{"id":"76df919b-a722-449a-bf94-c0de43b29789","arxiv_id":"2501.14748","paper_version":1,"verdict":"UNVERDICTED","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"An essay tracing the harms faced by 'Global Majority' internet users and moderators to a single root cause, exploitation, and urging nuanced, community-centered policy responses.","lead":"This essay argues that exploitation, rather than technical failure, is the root cause of harmful online experiences for 'Global Majority' users, from traumatic moderation work to weak platform enforcement. It calls on researchers and policymakers to move beyond blanket terms and center community-led design, and is published as a chapter in a UN Internet Governance Forum report.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'root cause' claim is underdetermined: the essay labels a mix of labor exploitation, neglect, and resource constraints as one exploitative mechanism, without ruling out non-exploitative explanations or specifying a causal mechanism.","rationale":"The essay is valuable as a policy-oriented synthesis, but its strongest claim—that exploitation is the root cause of the connected harms—is not established by the text. The cited evidence supports a weaker, still important claim: Global Majority users and moderators are systematically underserved and, in the moderator case, badly paid. But the essay treats 'exploitation' as covering both surplus extraction (moderator wages, data labeling) and mere neglect (under-enforcement, missing language resources). A root cause must be a mechanism, not a label; the text supplies no mechanism that distinguishes exploitation from the ordinary profit-seeking of a platform in markets with low revenue per user and scarce training data. The paper also concedes the 'Global Majority' is not a single thing, which raises the burden of showing a shared causal pathway. The proposed wage-ratio check is one way to test whether the flagship labor example actually demonstrates targeted exploitation rather than market equilibrium; if it fails, the title's strongest formulation should be softened. Since the piece is an argumentative essay rather than an empirical study, the reader's UNVERDICTED verdict remains appropriate even after this concern.","tokens_in":6109,"tokens_out":6869,"duration_ms":68574,"concrete_test":"Re-analyze the flagship moderator-wage example: compute the ratio of typical Nairobi content-moderator pay at the third-party vendor (e.g., Sama) to the Nairobi living wage, and the same ratio for equivalent US moderator compensation. Then adjust for Meta's per-user ad revenue in Kenya vs. the US. If the pay-ratio gap is fully explained by local market wages and lower per-user revenue, the decisive labor-exploitation example loses force; if the gap persists after controls, the exploitation claim is strengthened. Publish the calculation with sources and a sensitivity analysis on wage estimates to make the result checkable.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Central claim is that exploitation is the root cause that ties together harms to Global Majority users and moderators. This is the load-bearing assertion, and the essay does not establish it. The text shifts between two meanings of 'exploitation': (1) a transfer of surplus from low-paid moderators to platform companies, and (2) a broad label for any under-provision of care or enforcement ('exploitation and neglect', 'failure to effectively moderate'). The evidence displays the same split: the Kenyan moderators' lawsuits and ChatGPT data-labeling work are labor exploitation; the English-only moderation and YouTube Amharic harms are more naturally explained by data scarcity and resource allocation; the Myanmar/Ethiopia and WhatsApp cases are enforcement/policy failures. These may share a historical and economic ancestry, but showing that is an argument, not an observation. A root-cause claim must rule out alternative mechanisms that do not require exploitative intent: low expected revenue per user, missing language data, differing labor-law regimes, and ordinary cost minimization. The beneficiary of exploitation also needs to be identified in each case; the essay does not do that. Finally, the paper itself insists that 'the Global Majority is many' and that there is no single 'AI from the Global Majority', which raises the evidential burden for a single shared mechanism. Without a defined mechanism and a contrast class, 'exploitation all the way down' is an interpretive frame rather than a demonstrated cause.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This essay argues that the many harms experienced by \"Global Majority\" users and workers on online platforms—content-moderation failures, traumatic and exploitative labor conditions, surveillance risks, and exclusion from AI benefits—are not separate technical or policy failures but manifestations of one underlying cause: continued exploitation and neglect by Big Tech and by states outside the Global Majority, understood as continuous with colonial extraction. The paper draws on reported cases (Facebook in Myanmar and Ethiopia, WhatsApp misinformation in India, Kenyan and Ethiopian moderator labor, and the YouTube study of Ethiopian users), industry statistics (Meta's 87% US-focused moderation, YouTube's 89.2% English moderators), and prior scholarship on digital colonialism, and it closes by calling for nuance within blanket terminology, participatory design, and engagement with community organizations.","tokens_in":6356,"tokens_out":4505,"duration_ms":43115,"significance":"If the paper's framing were established, it would meaningfully shift policy and research attention from isolated moderation failures to the structural political economy that connects labor exploitation, underenforcement, and user harm. The essay's strength is its synthesis: it brings together labor-rights cases, user-facing harms, and language-technology