{"id":"e1510cf8-7f18-47d8-834d-8b390c6dd412","arxiv_id":"2502.03940","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Dominant media and corporate narratives about generative AI in creative work promote automation, efficiency, concept over execution, artifact over process, and short-term skills, marginalizing artists' perspectives.","lead":"This study analyzed 188 online media articles and product descriptions about generative AI in the arts. It finds that the dominant narratives frame AI as freeing creativity from labor, time, and skills, and that these stories are mostly told by tech companies and journalists, not artists.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The paper identifies five themes in its corpus, but the central claim that these are the 'dominant narratives' is never operationalized; no prevalence measure or systematic test distinguishes dominant from merely present.","rationale":"The reader's weakest assumption focuses on corpus representativeness: whether first-page search results restricted to sources with more than five articles capture dominant narratives rather than a tech-promotional slice. That is a genuine concern, but I see an even more direct gap in the argument. The paper's own analysis is purely qualitative: it identifies five themes through inductive coding and illustrates them with selected quotes, but it never counts how often the themes occur, nor does it systematically compare them with counter-narratives. The central claim is about what is 'dominant' in online media discourse, so the paper must supply some evidence of prevalence or cultural salience. Without a frequency measure or a structured comparison, the phrase 'dominant narratives' is an interpretive assertion rather than an empirical finding. This is load-bearing because if the five themes are only present in a minority of the corpus, or if counter-themes are equally common, the paper's conclusion about cultural influence is overstated. I partially agree with the reader because the corpus-construction choices exacerbate the problem, but the inferential gap from themes to dominance would remain even with a different corpus. The paper has strengths: it is transparent about its subjective method, it provides rich illustrative examples, and it situates the findings in a broader critical literature. These strengths support a conditional acceptance, but the dominance claim needs either additional evidence or a more modest framing. I therefore keep the reader's CONDITIONAL verdict unchanged.","tokens_in":14366,"tokens_out":4582,"duration_ms":47951,"concrete_test":"Take the final corpus of 188 articles plus 19 product descriptions and have two independent coders, blinded to the paper's conclusions, code each item for the presence or absence of each of the five value themes (automation over manual work, efficiency over exploration, concept over execution, artifact over process, short-term over long-term skills) and for any explicitly contrasting values (e.g., craft, exploration, process, long-term skill development). Report the proportion of items instantiating each theme and the proportion instantiating at least one counter-theme, with inter-coder reliability (e.g., Cohen's kappa). If the five themes together appear in a clear majority of items and counter-themes appear in a small minority, the dominance claim is supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim, in the abstract and throughout, is that the analyzed discourse 'promotes' specific values and that these are the 'dominant narratives' around generative AI and creative work. The method (§3) constructs a corpus of 188 articles plus 19 product descriptions, then uses inductive coding and thematic analysis (§3.2) to arrive at five themes. The findings (§4) present vivid examples for each theme, but the paper never reports how many articles instantiate each theme, whether any articles contain contrary narratives, or how the authors determined that these themes, rather than others, are dominant. The word 'dominant' is used as an interpretive label, not as a measured property. For instance, §5 states 'we have analyzed the dominant narratives' without any quantitative or systematically comparative evidence. The closest empirical support is the corpus-construction decision in §3.2 to keep sources publishing more than five relevant articles, which ensures that frequently covered topics dominate the sample, but it does not establish that the five values themselves are dominant within those articles. This is load-bearing because the paper's contribution is to identify the narratives that are shaping cultural values; if the themes are merely present in a subset of tech-promotional sources, the strong conclusion about the 'dominant' discourse overreaches. The paper is transparent about the subjectivity of its qualitative analysis (§5), but transparency does not close the gap between 'we identified these themes' and 'these are dominant.' A related but secondary issue is that the search queries in §3.1 include 'products' and 'companies,' and §3.2 explicitly adds company product descriptions, which likely skews the corpus toward vendor marketing messages; however, even with a perfectly representative corpus, the lack of a prevalence measure would remain a problem for the dominance claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies how generative AI is framed in online media and product descriptions with respect to creative work. The authors construct a corpus of 188 articles and 19 product descriptions from search-engine queries, conduct inductive coding, and identify five explicit values in the discourse: automation over manual work, efficiency over exploration, concept over execution, artifact over process, and short-term over long-term skills. They argue these values