Pith. sign in

REVIEW 5 major objections 6 minor 24 references

Multiverse Privacy Theory for Contextual Risks in Complex User-AI Interactions

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Multiverse Privacy Theory scores many futures and picks a privacy action by expected utility.

desk verdict A clearly written position paper whose simulation-based 'evidence' is circular: the correlations just recover the weights baked into the utility function. read the letter →

arxiv 2506.10042 v1 pith:2ZH55SQA submitted 2025-06-11 cs.CR cs.HC

Ece Gumusel This is my paper · ORCID
classification cs.CRcs.HC
keywords multiverseprivacytheorycontextualintegrityutilityexpecteddecision-makingusertrustriskAI
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 proposes Multiverse Privacy Theory, a framework in which every privacy decision branches into many possible futures—parallel universes—each shaped by the user's action and context. It claims that an AI system can estimate each future's probability, score it with a weighted utility built from privacy preference, security, risk, trust, demographics, and contextual integrity, and then choose the action with the highest expected utility. The payoff is a privacy mechanism that is dynamic, personalized, and explainable, instead of a static one-size-fits-all rule. The paper supports the model with a five-universe Monte Carlo simulation and reports correlations for five hypotheses, with contextual integrity showing the strongest association with utility.

What carries the argument

The machinery is a recursive expected-utility model over a set of simulated universes. Equation (2) defines the utility of an action as a weighted sum $\alpha \rho_t + \beta S_t - \gamma R_t + \delta T_t + \zeta g(D_t) + \theta CI(a_t, C_t)$, with tunable weights; Equation (4) averages this utility over the probability distribution of universes; and Equation (5) adds the discounted future value $\lambda V_{t+1}$. The paper operationalizes contextual integrity in the simulation as $CI_i(t) = \frac{P_i(t) + S_i(t) + T_i(t) + D_i(t)}{1 + R_i(t)}$, so that higher preference, security, trust, and demographic alignment raise the score while higher risk lowers it. This machinery lets a system compare alternative futures and choose the action with the largest expected, discounted utility.

What would settle it

Run a scenario-based survey in which users choose between two privacy actions with contrasting preference, risk, trust, security, demographic, and CI values; compute the expected utility of each action with Eq. (4) using the paper's weights; if a substantial share of users pick the lower-utility action, the expected-utility decision rule is falsified.

Watch

Extended reading notes

Core claim

The central claim is that privacy is not a single outcome to be optimized but a manifold of possible outcomes, and that the best privacy action can be found by weighing those outcomes explicitly. Formally, at time $t$ an action $a_t$ in context $C_t$ yields universes $U_t^i$ with probability $P(U_t^i | a_t, C_t)$; each universe is scored by a utility $UI(a_t, C_t, U_t^i)$ that combines privacy preference, security, risk, trust, demographic sensitivity, and a contextual-integrity score. The optimal action is $a_t^* = \arg\max_{a_t \in A_t} \mathbb{E}[UI(a_t, C_t)]$, and a recursive term $V_t = \mathbb{E}[UI(a_t, C_t)] + \lambda V_{t+1}$ carries future consequences into the present choice. In the simulation, contextual integrity correlates with utility at $r = 0.8129$, privacy preference at $0.6618$, and risk at $-0.6078$, while security level shows no significant correlation.

Load-bearing premise

The load-bearing premise is that inherently qualitative constructs, especially contextual integrity, can be collapsed into scalar numbers and added together into a single utility that real users maximize.

Editorial extensions

If this is right

  • A privacy-aware system could evaluate several candidate actions at each step and pick the one with the highest expected utility, making privacy decisions adaptive as preferences and contexts shift.
  • Because contextual integrity entered the model with the strongest correlation to utility, MPT predicts that aligning a decision with context matters more to users than abstract security metrics.
  • The strong negative risk correlation implies users should become more conservative as contextual risk rises, since privacy utility drops when risk is high.
  • The non-significant security result suggests that security level alone may not drive perceived privacy utility, a claim that could be tested in real deployments.

