{"id":"1cbb24a1-af7f-4b89-9172-aedf2059a383","arxiv_id":"2412.04575","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Queer and disabled fiction writers report greater pessimism about generative AI's industry impact than other writers and describe precarity, ethical objections, and collective boundary-setting strategies.","lead":"A survey of 137 fiction writers, focused on queer and disabled writers, finds these writers are more pessimistic than others about how generative AI will affect publishing, though most writers in both groups are pessimistic. The study gives a qualitative account of why: AI is seen as unhelpful, ethically unacceptable because of training data, and a threat to an already precarious industry.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The comparative pessimism claim is threatened by differential snowball recruitment: the gap may reflect the first author's online queer/disabled network rather than marginalization per se; the paper's own §9.2 limitation is acknowledged but untested.","rationale":"The reader's weakest assumption — that the snowball sample represents queer and disabled professional fiction writers generally — is indeed the most load-bearing concern. I go further by specifying the internal-validity threat: the two comparison groups may be differentially recruited from the first author's queer/disabled online network, which could create an artificial pessimism gap. This is a testable proposition, and the paper already contains the seed of the test (recruitment through Bluesky vs. other channels) but does not execute it. My proposed sensitivity analysis would settle whether the gap survives outside the first author's network. The statistical fragility (multiple comparisons, borderline p-values) is secondary but reinforces the need for such a test. Because the paper is explicitly preliminary and honestly states its limitations, the conditional verdict remains appropriate; the new contribution is the concrete test that should be a condition of acceptance.","tokens_in":24448,"tokens_out":9352,"duration_ms":92079,"concrete_test":"Perform a sensitivity analysis using only respondents recruited through non-Bluesky channels (or, if that data is absent, re-run the Mann-Whitney tests excluding the first author's direct followers). If the group difference in pessimism on the 'Effects of genAI' and the 'Making money (all, next five)' items is no longer significant, the selection-bias concern lands; if it persists, the finding is more robust. This directly tests the §9.2 limitation without collecting new data.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is comparative: queer/disabled writers are markedly more pessimistic than others about AI's impact on their industry. The quantitative evidence consists of Mann-Whitney tests in Table 3, but the sample was recruited via snowball through the first author's own professional networks (§3). Because the first author is a queer/disabled fiction writer, their Bluesky account is likely followed by a self-selected, highly online segment of queer/disabled writers already engaged in AI discourse, potentially more uniformly pessimistic. The control group, drawn partly from general writing Slacks/Discords, may be less engaged. This differential selection directly threatens the internal validity of the comparison: the observed difference could be an artifact of recruitment channel rather than a consequence of queer/disabled identity. The paper acknowledges this in §9.2 ('may overemphasize the importance of online discourse... may be more polarized than a fully random sample') but does not test it. The post-hoc decision to split the sample after observing the overrepresentation (§3.1) compounds the issue: with 12 tests, only 4 reach p<0.05, three of which are near the threshold; the strongest (p=0.0035) is a global 'Effects of genAI' item, while the abstract's industry-specific claim relies on p=0.045 and p=0.049, which would not survive a Bonferroni correction. Unless the group difference is shown to persist in less online or differently recruited samples, the central claim is not yet securely established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This mixed-methods paper surveys 137 professional fiction writers recruited by snowball sampling, with an overrepresentation of queer and disabled writers. The authors split the sample into 54 queer disabled writers and 75 non-queer/non-disabled writers, report Mann-Whitney comparisons on Likert-scaled attitudes toward the fiction-writing industry and generative AI, and develop a grounded-theory analysis of 16 queer/disabled writers' free-text responses. The paper's central claim is that queer and disabled writers are markedly more pessimistic than other writers about the impact of AI on their industry, and that this pessimism is grounded in existing precarity, ethical objections to copyrighted training data, and a view of AI writing as lacking