form a dominant narrative in which creativity is freed from material realization through automation, and that this narrative primarily originates from technology companies and journalists rather than artists, with consequences for cultural values and artistic labor.","tokens_in":14633,"tokens_out":3667,"duration_ms":33243,"significance":"The paper addresses an important and timely topic and makes a credible interpretive contribution by synthesizing a corpus of contemporary media coverage and product positioning around generative AI in the arts. Its strengths include a transparent description of corpus construction, the inclusion of product descriptions alongside editorial articles, and explicit acknowledgement of the subjectivity of the qualitative analysis. If the empirical claims are supported, the five-theme framework could be a useful starting point for future quantitative content analysis and for critical scholarship on AI narratives. The paper does not provide machine-checked results or quantitative predictions; its contribution is qualitative and interpretive.","major_comments":[{"comment":"The central claim that the analyzed discourse contains 'dominant narratives' is not operationalized. The paper reports 188 articles and 19 product descriptions (§3.2) and a codebook of 102 codes, but it never reports how many articles instantiate each of the five themes, whether any articles contain contrary or countervailing narratives, or how the authors decided that these five themes, rather than others, are dominant. In the limitations paragraph (§5) the authors acknowledge that the process 'is not systematic and calls on our subjectivity,' which directly undercuts the word 'dominant' as used in the abstract and throughout. To support the claim, the authors should either add a quantitative or systematic prevalence measure, or revise the claim to say that these are recurrent themes in a sample of frequently covered tech-oriented sources.","section":"Abstract and §5"},{"comment":"The corpus construction choices may systematically bias the sample toward technology-promotional content: the authors take the first 90 results per query from Google and Bing, then keep only sources that published more than five relevant articles over four years, and finally supplement the corpus with product descriptions from generative AI companies. This procedure makes it likely that large media outlets and company materials—rather than artist voices or critical publications—dominate the sample. The paper's broad conclusion that these are the 'dominant narratives' across online media outlets is therefore stronger than the sampling design can support. I ask the authors to report the distribution of sources by type, discuss how the filtering rules interact with the research question, and consider a sensitivity check on the >5 threshold, or to restrict the claims accordingly.","section":"§3.1–§3.2"},{"comment":"The analysis is not auditable as presented: the authors state that 102 codes were grouped into five themes, but no codebook is provided, no inter-coder reliability is reported, no definition of theme boundaries is given, and Table 1 summarizes only the final themes with their implicit narratives and impacts. A reader cannot tell whether the five themes were derived from the corpus or imposed by the authors' prior interpretive framework. I recommend including a supplementary codebook with representative quotes per code, documenting the grouping procedure, and, ideally, adding a second coder or a member-checking step so that the theme derivation can be independently assessed.","section":"§3.2 and Table 1"},{"comment":"The discussion's third point, that 'artists are generally absent from the discourse,' is an important and potentially falsifiable claim, but the paper does not provide evidence about the authorship of the 188 articles beyond mentioning a few exceptions such as Ted Chiang. Since this claim supports the paper's argument about who controls the narrative, the authors should report the proportion of articles authored by artists, creative practitioners, technologists, or journalists, or explicitly code authorship in the analysis.","section":"§5"}],"minor_comments":[{"comment":"The keyword 'Genrative AI' is misspelled and should be 'Generative AI.'","section":"Keywords"},{"comment":"In the sentence about racial biases, 'raciable' should be 'racial.'","section":"§2.1"},{"comment":"The phrase 'date is between 2020 and Today' should name the actual cutoff date; also 'Bing As request keywords' has a spacing and formatting error.","section":"§3.1"},{"comment":"The quoted statement from RunwayML, 'bring research into production within weeks instead of years,' is presented without a date or URL, making it hard to verify; please add retrieval dates and URLs for all product-description quotes.","section":"§4.2"}],"recommendation":"major_revision","confidential_remarks":"The paper's empirical basis is thinner than its framing suggests, but the topic is well suited to the journal and the interpretive contribution is potentially valuable. The main revision challenge is to align the 'dominant narratives' claim with the evidence actually presented; a supplementary codebook and prevalence counts would go a long way. I would not recommend rejection, as the issue is fixable through additional reporting or reframing."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis paper is worth a look. It analyzes 188 media articles plus 19 product descriptions about generative AI and creative work, and synthesizes five values: automation over manual work, efficiency over exploration, concept over execution, artifact over process, and short-term over long-term skills. That synthesis is the real contribution. It gives researchers a compact vocabulary for talking about how tech companies and mainstream media frame AI creativity, and the examples are well chosen.