Pith. sign in

REVIEW 3 major objections 6 minor 34 references

The paper argues that a user-owned, explainable AI layer running in the browser can mediate between platform algorithms and the interface, giving people practical, moment-to-moment control over safety and wellbeing without any platform coop

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

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

T0 review · deepseek-v4-flash

2026-08-03 23:24 UTC pith:WS2G2UOW

load-bearing objection A promising but overclaimed design-pattern paper: the five-pattern HL-AI taxonomy is worth discussing, but the working prototype is unsubstantiated and the optimization math is internally inconsistent. the 3 major comments →

arxiv 2511.05875 v1 pith:WS2G2UOW submitted 2025-11-08 cs.HC cs.AIcs.CV

Towards a Humanized Social-Media Ecosystem: AI-Augmented HCI Design Patterns for Safety, Agency & Well-Being

Mohd Ruhul Ameen , Akif Islam This is my paper
classification cs.HC cs.AIcs.CV
keywords Human-Layer AIdesign patternssocial media wellbeinguser agencybrowser extensioncontent moderationdigital wellbeinghuman-AI interaction
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to establish that social media's engagement-first problems can be addressed not by changing platforms, but by inserting a user-owned, explainable AI layer in the browser that mediates between platform content and the user. This 'Human-Layer AI' (HL-AI) works on top of any site without platform cooperation, and the authors claim a working prototype demonstrates feasibility today. The paper argues these intermediaries can rewrite harmful drafts, flag integrity issues, let users curate feeds, insert micro-withdrawals from compulsive scrolling, and provide recovery mode during harassment, all while preserving agency through explanations and overrides. A mathematical optimization balances user utility, agency compromise, and risk, providing a principled basis for future evaluation and adoption.

Core claim

The central claim is that a user-owned, explainable AI intermediary running in the browser can give individuals practical, moment-to-moment control over their social media experience without requiring platform cooperation. The paper contributes a working Chrome/Edge prototype implementing five representative design patterns—Context-Aware Post Rewriter, Post Integrity Meter, Granular Feed Curator, Micro-Withdrawal Agent, and Recovery Mode—alongside a unifying optimization objective J(a) = u(a) − λΩ(a) − β1_{r(a)>τ}r(a), which selects interventions by trading off user utility, agency compromise, and risk penalty above a user-set threshold. The argument is that this architecture bypasses the 's

What carries the argument

The central object is the Human-Layer AI (HL-AI) browser intermediary, a privacy-preserving extension that injects controls and explanations directly into platform pages. The key identity is the optimization objective J(a) = u(a) − λΩ(a) − β1_{r(a)>τ}r(a), which formalizes intervention selection as a trade-off among user utility, autonomy loss, and risk. Algorithm 1 implements this objective by initializing utility and risk estimators, computing an agency penalty, applying an extra penalty above the risk threshold τ, and returning the argmax action with an explanation and override option. Each of the five patterns instantiates the same three mechanisms: a transparency layer, user-adjustable

Load-bearing premise

The load-bearing premise is that the quantities in the objective—user utility, agency compromise, and risk—can be reliably estimated in real time from browser-visible content and user behavior, and that maximizing the scalarized objective actually improves wellbeing; if these estimates are inaccurate, the system selects the wrong interventions.

What would settle it

A randomized field trial comparing the full optimizer against a deliberately degraded version that ignores the risk penalty (or picks random low-cost interventions when risk exceeds τ) on measures of post-session regret, perceived control, and compulsivity would settle whether the objective function's risk term adds benefit beyond the interface affordances themselves. If the degraded version performs equally, the optimization's estimates are not capturing wellbeing.