Reading between the lines

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

  • The paper leaves implicit that the utility function could serve as a reward signal for reinforcement-learning agents, turning MPT from a decision rule into a policy objective for continuous privacy negotiation.
  • A natural extension is to test whether the additive form of Eq. (2) is the right combination rule; a multiplicative or thresholded version would likely change the optimal action in high-risk universes, and the two could be compared on stated user preferences.
  • The simulated H4 result could be sharpened in a survey: vary objective security while holding trust and risk constant, and measure whether users actually change their choices, which would separate perceived from actual security effects.
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 / 6 minor

Summary. The paper introduces Multiverse Privacy Theory (MPT), a decision-theoretic framework in which each privacy action generates multiple potential outcomes ("universes"), and the user/system selects actions by maximizing a discounted expected utility over those universes. The utility function in Eq. (2) is a weighted linear combination of privacy preference, security, risk, trust, demographics, and a contextual integrity score, with the CI score defined in Eq. (6) as a ratio of the same variables. To support the framework, the authors run a Monte-Carlo simulation across five universes and ten time steps, then report Pearson correlations between utility and its components as tests of five hypotheses (H1–H5). The paper concludes that AI systems can use MPT to continuously optimize privacy settings. The abstract explicitly states that real-world, scenario-based survey data are future work, so the present evidence is entirely simulation-based.

Significance. If the framework were validated, it could provide a structured way to reason about privacy decisions under uncertainty and to connect contextual integrity with quantitative utility models. The paper usefully surveys relevant privacy literature and makes an attempt at formalization. However, the current evidence does not support the central claims. The simulation and the hypothesis tests are circular because the utility function (Eq. 2) is defined as a weighted sum of the very inputs whose correlations with utility are then reported as empirical findings. There is no real user data, and the operationalization of Contextual Integrity in Eq. (6) is an ad hoc scalar index that diverges from Nissenbaum's theory. The paper's novelty relative to standard expected-utility models is not clearly articulated, and the formal model is incomplete in places. As a result, the paper's main contribution is a tentative conceptual framework rather than a validated theory.

major comments (5)
  1. [§2, Eq. (2), Eq. (6), Table 1] The reported hypothesis tests H1–H5 are tautological. UI(a_t,C_t) in Eq. (2) is defined as αρ_t + βS_t − γR_t + δT_t + ζg(D_t) + θCI(a_t,C_t), and CI in Eq. (6) is itself a function of the same random variables (P, S, T, D, R). Consequently, the Pearson correlations in Table 1 are algebraic consequences of the chosen weights, not empirical findings. For example, the positive H1 correlation (r=0.66) is implied by positive α, the negative H2 correlation (r=−0.61) by negative γ, and the very large H5 correlation (r=0.81) is near-forced because CI appears inside Eq. (2) and is built from the same inputs. These p-values therefore provide no evidence that the model describes real user behavior.
  2. [§2, H4 and the simulation] The non-significant result for security level (H4: r=0.1581, p=0.2728) is not informative. Because the data are generated with an arbitrary weight β=0.8 and random draws, the lack of correlation is a property of the simulation, not a finding about users' perceptions. The speculation that "users may lack sufficient understanding of security levels" is unsupported by the synthetic data and should be removed or explicitly labeled as conjecture.
  3. [§2, Eq. (6)] The proposed CI score is not a faithful operationalization of Contextual Integrity (CI) as defined in Nissenbaum's work [18]. CI concerns the appropriateness of information flows in contexts defined by actors, recipients, information types, transmission principles, and norms; it is not a scalar sum of privacy preference, security, trust, demographics, and risk. The paper uses the term "CI" for an ad hoc index without justification or any stated relation to the five parameters of CI mentioned in the introduction. This conflates a rich conceptual framework with a simple arithmetic ratio and undermines the claim to bridge CI and empirical metrics.
  4. [§2, Eqs. (1)–(5)] The formal model is incomplete in two important ways. First, the probability P(U_i^t | a_t, C_t) in Eq. (1) is never defined; without this distribution, the expected utility in Eq. (4) cannot be computed in practice. Second, Eq. (2) defines UI(a_t, C_t) without any universe index, yet Eq. (4) uses UI(a_t, C_t, U_i^t) and the text says utility "possibly adjusted for universe-specific consequences." The paper does not specify how the utility depends on U_i, so the summation in Eq. (4) is undefined. These gaps prevent the model from being applied or tested as stated.
  5. [§3, Conclusion] The conclusion overclaims the practical impact of the framework. It states that "AI systems can continuously optimize privacy settings" based on MPT, but the only evidence is a synthetic simulation with arbitrary weights and random inputs. The abstract itself acknowledges that real-world application is future work. The conclusion should be rephrased to describe MPT as a tentative theoretical proposal whose empirical assessment is pending rather than as an established method for system design.
minor comments (6)
  1. [§3] There is a typo in the sentence "Future work will explore MPT to test it to the real-world applicability of the model"; the intended phrase is likely "test its applicability to the real world."
  2. [References] Reference [14] contains a typo: "SJ Unviersity" should be "SJ University". Reference [4] misspells the author's name as "Apthrope" (should be "Apthorpe").
  3. [§2, Figure 1] Figure 1 is referenced in the text but is not included in the manuscript we reviewed, so the reader cannot verify the simulation output or the claim about utility evolution across risk bands.
  4. [§2, Table 1] The paper does not report the sample size used for the Pearson correlations, nor the method used to compute the 95% confidence intervals. These details are necessary for a reader to interpret the statistical results.
  5. [§1] The introduction cites the Governing Knowledge Commons (GKC) and GKC-CI model [21] but never explains how MPT integrates or extends GKC-CI beyond listing it as related work. The connection should be either elaborated or explicitly deferred to future work.
  6. [§2, Eq. (2)] The paper does not clarify whether the weights α, β, γ, δ, ζ, θ are intended to be user-specific parameters fitted from data or global constants set by the system designer. The simulation sets them arbitrarily, which limits the interpretability of the reported correlations.