\n\nThe method is transparent about its subjectivity, and the authors do not oversell the reliability of their coding. They report 102 codes grouped into five themes, acknowledge the process is not systematic, and cite relevant prior work.\n\nThe soft spot is the word 'dominant.' The paper claims to have analyzed 'the dominant narratives' but never measures prevalence. There is no codebook, no inter-coder reliability, no count of how many articles instantiate each value, and no comparison to contrary narratives. The corpus construction filters to sources that published more than five relevant articles, which skews toward outlets that cover the topic repeatedly (often tech-promotional), and the product descriptions are essentially vendor marketing. So the evidence supports 'values present in this corpus,' not necessarily 'dominant narratives' in any broader sense. The stress-test note has this right.\n\nThat said, the central critical argument does not depend on the word 'dominant.' If the authors loosened the claim to 'prevalent in the discourse we sampled' or 'frequently mobilized by commercial and mainstream sources,' the article would still be a useful contribution. The analysis of how each value carries implicit assumptions about creative work is thoughtful and grounded in the material.\n\nWho is this for? Researchers working on AI narratives, creative labor, or cultural studies of technology. It would be a reasonable paper in a venue like AI & Society or CSCW. For peer review, I would send it out, but I would ask the authors to either provide more transparency about theme prevalence or to moderate the dominance claim. With that revision, it is a solid conceptual mapping.\n\nRecommendation: deserving of serious review; the framing needs adjustment, not the whole edifice.","headline":"A useful five-value map of AI-in-creative-work discourse that overstates its case by calling the values 'dominant' without measuring prevalence.","tokens_in":15200,"tokens_out":1937,"would_cite":true,"duration_ms":18899,"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":"Online discourse about generative AI and creative work is built by tech firms and media, not artists, and promotes creativity as disembodied ideation freed from human labor.","keywords":["generative AI","creative work","AI narratives","media discourse analysis","automation","democratization of creativity","artistic labor","cultural values"],"falsifier":"Re-run the search protocol with artist-authored outlets or with queries grounded in artists' own vocabulary, such as 'AI art labor' or 'AI and craft,' and count how often the efficiency and automation frames appear; if they no longer dominate, the claim that the dominant discourse pushes disembodied creativity would be weakened.","tokens_in":14203,"feed_emoji":"🎨","tokens_out":8012,"duration_ms":65988,"temperature":0.7,"pith_summary":"This paper tries to establish that the public story about generative AI in creative work is a constructed narrative, not a neutral description: it is told mostly by technology companies and mainstream media, not by artists. Analyzing 188 media articles and 19 product descriptions, the authors find the dominant discourse presents creativity as an idea separated from its material execution, with automation speeding production and skill reframed as a barrier rather than a craft. If this is right, public debate about AI art is being shaped by the economic interests of tool vendors, and the values embedded in the discourse—speed, output volume, low barriers—are reshaping what counts as creative work and who can claim to do it.","feed_headline":"Media stories sell AI as creativity without the labor","feed_subtitle":"Analysis of 188 articles and 19 product pages finds tech firms and media, not artists, set the frame for what AI art means.","key_machinery":"The central analytic device is the narrative as a cultural artifact: a story that conveys a point of view or set of values. The paper operationalizes this by assembling a corpus of 188 articles from 19 online sources plus 19 product descriptions through 18 search-engine queries, filtering to sources that published more than five relevant articles, and coding the material inductively into 102 codes grouped into five themes. The five value oppositions—automation over manual work, efficiency over exploration, concept over execution, artifact over process, short-term over long-term skills—are the mechanism that carries the argument, translating scattered media language into a coherent value system with stated impacts on creative work.","core_discovery":"The central claim is that dominant online narratives about generative AI in creative work are anchored in five explicit values: automation over manual work, efficiency over exploration, concept over execution, artifact over process, and short-term over long-term skills. Each value carries implicit assumptions: creativity is a problem-solving process that can be rationalized and automated; time to output is the measure of creative value; craft and embodied skill are obstacles to democratization; and the final artifact matters more than the process that produces it. Because these narratives are told mainly by journalists and technology actors rather than by artists, the paper argues the discourse corresponds to a techno-positivist vision—a faith that technological progress is the measure of human progress—and asserts power over the creative economy and culture. Artists' own voices, when they appear, point the other way, valuing exploration, accidents, and small-scale choices.","pith_inferences":["The paper does not test this, but the same five value oppositions could be applied to discourse about AI in writing, coding, or