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

If this is right

  • A browser extension can add safety, agency, and wellbeing controls to existing social platforms without API access or platform cooperation.
  • The five patterns—post rewriting, integrity metering, granular feed curation, micro-withdrawal, and recovery mode—are feasible to implement today in Chromium browsers and cover a useful range of harms.
  • Users can calibrate the autonomy–risk balance to their own tolerance via parameters λ and τ, preserving override and explanation.
  • The design pattern language and prototype invite rigorous cross-cultural user evaluation, which the paper explicitly leaves as future work.
  • The approach offers an 'early blueprint' for retrofitting today's feeds, centering on user sovereignty rather than engagement optimization.

Where Pith is reading between the lines

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

  • If HL-AI works as claimed, the locus of algorithmic accountability shifts from platform servers to user-controlled code, potentially reshaping regulatory and auditing debates—a consequence the paper does not itself draw.
  • The same intermediary architecture could be extended beyond the five patterns to news reading, video streaming, or any algorithmically sorted interface, though only social feeds are discussed.
  • A testable extension would be measuring whether users' actual override rates and intervention acceptance in a field trial track the J(a) trade-off parameters (λ, τ) as predicted by the optimization.
  • The practical bottleneck for this approach is likely the real-time estimation of u(a), Ω(a), and r(a) from browser-visible signals, not interface design or pattern coverage.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes Human-Layer AI (HL-AI), a browser-extension intermediary that sits between social-media platforms and the user interface, applying HCI design patterns to improve safety, agency, and well-being. Five representative patterns are described (Context-Aware Post Rewriter, Post Integrity Meter, Granular Feed Curator, Micro-Withdrawal Agent, Recovery Mode), together with a fifteen-pattern taxonomy said to appear in a future journal publication. A unified optimization objective J(a)=u(a)−λΩ(a)−β1_{r(a)>τ}r(a) is introduced to balance user utility, agency compromise, and risk. The abstract claims a working Chrome/Edge prototype implementing the five patterns and evaluation spanning accuracy, usability, and behavioral outcomes; however, the body states that full empirical evaluation is reserved for future work, the only implementation figure is labeled "(placeholder)", and no code or repository is provided. The paper is best read as a design-pattern proposal with a planned evaluation.

Significance. If the proposed architecture and pattern language were implemented and validated, they would address a timely and important gap: giving users granular, user-owned control over social feeds without requiring platform cooperation. The pattern tables explicitly capture design tensions (e.g., reflection vs. speed, transparency vs. overload) and emphasize user override and explainability, which are constructive contributions. The paper is not circular: no parameters are fitted to outcomes, and the optimization objective is a proposed design with user-set weights. As submitted, however, the central advertised contributions—the working prototype and the evaluation—are unsubstantiated, and the mathematical core is internally inconsistent. The significance therefore rests on the plausibility of the design patterns and the promise of future validation.