Circularity Check

1 steps flagged · score 8.0 of 10

Hypothesis tests H1–H5 are tautological: utility is defined as a weighted sum of the very predictors being correlated with it, so the reported correlations are algebraic artifacts of the model.

  1. self definitional [Section 2, 'Example' and 'Hypotheses Testing'; Eqs. (2), (6); Table 1 (H1–H5)]
    "UI(at,Ct)=α·ρt+β·St−γ·Rt+δ·Tt+ζ·g(Dt)+θ·CI(at,Ct) (2) ... CIi(t)=Pi(t)+Si(t)+Ti(t)+Di(t) 1+Ri(t) (6) ... A utility value was computed for each outcome using a weighted combination of the input variables, with weights set to 1.0 for privacy preference (α), 0.8 for security level (β), -0.9 for contextual risk (γ), 0.6 for trust (δ), and 0.5 for demographic sensitivity (ζ). ... H5: CI strongly enhances privacy utility. A very strong positive correlation was found between CI and utility (r=0.8129, p<0.001)."

    By Eq. (2), UI is defined as the weighted sum of ρ, S, R, T, g(D), and CI. By Eq. (6), CI is itself defined from P, S, T, D, and R. The H1–H5 Pearson correlations in Table 1 are correlations of UI with its own additive ingredients, or with a function built from those same ingredients. With randomly generated inputs, the sign and approximate magnitude of each correlation is forced by the coefficient signs (α=+1, β=+0.8, γ=−0.9, δ=+0.6, ζ=+0.5, θ=+1). For example, H1's r=0.66 follows from positive α, H2's r=−0.61 from negative γ, and H5's r=0.81 from CI appearing inside Eq. (2) and being constructed from the same variables. Thus H1–H5 restate the definition of utility rather than testing behavioral hypotheses; no independent utility or choice measurement enters.

full rationale

The paper's central empirical support—the Monte Carlo example and the H1–H5 correlations—is circular by construction, so the circularity score is high. The utility function in Eq. (2) is explicitly a weighted sum of privacy preference, security, contextual risk, trust, demographic sensitivity, and CI, and CI in Eq. (6) is defined from the same normalized inputs. Correlating the computed utility with those same inputs therefore recovers the functional form and parameter signs of the model; it does not validate any claim about real users' privacy decisions. This is a specific equation-to-equation reduction (Eq. (2) + Eq. (6) imply the Table 1 correlations), not a vague impression. The paper also contains self-citations (e.g., reference [11]), but these are background literature and not load-bearing for the model's derivation, so they do not add to the circularity score. If the framework were checked against external scenario-based survey data (as the conclusion promises), that could provide independent evidence; the current manuscript contains no such data.