design work, predicting that efficiency and concept-over-execution frames dominate those fields too.","A natural extension would be a survey experiment measuring whether exposure to efficiency-framed AI marketing shifts artists' stated willingness to engage in open-ended exploration.","The corpus is drawn from English-language search results on US and Europe-focused engines; applying the same coding scheme to non-English or artist-run media could reveal whether these narratives are a regional phenomenon tied to tech-industry communication.","The paper implies but does not analyze a tension: the 'democratization via skill removal' narrative coexists with a growing market in prompt-engineering expertise, suggesting the skill is displaced rather than eliminated."],"forward_implications":["Public perception of AI art will keep centering speed and output volume, pushing process-oriented artistic practice to the margins of what is recognized as creative work.","The democratization framing that equates access with removing skill can weaken support for arts education and devalue craft, making creative careers harder to sustain for people without privileged starting points.","Freelancers and gig creatives will face pressure to meet AI-speed production benchmarks, which can lower prices and deepen inequality within the creative sector.","When artists do enter the discourse, their counter-narratives emphasize exploration, accidents, and small-scale choices, so the gap between maker experience and vendor messaging is likely to grow.","If narratives shape policy, the efficiency and automation frames may steer funding and regulation toward tool adoption rather than toward protections for artistic labor."],"supporting_citations":[{"why":"Defines narratives as cultural artifacts that carry points of view and values, grounding the paper's object of analysis.","marker":"[3]"},{"why":"Supplies the concept of 'enchanted determinism' to explain how tech narratives deflect accountability.","marker":"[7]"},{"why":"Establishes that dominant AI narratives are polarized between threat and solutionism, motivating the search for missing stories.","marker":"[15]"},{"why":"Provides evidence that artists value unpredictability, accidents, and process in AI art, the counterpoint to efficiency narratives.","marker":"[10]"},{"why":"A prominent artist-authored counter-narrative arguing that small-scale choices matter in art, used against concept-over-execution.","marker":"[14]"},{"why":"Documents artists' fears and anxieties about AI, supporting the claim that dominant narratives reinforce those fears.","marker":"[34]"},{"why":"Traces democratization rhetoric to cyberculture, used to critique the empowerment and access framing.","marker":"[61]"},{"why":"Supplies the critique that democratizing tools can reinforce exclusion, used to question the democratization narrative.","marker":"[47]"}],"fun_headline_variants":["AI creativity hype sells automation as democratization","Media and tech, not artists, set AI art's narrative","Study: AI discourse prioritizes efficiency over exploration","Five values define how AI creative work is framed","AI art stories push concept over craft and process"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper assumes that a corpus built from the first 90 search results of 18 queries on two search engines, filtered to sources publishing more than five relevant articles, captures the dominant public narratives rather than a self-selected slice of tech-promotional media.","fun_headline_variants_meta":{"raw":{"variants":["AI creativity hype sells automation as democratization","Media and tech, not artists, set AI art's narrative","Study: AI discourse prioritizes efficiency over exploration","Five values define how AI creative work is framed","AI art stories push concept over craft and process"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000679,"raw_usage":{"total_tokens":3051,"prompt_tokens":876,"completion_tokens":2175,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":492,"completion_tokens_details":{"reasoning_tokens":2102}},"tokens_in":492,"tokens_out":2175,"duration_ms":19318,"temperature":1.0,"reasoning_tokens":2102,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T00:09:21.521828+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the search protocol with artist-authored outlets or with queries grounded in artists' own vocabulary, such as 'AI art labor' or 'AI and craft,' and count how often the efficiency and automation frames appear; if they no longer dominate, the claim that the dominant discourse pushes disembodied creativity would be weakened.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines narratives as cultural artifacts that carry points of view and values, grounding the paper's object of analysis."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the concept of 'enchanted determinism' to explain how tech narratives deflect accountability."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"A prominent artist-authored counter-narrative arguing that small-scale choices matter in art, used against concept-over-execution."},{"cited_title":"Jiang, Lauren Brown, Jessica Cheng, Mehtab Khan, Abhishek Gupta, Deja Workman, Alex Hanna, Johnathan Flowers, and Timnit Gebru","cited_arxiv_id":null,"evidence_quote":"Documents artists' fears and anxieties about AI, supporting the claim that dominant narratives reinforce those fears."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Traces democratization rhetoric to cyberculture, used to critique the empowerment and access framing."},{"cited_title":"Mowery and Nathan Rosenberg","cited_arxiv_id":null,"evidence_quote":"Supplies the critique that democratizing tools can reinforce exclusion, used to question the democratization narrative."}],"review_version":1}