major comments (3)
  1. [Abstract vs. Sections I and VI] The abstract states that the paper contributes "a working Chrome/Edge prototype implementing five representative pattern frameworks" and that "evaluation spans technical accuracy, usability, and behavioral outcomes." But Section I says "full empirical evaluation is reserved for future work," and Section VI's Fig. 7 is captioned "(placeholder)" with no actual screenshots, architecture diagram, code repository, or extension package. This is not a minor overstatement: the paper's central claim is that HL-AI provides "practical, moment-to-moment control" today. Please either provide the working prototype and its evaluation data, or re-scope the claims to a design-pattern proposal with a planned evaluation.
  2. [§IV Eq. (1) and §V Algorithm 1] Equation (1) defines J(a)=u(a)−λΩ(a)−β1_{r(a)>τ}r(a). Algorithm 1 computes, at line 4, J←λ·û+(1−λ)·r̂−Ω(a). This differs from Eq. (1) in three substantive ways: the agency penalty is not multiplied by λ, risk enters as a positive weighted term (1−λ)·r̂ instead of a conditional negative penalty, and the β/τ mechanism is missing. Moreover, Algorithm 1's Require clause does not list β, although line 6 uses it. Since the paper presents Eq. (1) as the unifying mathematical basis of the framework, this inconsistency cannot be left as a typo; the algorithm and equation must be aligned.
  3. [§IV and Algorithm 1 (estimators)] The framework's practical viability depends on real-time estimates of user utility u(a), agency compromise Ω(a), and risk r(a). Algorithm 1 simply says "Initialize utility estimator û(a) and risk estimator r̂(a)" without specifying how these are implemented, calibrated, or validated. Algorithm 2 partially addresses risk for posts by referencing an external fact-check database and an AI detection model, but does not name the database/model, their accuracy, or how a browser extension accesses them. If these estimates are unreliable, the optimization will select the wrong interventions. At minimum, the manuscript should clearly identify these as assumptions, separate demonstrated implementation from planned methodology, and describe a concrete estimation strategy.
minor comments (6)
  1. [§V.C] Section V.C says Algorithm 2 details "Pattern P3 (Post Integrity Meter)," but Post Integrity Meter is Pattern P2 in Section III; P3 is Granular Feed Curator. Please fix the numbering.
  2. [Algorithm 1] The Require clause lists λ and τ but not β, even though β appears in the body. Add β to the inputs, or explain how it is obtained.
  3. [§VI] The text says "Figure 7 presents the system architecture along with prototype screenshots," but the caption is "(placeholder)." If no real artifact is available, say so explicitly and remove the implication of a completed implementation.
  4. [§VI / Abstract] The abstract says "we release design patterns and implementation details to support adoption," but no repository, artifact link, or implementation document is provided. Clarify what exactly is released.
  5. [Figures 1–5] Several figures are referenced only by captions and appear to be illustrative wireframes. Please state explicitly whether these are mockups or actual system screenshots, since the distinction matters for the prototype claim.
  6. [Introduction, static statistics] The statistics on post-session regret, trust, and misinformation sharing are presented without survey details or sample sizes. If they are used to motivate the work, cite the original surveys directly rather than secondary summaries.

Circularity Check

0 steps flagged

No significant circularity: Eq. (1) is a proposed design objective with user-set weights, the prototype/evaluation gaps are evidentiary rather than circular, and there is no self-citation chain.

full rationale

The paper's central derivation chain is limited to a proposed scalarization objective J(a)=u(a)−λΩ(a)−β1_{r(a)>τ}r(a) (Eq. 1) and Algorithm 1. This is a design commitment, not a prediction or a fit: u, Ω, r are inputs to be estimated, λ, β, τ are user-set parameters, and the paper reports no empirical outcome that the objective is supposed to reproduce. No parameter is fitted to a subset of data and then treated as a prediction of a related quantity. There are no self-citations by Ameen/Islam; references [1]–[34] are external reports, prior HCI/XAI literature, and news sources, and no uniqueness theorem or prior-work ansatz is imported. The fifteen-pattern taxonomy is presented as a design-pattern language, not as a derivation of a known result under new coordinates. Reviewing-rule flags: Section I states 'full empirical evaluation is reserved for future work'; Fig. 7 is captioned '(placeholder)'; Algorithm 2 references an unspecified fact-check database D and AI detection model M_AI; and Eq. (1) is not syntactically equivalent to Algorithm 1's line 4 (which uses λû+(1−λ)r̂−Ω). These are evidentiary/completeness problems and a formal inconsistency, but none of them reduces a claimed result to its own inputs, so they do not constitute circularity.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 1 invented entities

The central claim rests on quantifiability of human wellbeing and browser-level access; these are domain assumptions rather than derived results. The model's free parameters λ, β, τ are user-set and not fitted, but they are still free knobs the framework depends on.