exclusions in one narrative, and it explicitly acknowledges that the Global Majority is internally diverse—an important corrective to uniform policy proposals. The published venue (a UN IGF coalition report chapter) gives it policy visibility. The paper is also honest about the limits of its own proposals, grounding them in participatory design rather than top-down regulation. That said, it is an essay rather than a systematic study, and its central causal claim about a single root cause is asserted more than demonstrated.","major_comments":[{"comment":"The paper's central claim—that \"at the center of this cycle of harm is the exploitation and neglect of the wide swath of communities\"—is asserted but not established with a defined causal mechanism. The text shifts between two senses of exploitation: surplus extraction from low-paid moderators (Kenyan lawsuits, ChatGPT data labeling) and the broader neglect of non-English-speaking users (English-only moderators, resource scarcity, enforcement failures). These are different mechanisms with different beneficiaries; for example, the under-provision of moderation for low-resource languages is more parsimoniously explained by cost minimization and missing training data than by an intent to exploit. A root-cause claim must name the mechanism, identify the beneficiary in each case, and specify a contrast class of plausible non-exploitative explanations that are ruled out. Without that, the central claim is underdetermined rather than demonstrated.","section":"The Cycle of Harm in Moderation and Inclusion"},{"comment":"The essay explicitly states that \"there is no single 'AI from the Global Majority' because the 'Global Majority' is many\" and that class, ethnicity, and political context create different layers of harm. This acknowledgment raises the evidential burden for the preceding claim that a shared mechanism—exploitation—ties all these communities' experiences together. If the realities of, say, an Ethiopian domestic worker in Lebanon, an Ethiopian software engineer in the US, and an Indigenous community on occupied land are as different as the paper says, the reader needs a specification of how each harm traces to the same exploitative root rather than to distinct local conditions. The paper should either reconcile this tension by identifying the common causal pathway at the right level of abstraction or reframe the claim as an interpretive hypothesis rather than a root-cause finding.","section":"What Lies Under Blanket Terminologies?"},{"comment":"The section on \"False Dichotomies of Harm\" asserts that current moderation alternatives present a false choice between surveillance and neglect, calling surveillance an \"inevitable consequence\" of effective moderation. However, the alleged dichotomy is never precisely defined, and no argument is given for why these two options exhaust the relevant space or why alternatives such as community-based moderation, cooperative governance, or non-profit infrastructure are excluded. The rhetorical question \"If the premise of moderation was not capitalistic and exploitative, could we have safer online experiences?\" presupposes exactly the exploitative premise that the essay is trying to establish. This is a load-bearing gap, because the \"exploitation all the way down\" thesis depends on showing that the observed harms are not merely the result of resource constraints or poor implementation, but of an exploitative structure that makes those the only options.","section":"False Dichotomies of Harm"},{"comment":"The evidence is curated rather than systematic. The paper relies on a small set of case studies (Ethiopian YouTube users, Kenyan moderators, Facebook in Myanmar/Ethiopia, WhatsApp in India) and selected statistics (90% of users outside the US/Canada, 87% of moderation time on US posts, 89.2% of moderators operating in English, 60% higher regret exposure for non-English users). These are cited without addressing whether they are representative of the Global Majority as a whole. Given that the paper itself insists on the diversity of the Global Majority and criticizes blanket terminology, the inference from selected examples to a single root cause needs explicit justification—or the claim should be softened to a common pattern rather than a root cause. At minimum, the paper should acknowledge the selection basis and state what kind of evidence would count against the root-cause claim.","section":"Introduction and Discussion"}],"minor_comments":[{"comment":"The title as printed in the manuscript header reads \"Calling out theRoot Cause\"—missing a space—and the footnote contains \"Global Majoirty.\" These should be corrected.","section":"Title"},{"comment":"The in-text citation \"Dias Oliva, 2020\" does not match the reference list entry \"Dias Oliva, T., Antonialli, D. M., & Gomes, A. (2021)\"; please harmonize the year and author list.","section":"References"},{"comment":"The sentence \"which has lead to a lack of data\" should read \"led\" rather than \"lead.\"","section":"Introduction"},{"comment":"The phrase \"users resort toVPNservicesto access the platforms\" is missing spaces around \"VPN\" and reads as a typographical error rather than an intentional formatting choice.","section":"What Lies Under Blanket Terminologies?"}],"recommendation":"major_revision","confidential_remarks":"This appears to be an invited chapter for a UN IGF report, and it functions as a critical essay rather than a conventional empirical paper. Given that venue, some reviewers might accept the rhetorical style; however, for a journal audience, the central claim needs to be reframed as a hypothesis or interpretive frame, or supported with a defined mechanism and a contrast class. The self-citation to Nigatu & Raji (2024) is appropriate, as that work supplies independent user-interview evidence, and no circularity concern arises."