intentional 'soul.' The authors describe individual and collective strategies of boundary-setting, momentum-building, and public backlash, and offer policy recommendations.","tokens_in":24730,"tokens_out":5476,"duration_ms":57273,"significance":"If the comparative pessimism claim holds, the paper provides valuable evidence that generative AI's harms are experienced disproportionately by already-marginalized writers, and it reframes online backlash as a material self-protection strategy rather than mere polarization. The study's strengths include its genuinely understudied population, honest disclosure of recruitment and analysis limitations, rich qualitative material, and engagement with existing professional-writer surveys and AI-harm literature. The contribution is potentially important for conversations about AI labor impacts and for designers and policymakers. However, the quantitative support for the headline comparative claim is more fragile than the abstract suggests, and the recruitment strategy introduces a direct threat to the internal validity of the group comparison.","major_comments":[{"comment":"The comparative quantitative claim rests on 15 Mann-Whitney tests with no multiple-comparison correction and no reported effect sizes. Only one p-value (p=0.0035 for the global 'Effects of genAI' item) is close to the Bonferroni threshold of 0.05/15≈0.0033 and does not survive it; the industry-specific comparisons cited in the abstract (p=0.049, p=0.045, and p=0.033) clearly do not survive even a mild correction. The word 'markedly' in the abstract therefore outruns the evidence as reported. Please report effect sizes with confidence intervals (for example Cliff's delta or rank-biserial correlation), apply or justify a correction for multiple testing, and either soften the 'markedly' claim or show that the aggregate pattern is robust.","section":"Table 3 / §4"},{"comment":"The differential recruitment design threatens the internal validity of the central comparison. Recruitment began from the first author's own Bluesky account, which is likely followed by a self-selected, highly online segment of queer and disabled writers already engaged in AI discourse, while the control group was drawn more heavily from general writing Slacks and Discords. The paper's own §9.2 acknowledges that the sample may be more polarized than a random sample, but it does not test the alternative explanation that the observed group difference is an artifact of recruitment channel. Please include a sensitivity analysis using any available source-of-recruitment information, or otherwise demonstrate that the pessimism gap persists after accounting for recruitment channel.","section":"§3 and §9.2"},{"comment":"The comparison groups were defined after observing the overrepresentation of queer and disabled writers in the sample, and the hypothesis of greater pessimism was then tested on the same data. This post-hoc decision inflates the evidentiary value of the p-values in Table 3. The paper is transparent about the decision, which is commendable, but the statistical language should be adjusted to treat the finding as exploratory rather than confirmatory. At a minimum, the Discussion and Limitations should state explicitly that these p-values are not corrected for the data-dependent choice of grouping and hypothesis.","section":"§3.1"},{"comment":"The authors themselves note that queer and disabled writers show no significant differences from the control group on any of the individual-career questions (self rows in Table 3), despite dramatically lower reported income. The abstract's phrase 'impact of AI on their industry' is technically aligned with the significant industry-wide items, but the gap between individual- and industry-level results deserves more interpretive attention in the main text, including a discussion of whether the 'markedly' framing is warranted when the significant differences are small in magnitude and confined to industry-wide perceptions.","section":"§4 / Table 3"}],"minor_comments":[{"comment":"The research question contains a typo: 'navigate and mange' should be 'navigate and manage.'","section":"§1"},{"comment":"There is a grammatical error: 'an writer' should be 'a writer.'","section":"§6.2"},{"comment":"The phrase 'of of harm to their mental health' should be 'harm to their mental health.'","section":"§8.3"},{"comment":"The word 'longditudinal' should be 'longitudinal.'","section":"§9.3"},{"comment":"The final row of Table 2 reports percentages (19%, 38%, 9%, 53%, 4%) that sum to 123%, which suggests an alignment or transcription error; please verify and correct the table.","section":"Table 2"},{"comment":"The reference '(in Fiction [2023])' should be formatted consistently with the full name 'Humanity in Fiction,' and the same consistency check is needed for other civil-society survey citations.","section":"References"},{"comment":"Since the grounded-theory analysis is explicitly