free parameters (3)
  • λ (autonomy weight) = user-set; no default specified
    Controls the trade-off between utility and agency cost in Eq. (1)/Algorithm 1; no recommended value or fitting procedure is given.
  • β (risk penalty weight) = user-set; no default specified
    Scales the penalty applied when risk exceeds threshold in Eq. (1); no default or fitting procedure is given.
  • τ (risk tolerance threshold) = user-set; no default specified
    Risk level above which interventions incur an extra penalty; described as user-adjustable but never operationalized.
axioms (4)
  • domain assumption Engagement-first algorithms cause measurable psychological and informational harm that user control can mitigate.
    Motivates the entire design; supported only by cited third-party surveys (e.g., 62% post-session regret, 28% trust, 38.2% accidental misinformation) and event anecdotes, not by this paper's own data.
  • ad hoc to paper u(a), Ω(a), r(a) are quantifiable scalar values that can be compared by subtraction in Eq. (1).
    No operationalization, units, or measurement methods are provided; Algorithm 1 simply assumes initializable estimators û and r̂.
  • domain assumption A browser extension can access, parse, and modify platform content and inject UI in real time without platform cooperation and without breaking.
    Section VI describes the architecture, but this is not tested against real platform DOM restrictions, anti-bot measures, or terms-of-service constraints.
  • ad hoc to paper The full fifteen-pattern taxonomy exists and the five presented are representative.
    Section I says the complete set will appear in a forthcoming journal publication, which cannot be independently checked in this preprint.
invented entities (1)
  • Human-Layer AI (HL-AI) intermediary no independent evidence
    purpose: User-owned browser agent that mediates between platform algorithms and the user interface to apply safety, agency, and wellbeing patterns
    Central proposed entity; no deployed implementation, benchmark, or third-party verification is provided, and the only screenshots are labeled placeholder.

pith-pipeline@v1.3.0-alltime-deepseek · 7439 in / 10865 out tokens · 100520 ms · 2026-08-03T23:24:10.748118+00:00 · methodology

0 comments
read the original abstract

Social platforms connect billions of people, yet their engagement-first algorithms often work on users rather than with them, amplifying stress, misinformation, and a loss of control. We propose Human-Layer AI (HL-AI)--user-owned, explainable intermediaries that sit in the browser between platform logic and the interface. HL-AI gives people practical, moment-to-moment control without requiring platform cooperation. We contribute a working Chrome/Edge prototype implementing five representative pattern frameworks--Context-Aware Post Rewriter, Post Integrity Meter, Granular Feed Curator, Micro-Withdrawal Agent, and Recovery Mode--alongside a unifying mathematical formulation balancing user utility, autonomy costs, and risk thresholds. Evaluation spans technical accuracy, usability, and behavioral outcomes. The result is a suite of humane controls that help users rewrite before harm, read with integrity cues, tune feeds with intention, pause compulsive loops, and seek shelter during harassment, all while preserving agency through explanations and override options. This prototype offers a practical path to retrofit today's feeds with safety, agency, and well-being, inviting rigorous cross-cultural user evaluation.

Figures

Figures reproduced from arXiv: 2511.05875 by Akif Islam, Mohd Ruhul Ameen.

Figure 1
Figure 1. Figure 1: Example of P1. Context-Aware Post Rewriter [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Example of P2. Post Integrity Meter interpreted, and provides neutral or empathetic alternatives while leaving final choice to the user. B. P2. Post Integrity Meter This pattern provides real-time integrity signals for posts by checking factual consistency, AI-generation likelihood, and political bias. Instead of restricting access, the intermediary offers transparent cues that help users judge credibility… view at source ↗
Figure 3
Figure 3. Figure 3: Example of P3. Granular Feed Curator TABLE II P2 — INTEGRITY METER Intent Provide clear, real-time integrity cues showing whether content may be fake, AI-generated, or biased, supporting informed decision-making. Problem Users struggle to distinguish reliable posts from mis￾information or synthetic media in fast-moving environ￾ments. Context High-stakes domains such as health, elections, and crises; mixed … view at source ↗
Figure 5
Figure 5. Figure 5: Example of P5. Recovery Mode TABLE V P5 — RECOVERY MODE Intent Provide an immediate, protective mode after harassment or overload that reduces exposure, limits inbound inter￾actions, and surfaces appropriate support. Problem Following negative events, users face toxic replies, brigading, and decision paralysis. Default feeds continue amplifying harm instead of stabilizing the situation. Context Online pile… view at source ↗
Figure 6
Figure 6. Figure 6: HL-AI master decision flow: pattern coordination and optimization. [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Prototype screenshots of HL-AI browser-based intermediary (place [PITH_FULL_IMAGE:figures/full_fig_p005_7.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