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The short version: this is a good critical essay, not a research paper, and the central causal claim is more asserted than demonstrated. But the synthesis is genuinely useful, and the call to action is sensible.\n\nWhat it does well: it connects content moderation harms, moderator labor conditions, and language coverage gaps under one framework. The citation base is solid — Kwet, Roberts, Shahid & Vashistha, Sambasivan et al., Nigatu & Raji — and the concrete Ethiopian women YouTube example gives texture to the argument that blanket policies fail. The critique of the \"surveillance or neglect\" false dichotomy is fair, and the final recommendation to center community organizations and participatory design is pragmatic. For a policy-facing essay, the writing is clear and the structure works.\n\nThe soft spot is real, though: \"exploitation all the way down\" is a framing device, not an established mechanism. The evidence splits. Moderator lawsuits and ChatGPT data labor are literal exploitation — low-paid workers producing surplus for platform owners. But English-only moderation and under-resourced language support are more naturally explained by ordinary cost minimization and data scarcity; the Myanmar/Ethiopia and WhatsApp cases are enforcement and policy failures. The essay never rules out these alternative mechanisms or identifies the specific beneficiary in each case. It also insists, correctly, that \"the Global Majority is many,\" which undercuts the idea of a single shared mechanism. That tension is acknowledged but not resolved. As a research claim, the root-cause thesis is underdetermined; as an interpretive frame for policy advocacy, it is workable.\n\nThe paper also presents no new data — it is a synthesis of prior work, including the authors' own, which is fine for this genre. No machine-checked proofs or reproducibility artifacts, obviously; that is not the point here.\n\nWho gets value from this: people in AI governance, content moderation policy, FAccT-type audiences, and anyone looking for a compact, well-referenced statement of the structural critique. It is a position piece, and a decent one.\n\nRecommendation: if the venue accepts essays, send it to peer review. A good referee will push the authors to sharpen the causal claim or reframe it as a diagnostic lens rather than a root cause, and that would make the piece stronger. Desk rejection would be a miss.","headline":"A clear, well-cited critical essay whose 'exploitation all the way down' framing is more an interpretive lens than a demonstrated root cause, but it deserves a serious referee as a policy intervention.","tokens_in":6890,"tokens_out":1426,"would_cite":false,"duration_ms":14000,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This essay argues that the common root of the online harms faced by users from the Global Majority is the continued exploitation and disregard for their well-being by large tech companies and outside states.","keywords":["content moderation","global majority","online harm","exploitative labor","AI governance","language inequality","digital colonialism","participatory design"],"falsifier":"A representative cross-platform audit of moderation outcomes across many Global Majority languages, controlling for how much each platform invests in moderating that language and how moderators are compensated, would settle the claim: if harm rates track resource levels rather than exploitative labor arrangements, the root-cause thesis fails.","tokens_in":5893,"feed_emoji":"🌍","tokens_out":7078,"duration_ms":54199,"temperature":0.7,"pith_summary":"This essay tries to establish that the many online harms experienced by people in the \"Global Majority\"—harmful content they see, traumatic moderation work they perform, and AI systems that exclude them—share a single root cause: ongoing exploitation and disregard for their well-being by large tech companies and by states outside the Global Majority. The authors argue that current moderation systems do not simply fail; they are built on a cycle in which the same communities are exposed to harm as users, hired under exploitative conditions to moderate that harm, and then excluded from the benefits of the models they helped train. They also claim that the blanket term \"Global Majority\" hides the very different social, political, and economic realities within these communities, so one-size-fits-all policies cannot work. The paper matters because it reframes policy debates from isolated moderation failures to a structural pattern with historical roots in colonialism, and it calls for interventions that center community needs.","feed_headline":"One root cause explains online harms for the global majority","feed_subtitle":"The paper traces moderation failures, labor abuse, and AI exclusion to one exploitative cycle.","key_machinery":"The carrying mechanism is the \"cycle of harm,\" a four-stage account that connects user-facing harm, moderator trauma and exploited labor, exclusion from technological benefits, and a false dichotomy between surveillance and neglect. The authors use this cycle, together with a critique of blanket terminology, to argue that exploitation is the root cause rather than a side effect. The cycle is what makes the argument cohere: each stage is treated as a manifestation of the same structural neglect, and the term \"Global Majority\" is examined as a lens that both mobilizes and obscures.","core_discovery":"The paper's central claim is that the common denominator tying the experiences of Global Majority people is \"the continued exploitation and disregard for well-being by Big Tech and states outside of the Global Majority.