not saturated and covers only 16 of the 54 queer/disabled writers, the sentence in §5 that 'our confidence in this preliminary theory is strong' should be paired with an explicit reminder of the unsaturation and of the fact that five control-group analyses were set aside rather than integrated.","section":"§3.2 and §5"}],"recommendation":"major_revision","confidential_remarks":"This is an unusually honest and useful preliminary study, and I do not see evidence of bad faith in the treatment of limitations. The central difficulty is that the abstract's comparative claim is stronger than the statistical analysis currently supports, and the recruitment-based selection threat is acknowledged but not addressed empirically. Both problems are fixable with additional analysis and revised framing, so a major-revision path is appropriate rather than rejection. Given the paper's explicitly preliminary status, the editor may also wish to invite the authors to describe this as an exploratory report with confirmatory analysis deferred to the planned full study."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a worthwhile qualitative study of an overlooked group, but the headline quantitative claim is not yet as solid as the abstract implies.\n\nWhat's new: prior surveys (Author's Guild, ITW, Humanity in Fiction) measured writer concern at scale, and the qualitative work on genAI covered visual artists, game workers, and designers. This paper centers queer and disabled fiction writers and gives them room to explain why they oppose genAI: precarity, the \"soul\" concept, and collective boundary-setting as self-protection rather than mere polarization. That material is genuinely missing from the literature, and the grounded theory is plausible and well supported by quotes. The authors are transparent about the recruitment path, the lack of saturation, and the fact that the control group's qualitative responses remain unanalyzed. The policy section is measured. These are real strengths.\n\nWhere it wobbles: the abstract says 'markedly more pessimistic,' but the statistical support is four significant Mann-Whitney tests out of seventeen, with no multiple-comparison correction and no effect sizes. Three of the four p-values are in the 0.03–0.05 range and would not survive a Bonferroni correction; the strongest (p=0.0035) is a general 'effects of genAI' item rather than an industry-specific one. The individual-career questions show no significant difference, which undercuts the 'markedly' framing. The income gap is described as significant but no test is reported. The snowball recruitment through the first author's own queer/disabled network on Bluesky is a genuine threat to the comparative claim: the gap could be a selection effect. The paper acknowledges this in §9.2 but does not test it. These are addressable, not fatal.\n\nVerdict: the qualitative core is a real contribution, and the authors are clearly thinking carefully. Send it to peer review, but with the expectation of major revision: either soften the comparative claim or provide robustness checks. This is exactly the kind of paper a good referee can push into a stronger form. I'd take it to reading group, though not as a model of quantitative practice.","headline":"A genuinely useful qualitative study of queer and disabled fiction writers, burdened by a comparative headline that its small, snowball-recruited sample and uncorrected tests do not yet support.","tokens_in":25227,"tokens_out":3297,"would_cite":true,"duration_ms":99519,"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":"A survey of 137 professional fiction writers finds queer and disabled authors markedly more pessimistic than others about generative AI, and argues that their backlash is a self-protective collective response to deepening precarity.","keywords":["generative AI","queer studies","disability studies","fiction authors","publishing industry","precarity","grounded theory","collective action"],"falsifier":"Conduct the same survey on a probability-based sample of professional fiction writers drawn outside AI-discussion-heavy social platforms, for example through a national writers' register, professional organizations' membership rolls, or in-person conventions, and compare the queer and disabled responses to the control group. If the significant pessimism gap on the four Mann-Whitney questions disappears, or if the free-text responses do not reliably organize around precarity, soul, and boundary-setting, the central claim as stated would not survive.","tokens_in":24255,"feed_emoji":"📚","tokens_out":11107,"duration_ms":106419,"temperature":0.7,"pith_summary":"The paper reports a mixed-methods survey of 137 professional fiction writers, with qualitative analysis centered on the 54 who identify as both queer and disabled. Its central claim is that generative AI deepens an already precarious publishing industry, and that queer and disabled writers, who earn a