Assumptions & free parameters 7 free parameters · 4 assumptions · 2 invented entities

The central claim rests on unexamined behavioral and measurement assumptions: users maximize a scalar expected utility, privacy constructs are commensurable numbers, and Eq. (6) faithfully implements contextual integrity. The free parameters are hand-chosen weights, and the invented universe construct has no independent evidence.

free parameters (7)
  • alpha (privacy preference weight) = 1.0
    Hand-chosen in the simulation; because it multiplies rho in Eq. (2), H1's positive correlation is virtually guaranteed.
  • beta (security level weight) = 0.8
    Hand-chosen; contributes to H4's weak correlation depending on simulated variance.
  • gamma (contextual risk coefficient) = 0.9 (applied as -gamma * R in Eq. 2)
    Hand-chosen; enforces H2's negative correlation with utility.
  • delta (trust weight) = 0.6
    Hand-chosen; shapes H3's moderate positive correlation.
  • zeta (demographic sensitivity weight) = 0.5
    Hand-chosen; appears in the simulation but no dedicated hypothesis tests it directly.
  • theta (CI weight) = not specified in simulation
    Defined in Eq. (2) but absent from the simulation description; unclear how H5's CI-utility correlation arises.
  • lambda (discount factor) = not specified
    Defined in Eq. (5) but never assigned or used in the simulation.
assumptions (4)
  • domain assumption Users maximize expected utility over possible outcomes.
    Eqs. (3)-(4) select actions by maximizing expected utility; no empirical justification is offered for this behavioral model.
  • ad hoc to paper Privacy-relevant constructs are scalar, commensurable, and linearly additive.
    Eq. (2) sums preference, security, risk, trust, demographics, and CI into one number, without grounding in measurement theory.
  • ad hoc to paper Eq. (6) is a valid operationalization of Contextual Integrity.
    Nissenbaum's CI is a qualitative theory about norms and information flows; the paper reduces it to a normalized ratio without derivation or validation.
  • domain assumption Random values in the simulation represent realistic user/system states.
    The example draws all variables randomly, so the resulting correlations have no external validity without calibration to real users.
invented entities (2)
  • Parallel universe U_i_t
    purpose: Represents a distinct possible outcome of a privacy action at time t.
    A modeling construct; no falsifiable prediction is attached outside the simulation, so it functions as a rhetorical device rather than a measurable entity.
  • MPT contextual integrity score CI_i(t)
    purpose: Quantifies contextual alignment as a scalar between 0 and 1 for use in the utility function.
    A new composite metric (Eq. 6) that is not derived from Nissenbaum's theory and is never validated against real privacy norms.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multiverse Privacy Theory for Contextual Risks in Complex User-AI Interactions." pith.science (2026). https://pith.science/paper/2ZH55SQA

@misc{pith2026250610042,
  author       = {Pith},
  title        = {Pith review of: Multiverse Privacy Theory for Contextual Risks in Complex User-AI Interactions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2ZH55SQA}},
  note         = {Machine review of arXiv:2506.10042}
}
read the original abstract

In an era of increasing interaction with artificial intelligence (AI), users face evolving privacy decisions shaped by complex, uncertain factors. This paper introduces Multiverse Privacy Theory, a novel framework in which each privacy decision spawns a parallel universe, representing a distinct potential outcome based on user choices over time. By simulating these universes, this theory provides a foundation for understanding privacy through the lens of contextual integrity, evolving preferences, and probabilistic decision-making. Future work will explore its application using real-world, scenario-based survey data.

Figures

Figures reproduced from arXiv: 2506.10042 by the authors.