\" The authors present a cycle of harm with four linked faces: users are exposed to harmful content because moderation is under-enforced and under-resourced in their languages; people from these communities are hired as content moderators and subjected to traumatic, poorly compensated work; their labor is used to train models that do not serve their languages or are unavailable in their countries; and existing moderation alternatives force a false choice between being surveilled or being left unprotected. The essay further argues that this structure resembles colonial exploitation and that the term \"Global Majority,\" while useful for reclaiming power, obscures internal diversity that makes uniform policies ineffective. The remedy proposed is not to extend current Global Minority policies to previously ignored communities, but to support community-driven interventions and participatory design.","pith_inferences":["A testable consequence the paper does not spell out: if the exploitation thesis is right, then a cross-platform audit controlling for language and region should find that harm rates and moderation failures are predicted by the exploitativeness of labor practices (e.g., third-party subcontracting) rather than by raw resource levels.","The cycle-of-harm account suggests a concrete design target: platform policies that combine community-owned moderation with transparent appeals could break the false dichotomy between surveillance and neglect, a direction the paper gestures at but does not fully develop.","The \"Global Majority\" terminology critique implies that future research should disaggregate outcomes by country, class, and ethnicity; a useful next step would be to build a comparative dataset of moderation harms across several Global Majority communities to test which layers of difference matter most.","The paper's colonial-continuity analogy could be extended to labor externalization in other industries; if the moderation case follows the same pattern as outsourced manufacturing, we would expect repeated cycles of subcontracting, union-busting, and relocation to lower-cost regions."],"forward_implications":["If exploitation is the root cause, then improving automated detection models for more languages will not solve the problem by itself; the labor conditions and policy enforcement structures also need to change.","Hiring moderators from the Global Majority without fair compensation and support cannot count as inclusion; it extends the same cycle of harm.","Extending existing Western-derived moderation policies to Global Majority communities would likely reproduce harm, since the term hides important differences in class, ethnicity, and political context.","Effective interventions should start from community organizing and participatory design rather than from platform-centric policy.","The surveillance-versus-neglect framing implies that any moderation solution that increases surveillance without addressing exploitation will fail the communities it claims to protect."],"supporting_citations":[{"why":"Supplies the statistic that 89.2% of human moderators operate in English and the platform's stated commitment to positive experiences, which the paper uses to show the gap between claim and practice.","marker":"Google, 2023"},{"why":"Provides the statistic that in 2020, 87% of moderation effort was spent on US posts despite 90% of users living outside the US and Canada, establishing America-centric moderation.","marker":"Tworek, 2021"},{"why":"Documents the dire working conditions, mass layoffs, and union-busting threats faced by third-party content moderators in Kenya, supporting the exploited-labor stage.","marker":"Perrigo, 2022"},{"why":"Presents the Ethiopian YouTube user study showing policy-violating content in Amharic and the different experiences of migrant domestic workers and a US-based software engineer.","marker":"Nigatu & Raji, 2024"},{"why":"Supplies examples of cultural nuances in moderation, such as the removal of a photo of village children swimming in a pond, showing that one-size-fits-all policies fail.","marker":"Shahid & Vashistha, 2023"},{"why":"Frames platform power as a form of digital colonialism, supporting the paper's historical-continuity claim.","marker":"Kwet, 2019"},{"why":"Provides the \"bottom billion petri dishes\" characterization, illustrating how Global Majority communities are treated as low-consequence testbeds.","marker":"Sambasivan et al, 2021"}],"fun_headline_variants":["Exploitation is the root cause of global majority online harm","One exploitative cycle links moderation, labor, and AI exclusion","Global majority harms trace back to a single exploitative cause","The exploitation cycle: why global majority users can't win online"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the cited cases and statistics are representative of the broader Global Majority; if those examples are atypical, or if under-moderation stems more from resource constraints than exploitative intent, the causal claim loses its basis.","fun_headline_variants_meta":{"raw":{"variants":["Exploitation is the root cause of global majority online harm","One exploitative cycle links moderation, labor, and AI exclusion","Global majority harms trace back to a single exploitative cause","The exploitation cycle: why global majority users can't win online"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000845,"raw_usage":{"total_tokens":3611,"prompt_tokens":811,"completion_tokens":2800,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":427,"completion_tokens_details":{"reasoning_tokens":2731}},"tokens_in":427,"tokens_out":2800,"duration_ms":19267,"temperature":1.0,"reasoning_tokens":2731,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T10:44:44.770007+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A representative cross-platform audit of moderation outcomes across many Global Majority languages, controlling for how much each platform invests in moderating that language and how moderators are compensated, would settle the claim: if harm rates track resource levels rather than exploitative labor arrangements, the root-cause thesis fails.","supporting_citations":[],"review_version":1}