lower median writing income despite similar experience, feel and expect its harms more acutely than other writers, although pessimism is the majority view in both groups. The paper argues that writers' opposition to genAI is not reflexive polarization but a self-protective collective strategy, grounded in two objections: the technology is useless for writing as intentional self-expression, and it is unethical because it trains on copyrighted work without consent or compensation. The study contributes first-hand accounts from a marginalized population of creative workers to the debate over genAI's industry effects.","feed_headline":"Queer and disabled fiction writers see more AI harm ahead","feed_subtitle":"A 137-writer survey ties their deeper pessimism to lower income and a backlash rooted in precarity, not polarization.","key_machinery":"The machinery is a grounded-theory qualitative analysis built around three concepts the writers themselves use. Precarity is the central category: the industry's pre-existing instability, including low advances, shrinking publisher and magazine counts, flooding of self-published books, self-marketing burdens, and dependence on platforms that change their business models, against which every genAI effect is measured. Soul names what writers say AI cannot supply: conscious, intentional communication of the writer's own experience, which is why even useful AI output is rejected. Momentum is the one individual protective factor writers describe, the way past successes make future ones easier. The quantitative results are carried by Mann-Whitney U tests comparing Likert-scale responses between the queer disabled group and the control group, while the qualitative coding supplies the mechanism that explains the pessimism gap.","core_discovery":"The paper's core discovery, stated on its own terms, is that queer and disabled fiction writers are markedly more pessimistic than non-queer and non-disabled writers about the effects of generative AI on the fiction industry, and that this pessimism tracks measurable precarity. The 54 queer disabled writers in the sample had a median writing income of 695.50 US dollars versus 3000 US dollars for the 75-writer control group, with similar age and experience, and 72% rated genAI's effects as very negative, against 52% in the control group; Mann-Whitney U tests found significant differences on four questions, all about conditions for writers in general rather than the respondent's own career. Qualitatively, the free-text responses organized around precarity: genAI floods self-publishing and short-fiction markets with low-quality submissions, introduces AI-related contract clauses, threatens cover-art and manuscript-evaluation work, and forces writers into extra labor to avoid AI-infused software. Writers describe genAI as soulless, meaning incapable of the intentional, effortful communication they value, and as plagiarism because of unlicensed training data, and they respond with personal boundaries and public backlash that have already pressed some publishers to retract AI plans. The paper interprets this backlash as a conscious, self-protective strategy by a vulnerable population, not as identity-driven polarization.","pith_inferences":["The paper does not test whether writers' rejection of genAI would soften under a fully consensual, compensated training-data regime; the strength of the soul objection suggests it would not, so technical fixes like watermarking would address only part of the harm described.","If precarity rather than identity is the operative mechanism, the qualitative theory may generalize to other freelance creative workers who labor alone and depend on copyright, such as illustrators, translators, and editors, a hypothesis the authors' planned control-group analysis can partially check.","Recruiting the same survey from venues where AI is not a central topic would be a direct test of the sample-dependence concern the authors flag; a large drop in the very negative responses would show that the result partly reflects the recruitment channel.","Because writers already refuse to work with AI-using publishers and platforms, genAI adoption may split the market into explicitly human-labelled work and openly AI-produced work, rather than uniformly changing one industry."],"forward_implications":["Backlash against genAI by fiction writers should be read as a self-protective labor strategy, since informal social pressure has already led some publishers to retract AI plans.","Marginalized writers are the ones most exposed to genAI's harms, so technology deployment that ignores material conditions in publishing would widen existing income and stability gaps.","Consent-and-compensation regimes for copyrighted training data would directly address the ethical objection most writers in the study raise.","Claims that genAI will democratize creativity run against the barriers writers actually name: precarity, low