the 38th International Conference on Machine Learning, ICML 2021, 18-24 July 2021, Virtual Event , volume 139 of Proceedings of Mach...

    2021

  119. [128]

    Tarr, P.; Ossher, H.; Harrison, W.; and Sutton Jr, S. M. 1999. N degrees of separation: Multi-dimensional separation of concerns. In Proceedings of the 21st international conference on Software engineering, 107--119

    1999

  120. [129]

    Thaler, J.; et al. 2022. Proofs, arguments, and zero-knowledge. Foundations and Trends in Privacy and Security , 4(2--4): 117--660

    2022

  121. [130]

    Tighanimine, M. 2025. Le travail journalistique entre priorisation de l’information et r \'e action aux algorithmes de recommandation des r \'e seaux sociaux. Socio. La nouvelle revue des sciences sociales, (20): 127--158

    2025

  122. [131]

    Treisman, D. 2007. The Architecture of Government: Rethinking Political Decentralization. Cambridge University Press

    2007

  123. [132]

    Turing, A. M. 1936. On computable numbers, with an application to the Entscheidungsproblem. J. of Math, 58(345-363): 5

    1936

  124. [133]

    Ungarino, R. 2020. Here are 9 fascinating facts to know about BlackRock, the world's largest asset manager. Business Insider

    2020

  125. [134]

    S.; Hutter, M.; and Silver, D

    Veness, J.; Ng, K. S.; Hutter, M.; and Silver, D. 2010. Reinforcement learning via AIXI approximation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 24, 605--611

    2010

  126. [135]

    Vergara, R. A. G.; and Del Rosario, R. 2012. Samsung Electronics and Apple, Inc.: A Study in Contrast in Vertical Integration in the 21 st Century. American International Journal of Contemporary Research, 2(9): 77--81

    2012

  127. [136]

    Watson, D. 2019. Fordism: A review essay. Labor History, 60(2): 144--159

    2019

  128. [137]

    Williams, A.; Miceli, M.; and Gebru, T. 2022. The exploited labor behind artificial intelligence. Noema Magazine, 22

    2022

  129. [138]

    W \"o rsd \"o rfer, M. 2024. Apple’s antitrust paradox. European Competition Journal, 20(1): 113--146

    2024

  130. [139]

    Yao, S.; Zhao, J.; Yu, D.; Du, N.; Shafran, I.; Narasimhan, K.; and Cao, Y. 2023. React: Synergizing reasoning and acting in language models. In International Conference on Learning Representations (ICLR)

    2023

  131. [140]

    Zhu, T.; Xiong, P.; Li, G.; and Zhou, W. 2015. Correlated Differential Privacy: Hiding Information in Non-IID Data Set. IEEE Trans. Inf. Forensics Secur. , 10(2): 229--242

    2015

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.