34 extracted references · 3 linked inside Pith

  1. [1]

    Digital 2024: Global overview report,

    Kepios, “Digital 2024: Global overview report,” DataReportal, Mar. 2024, [Online]. Available: https://datareportal.com/reports/digital- 2024-global-overview-report

    2024

  2. [2]

    Social drivers and algorithmic mechanisms on digital media,

    H. Metzler and D. Garcia, “Social drivers and algorithmic mechanisms on digital media,”Perspectives on Psychological Science, vol. 19, no. 5, pp. 735–748, 2024

    2024

  3. [3]

    The right to know social media algorithms,

    H. Sun, “The right to know social media algorithms,”Harvard Law & Policy Review, vol. 18, no. 1, pp. 1–47, 2024

    2024

  4. [4]

    Milestone tragedy: Survivors face trauma beyond injuries,

    Dhaka Tribune, “Milestone tragedy: Survivors face trauma beyond injuries,” Dhaka Tribune, Jul. 2025, [Online]. Available: https://www. dhakatribune.com/feature/health-wellness/387147/milestone-tragedy- survivors-face-trauma-beyond

    2025

  5. [5]

    Bangladesh air force jet crash: Distraught students demand answers after crash turned dhaka school into ’death trap’,

    CNN, “Bangladesh air force jet crash: Distraught students demand answers after crash turned dhaka school into ’death trap’,” CNN, Jul. 2025, [Online]. Available: https://www.cnn.com/2025/07/22/asia/bangla desh-plane-crash-witnesses-intl-hnk

    2025

  6. [6]

    Survivors of the milestone jet crash and the road to emotional recovery,

    The Daily Star, “Survivors of the milestone jet crash and the road to emotional recovery,” The Daily Star, Jul. 2025, [Online]. Available: ht tps://www.thedailystar.net/rising-stars/news/survivors-the-milestone- jet-crash-and-the-road-emotional-recovery-3951551

    2025

  7. [7]

    How bangladesh’s ’gen z’ protests brought down pm sheikh hasina,

    Al Jazeera, “How bangladesh’s ’gen z’ protests brought down pm sheikh hasina,” Al Jazeera, Aug. 2024, [Online]. Available: https://www.alja zeera.com/news/longform/2024/8/7/how-bangladeshs-gen-z-protests- brought-down-pm-sheikh-hasina

    2024

  8. [8]

    July revolution (bangladesh),

    Wikipedia, “July revolution (bangladesh),” 2024, [Online]. Available: https://en.wikipedia.org/wiki/July_Revolution_(Bangladesh)

    2024

  9. [9]

    Bangladesh’s accidental revolutionaries topple sheikh hasina — what’s next?

    G. Macdonald, “Bangladesh’s accidental revolutionaries topple sheikh hasina — what’s next?” United States Institute of Peace, Aug. 2024, [Online]. Available: https://www.usip.org/publications/2024/08/bangla deshs-accidental-revolutionaries-topple-sheikh-hasina-whats-next

    2024

  10. [10]

    ’more egalitarian’: How nepal’s gen z used gaming app discord to pick pm,

    Al Jazeera, “’more egalitarian’: How nepal’s gen z used gaming app discord to pick pm,” Al Jazeera, Sep. 2025, [Online]. Available: https: //www.aljazeera.com/news/2025/9/15/more-egalitarian-how-nepals- gen-z-used-gaming-app-discord-to-pick-pm