income, and uncertain career prospects, not lack of skill or tools.","If soul is the source of value in fiction for these writers, then labelling AI-generated work and watermarking would help readers who want human-written fiction find it."],"supporting_citations":[{"why":"This industry income survey supplies the demographic and income benchmark against which the sample's representativeness is judged.","marker":"Guild [2023a]"},{"why":"This methodology text supplies the grounded-theory procedures the qualitative analysis follows.","marker":"Corbin and Strauss [2015]"},{"why":"This prior account of genAI harms to artists supplies the intentional-communication view of art that the paper's soul concept extends.","marker":"Jiang et al. [2023]"},{"why":"This editor's report documents the flood of AI-generated submissions that forced a major magazine to close briefly, supporting the market-flooding claim.","marker":"Clarke [2023]"},{"why":"This security study shows that language models can memorize training text, grounding the plagiarism objection.","marker":"Carlini et al. [2021]"},{"why":"This journalistic investigation reveals that pirated books are included in training corpora, grounding the unauthorized-use objection.","marker":"Reisner [2023]"},{"why":"This news account of the Hollywood writers' strike supplies the collective-bargaining model that prose fiction writers lack.","marker":"Anguiano and Beckett [2023]"},{"why":"This news report documents AI-generated knock-offs of authors' books being sold on a major bookstore platform, supporting the direct-competition harm.","marker":"Tapper [2023]"}],"fun_headline_variants":["Queer disabled writers' AI pessimism tracks pay gap","Precarity drives deeper AI fears for marginalized writers","Marginalized fiction writers see genAI as a precarity threat","Backlash by queer, disabled writers tied to AI precarity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the writers recruited through the researchers' own social-media networks and writing communities represent queer and disabled professional fiction writers generally; if that snowball sample is systematically unrepresentative, for instance more enmeshed in online AI discourse than the wider population, then the pessimism gap and the qualitative theory could be an artifact of recruitment.","fun_headline_variants_meta":{"raw":{"variants":["Queer disabled writers' AI pessimism tracks pay gap","Precarity drives deeper AI fears for marginalized writers","Marginalized fiction writers see genAI as a precarity threat","Backlash by queer, disabled writers tied to AI precarity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000526,"raw_usage":{"total_tokens":2530,"prompt_tokens":923,"completion_tokens":1607,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":539,"completion_tokens_details":{"reasoning_tokens":1538}},"tokens_in":539,"tokens_out":1607,"duration_ms":10990,"temperature":1.0,"reasoning_tokens":1538,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T21:23:09.772562+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Conduct the same survey on a probability-based sample of professional fiction writers drawn outside AI-discussion-heavy social platforms, for example through a national writers' register, professional organizations' membership rolls, or in-person conventions, and compare the queer and disabled responses to the control group. If the significant pessimism gap on the four Mann-Whitney questions disappears, or if the free-text responses do not reliably organize around precarity, soul, and boundary-setting, the central claim as stated would not survive.","supporting_citations":[{"cited_title":"Basics of Qualitative Research: Techniques and Procedures for Developing Grounded Theory","cited_arxiv_id":null,"evidence_quote":"This methodology text supplies the grounded-theory procedures the qualitative analysis follows."},{"cited_title":"Editor's Desk: W ritten by a human","cited_arxiv_id":null,"evidence_quote":"This editor's report documents the flood of AI-generated submissions that forced a major magazine to close briefly, supporting the market-flooding claim."},{"cited_title":"Revealed: The authors whose pirated books are powering generative AI","cited_arxiv_id":null,"evidence_quote":"This journalistic investigation reveals that pirated books are included in training corpora, grounding the unauthorized-use objection."},{"cited_title":"How Hollywood writers triumphed over AI - and why it matters","cited_arxiv_id":null,"evidence_quote":"This news account of the Hollywood writers' strike supplies the collective-bargaining model that prose fiction writers lack."},{"cited_title":"Authors shocked to find AI ripoffs of their books being sold on Amazon","cited_arxiv_id":null,"evidence_quote":"This news report documents AI-generated knock-offs of authors' books being sold on a major bookstore platform, supporting the direct-competition harm."}],"review_version":1}