Figure 1
Figure 1. Simulation showing how user privacy utility evolves across multiple universes and three contextual risk bands. Hypotheses Testing. We also tested the relationships be￾tween different factors and the utility score using Pearson’s correlation. The results are summarized as follows: H1: Privacy preferences significantly affect utility. A strong positive correlation was found between privacy preference and utility (r = … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 23 canonical work pages

  1. [18]

    Stanford University Press, Stanford, 2009

    Helen Nissenbaum.Privacy in Context: Technology, Pol- icy, and the Integrity of Social Life. Stanford University Press, Stanford, 2009

    work page 2009

  2. [1]

    Noah Apthorpe, Yan Shvartzshnaider, Arunesh Mathur, Dillon Reisman, and Nick Feamster. Discovering smart home internet of things privacy norms using contextual integrity.Proceedings of the ACM on Interactive, Mo- bile, Wearable and Ubiquitous Technologies, 2(2):1–23, 2018

    work page 2018

  3. [2]

    Michaela Benk, Léane Wettstein, Nadine Schlicker, Florian von Wangenheim, and Nicolas Scharowski. Bridging the knowledge gap: Understanding user ex- pectations for trustworthy llm standards.Proceed- ings of the AAAI Conference on Artificial Intelligence, 39(2626):27197–27205, April 2025

    work page 2025

  4. [3]

    Artificial intelligence for cybersecurity: Threats, attacks and mitigation

    Abhilash Chakraborty, Anupam Biswas, and Ajoy Ku- mar Khan. Artificial intelligence for cybersecurity: Threats, attacks and mitigation. InArtificial intelligence for societal issues, pages 3–25. Springer, 2023

    work page 2023

  5. [4]

    Automating Governing Knowledge Com- mons and Contextual Integrity (GKC-CI) Privacy Policy Annotations with Large Language Models.Proceedings on Privacy Enhancing Technologies, 2025

    Jake Chanenson, Madison Pickering, and Noah Apthrope. Automating Governing Knowledge Com- mons and Contextual Integrity (GKC-CI) Privacy Policy Annotations with Large Language Models.Proceedings on Privacy Enhancing Technologies, 2025

    work page 2025

  6. [5]

    Rethink- ing youth privacy.Virginia Public Law and Legal Theory Research Paper, (2025-15), 2025

    Danielle Keats Citron and Ari Ezra Waldman. Rethink- ing youth privacy.Virginia Public Law and Legal Theory Research Paper, (2025-15), 2025

    work page 2025

  7. [6]

    Apart from Universes.Many Worlds, pages 542–552, 2010

    David Deutsch. Apart from Universes.Many Worlds, pages 542–552, 2010

    work page 2010

  8. [7]

    The algorithmic foundations of differential privacy.Foundations and Trends® in Theoretical Computer Science, 9(3–4):211– 407, 2014

    Cynthia Dwork, Aaron Roth, et al. The algorithmic foundations of differential privacy.Foundations and Trends® in Theoretical Computer Science, 9(3–4):211– 407, 2014

    work page 2014

Show all 24 references
  1. [8]

    Investigating Software Development Teams Members’ Perceptions of Data Privacy in the Use of Large Language Models (LLMs)

    Fabiano Damasceno Sousa Falcão and Edna Dias Canedo. Investigating Software Development Teams Members’ Perceptions of Data Privacy in the Use of Large Language Models (LLMs). InProceedings of the XXIII Brazilian Symposium on Software Quality, pages 373–382, 2024

    2024

  2. [9]

    Privacy and Secu- rity Threat Models and Mitigation Strategies of Older Adults

    Alisa Frik, Leysan Nurgalieva, Julia Bernd, Joyce Lee, Florian Schaub, and Serge Egelman. Privacy and Secu- rity Threat Models and Mitigation Strategies of Older Adults. InFifteenth Symposium on Usable Privacy and Security (SOUPS 2019), page 21–40, 2019

    2019

  3. [10]

    like lesbians walking the perime- ter

    Christine Geeng, Mike Harris, Elissa Redmiles, and Franziska Roesner. " like lesbians walking the perime- ter": Experiences of{US}.{LGBTQ+} folks with online security, safety, and privacy advice. In31st USENIX Security Symposium (USENIX Security 22), pages 305– 322, 2022

    2022

  4. [11]

    Ece Gumusel. A literature review of user privacy con- cerns in conversational chatbots: A social informatics approach: An Annual Review of Information Science and Technology (ARIST) paper.Journal of the Associa- tion for Information Science and Technology, 76(1):121– 154, 2025. 4

    2025

  5. [12]

    From chatgpt to threat- gpt: Impact of generative ai in cybersecurity and privacy