    2025

  11. [11]

    2025 nepalese gen z protests,

    Wikipedia, “2025 nepalese gen z protests,” 2025, [Online]. Available: https://en.wikipedia.org/wiki/2025_Nepalese_Gen_Z_protests

    2025

  12. [12]

    How gen z-led protests put nepal’s 1st female prime minister in power,

    NPR, “How gen z-led protests put nepal’s 1st female prime minister in power,” NPR, Sep. 2025, [Online]. Available: https://www.npr.org/2025 /09/20/nx-s1-5545760/nepal-protests-gen-z

    2025

  13. [13]

    Trace of misinformation about bangladesh in 72 indian media outlets in 2024,

    Rumor Scanner, “Trace of misinformation about bangladesh in 72 indian media outlets in 2024,” Rumor Scanner, Jan. 2025, [Online]. Available: https://rumorscanner.com/en/india-en/indian-misinformation-on- bangladesh-2024/135322

    2024

  14. [14]

    Tensions peak as bangladesh blames india for ’spreading misinformation’,

    V oice of America, “Tensions peak as bangladesh blames india for ’spreading misinformation’,” V oice of America, Dec. 2024, [Online]. Available: https://www.voanews.com/a/tensions-peak-as-bangladesh- blames-india-for-spreading-misinformation-/7895753.html

    arXiv 2024

  15. [15]

    Understanding social media recommendation algorithms: Towards a better-informed debate on the effects of social media,

    A. Narayanan, “Understanding social media recommendation algorithms: Towards a better-informed debate on the effects of social media,” Knight First Amendment Institute, Columbia University, New York, NY , USA, Essay, Mar. 2023, visiting Senior Research Scientist Essay. [Online]. Available: https://knightcolumbia.org/content /understanding-social-media-rec...

    2023

  16. [16]

    Share of people who have ever accidentally shared fake news or information on social media in the united states as of december 2020,

    A. Watson, “Share of people who have ever accidentally shared fake news or information on social media in the united states as of december 2020,” Statista, Mar. 2023, statista Statistic No. 657111. Available: https: //www.statista.com/statistics/657111/fake-news-sharing-online/

    2020

  17. [17]

    ‘persecution’ of hindus in bangladesh: Fake posts uncov- ered by bbc,

    BBC Verify, “‘persecution’ of hindus in bangladesh: Fake posts uncov- ered by bbc,” The Business Standard, Aug. 2024, [Online]. Available: https://www.tbsnews.net/bangladesh/persecution-hindus-bangladesh- fake-posts-uncovered-bbc-914141

    2024

  18. [18]

    Six human-centered artificial intelligence grand challenges,

    O. O. Garibayet al., “Six human-centered artificial intelligence grand challenges,”International Journal of Human-Computer Interaction, vol. 39, no. 3, pp. 391–437, 2023

    2023

  19. [19]

    Seven hci grand challenges,

    C. Stephanidiset al., “Seven hci grand challenges,”International Jour- nal of Human-Computer Interaction, vol. 39, no. 4, pp. 1–47, 2023

    2023

  20. [20]

    Digital well-being theory and research,

    M. Büchi, “Digital well-being theory and research,”New Media & Society, vol. 26, no. 1, pp. 155–183, 2024

    2024

  21. [21]

    Children’s wellbeing in a digital world: Year three index report 2024,

    Internet Matters, “Children’s wellbeing in a digital world: Year three index report 2024,” Internet Matters, London, UK, Tech. Rep., Sep. 2024, developed with BMG Research. [Online]. Available: https://www.internetmatters.org/hub/research/childrens-wellbeing-in-a- digital-world-index-report-2024/

    2024

  22. [22]