    Maanak Gupta, CharanKumar Akiri, Kshitiz Aryal, Eli Parker, and Lopamudra Praharaj. From chatgpt to threat- gpt: Impact of generative ai in cybersecurity and privacy. IEEE Access, 11:80218–80245, 2023

    2023

  6. [13]

    Pri- vacy by designers: software developers’ privacy mindset

    Irit Hadar, Tomer Hasson, Oshrat Ayalon, Eran Toch, Michael Birnhack, Sofia Sherman, and Arod Balissa. Pri- vacy by designers: software developers’ privacy mindset. Empirical Software Engineering, 23:259–289, 2018

    2018

  7. [14]

    Contexlot: Towards providing contextual integrity to appified iot platforms

    Yunhan Jack Jia, Qi Alfred Chen, Shiqi Wang, Amir Rahmati, Earlence Fernandes, Zhuoqing Morley Mao, Atul Prakash, and SJ Unviersity. Contexlot: Towards providing contextual integrity to appified iot platforms. Inndss, volume 2, pages 2–2. San Diego, 2017

    2017

  8. [15]

    no telling passcodes out because they’re private

    Priya Kumar, Shalmali Milind Naik, Utkarsha Ramesh Devkar, Marshini Chetty, Tamara L. Clegg, and Jessica Vitak. “no telling passcodes out because they’re private”: Understanding children’s mental models of privacy and security online.Proceedings of ACM Human-Computer Interacti...

    2017

  9. [16]

    Privacy and vulnera- ble populations

    Nora McDonald and Andrea Forte. Privacy and vulnera- ble populations. InModern Socio-technical Perspectives on Privacy, pages 337–363. Springer International Pub- lishing Cham, 2022

    2022

  10. [17]

    dark matter

    Michael B Mensky. Phenomenology of "dark matter"- from the everett’s quantum cosmology.arXiv preprint arXiv:1105.3696, 2011

    2011 arXiv

  11. [19]

    Security and pri- vacy for artificial intelligence: Opportunities and chal- lenges.arXiv preprint arXiv:2102.04661, 2021

    Ayodeji Oseni, Nour Moustafa, Helge Janicke, Peng Liu, Zahir Tari, and Athanasios Vasilakos. Security and pri- vacy for artificial intelligence: Opportunities and chal- lenges.arXiv preprint arXiv:2102.04661, 2021

    2021 arXiv

  12. [20]

    Ghadeer Ghazi Shayea, Mohd Hazli Mohammed Zabil, Mustafa Abdulfattah Habeeb, Yahya Layth Khaleel, and A. S. Albahri. Strategies for protection against adversar- ial attacks in ai models: An in-depth review.Journal of Intelligent Systems, 34(1), January 2025

    2025

  13. [21]

    GKC-CI: A unifying framework for contextual norms and information governance.Journal of the Association for Information Science and Technol- ogy, 73(9):1297–1313, 2022

    Yan Shvartzshnaider, Madelyn Rose Sanfilippo, and Noah Apthorpe. GKC-CI: A unifying framework for contextual norms and information governance.Journal of the Association for Information Science and Technol- ogy, 73(9):1297–1313, 2022

    2022

  14. [22]

    Statistical data privacy: A song of privacy and utility.Annual Review of Statistics and Its Application, 10(1):189–218, 2023

    Aleksandra Slavkovi´c and Jeremy Seeman. Statistical data privacy: A song of privacy and utility.Annual Review of Statistics and Its Application, 10(1):189–218, 2023

    2023

  15. [23]

    It’s a Fair Game

    Zhiping Zhang, Michelle Jia, Hao-Ping Lee, Bingsheng Yao, Sauvik Das, Ada Lerner, Dakuo Wang, and Tian- shi Li. “It’s a Fair Game”, or Is It? Examining How Users Navigate Disclosure Risks and Benefits When Using LLM-Based Conversational Agents. InProceed- ings of the 2024 CHI ...

    2024

  16. [24]

    Addressing conceptual gaps in big data research ethics: An application of contextual integrity.Social Media+ Society, 4(2):2056305118768300, 2018

    Michael Zimmer. Addressing conceptual gaps in big data research ethics: An application of contextual integrity.Social Media+ Society, 4(2):2056305118768300, 2018. 5

    2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.