    Exploring the effects of human-centered ai explanations on trust and reliance,

    N. Scharowski, S. A. C. Perrig, M. Svab, K. Opwis, and F. Brühlmann, “Exploring the effects of human-centered ai explanations on trust and reliance,”Frontiers in Computer Science, vol. 5, Jul. 2023

    2023

  23. [23]

    Human-centered explainable ai (xai): From algorithms to user experiences,

    Q. V . Liao and K. R. Varshney, “Human-centered explainable ai (xai): From algorithms to user experiences,” 2022, arXiv preprint arXiv:2110.10790

    Pith/arXiv arXiv 2022

  24. [24]

    Explainable ai for improved human-computer interaction,

    C. Molnar, “Explainable ai for improved human-computer interaction,” inProceedings of the Second World Conference on Explainable Arti- ficial Intelligence (xAI 2024), ser. Communications in Computer and Information Science, vol. 2156. Valletta, Malta: Springer, Jul. 2024

    2024

  25. [25]

    Generative language models and automated influence operations: Emerging threats and potential mitigations,

    J. A. Goldsteinet al., “Generative language models and automated influence operations: Emerging threats and potential mitigations,”arXiv preprint arXiv:2301.04246, 2023

    Pith/arXiv arXiv 2023

  26. [26]

    The use of artificial intelligence in counter- disinformation: a world wide (web) mapping,

    F. Pilati and T. Venturini, “The use of artificial intelligence in counter- disinformation: a world wide (web) mapping,”Frontiers in Political Science, vol. 7, Jan. 2025

    2025

  27. [27]

    A comprehensive evaluation of explainable artificial intelligence applications: a systematic review,

    G. Vilone and L. Longo, “A comprehensive evaluation of explainable artificial intelligence applications: a systematic review,”Frontiers in Artificial Intelligence, vol. 7, Sep. 2024

    2024

  28. [28]

    An approach to sociotechnical transparency of social media algorithms using agent-based modelling,

    A. Gausen, C. Guo, and W. Luk, “An approach to sociotechnical transparency of social media algorithms using agent-based modelling,” AI and Ethics, vol. 5, pp. 1827–1845, 2025

    2025

  29. [29]

    Ai algorithm transparency, pipelines for trust not prisms: Mitigating general negative attitudes and enhancing trust toward ai,

    K. Park and H. Y . Yoon, “Ai algorithm transparency, pipelines for trust not prisms: Mitigating general negative attitudes and enhancing trust toward ai,”Humanities and Social Sciences Communications, vol. 12, no. 1, p. 1160, Jul. 2025

    2025

  30. [30]

    Achieving digital wellbe- ing through digital self-control tools: A systematic review and meta- analysis,

    A. Monge Roffarello and L. De Russis, “Achieving digital wellbe- ing through digital self-control tools: A systematic review and meta- analysis,”ACM Transactions on Computer-Human Interaction, 2023

    2023

  31. [31]

    Digital wellbeing as a dynamic construct,

    M. Vanden Abeele, “Digital wellbeing as a dynamic construct,”Com- munication Theory, vol. 31, no. 4, pp. 932–955, 2020

    2020

  32. [32]

    Designing for motivation, en- gagement and wellbeing in digital experience,

    D. Peters, R. A. Calvo, and R. M. Ryan, “Designing for motivation, en- gagement and wellbeing in digital experience,”Frontiers in Psychology, vol. 9, p. 797, 2018

    2018

  33. [33]

    Theory and practice of social media’s content moderation by artificial intelligence in light of european union’s ai act and digital services act,

    “Theory and practice of social media’s content moderation by artificial intelligence in light of european union’s ai act and digital services act,” European Journal of Law and Political Science, Feb. 2025

    2025

  34. [34]

    Characterizing ai-generated misinformation on social media,

    N. Pröllochset al., “Characterizing ai-generated misinformation on social media,”arXiv preprint arXiv:2505.10266, 2025

    Pith/arXiv arXiv 2025