Precarity and Solidarity: Preliminary Results on a Study of Queer and Disabled Fiction Writers’ Experiences with Generative AI

1 Introduction

We surveyed 137 professional fiction writers about their experiences with generative AI (genAI). We used descriptive statistics for the survey's quantitative questions and a grounded theory approach to analyze free-text responses. Our primary research questions are: What effect is genAI currently having on professional fiction writers? And what strategies do such writers employ to navigate these effects?

Our sampling method overrepresented writers who are queer and/or disabled. Rather than correct this overrepresentation, we used it to analyze queer and disabled writers specifically. Our data show that queer and disabled writers are more precarious than other writers, and more pessimistic about the effects of genAI.

Qualitatively, queer and disabled writers provide a coherent account of their attitudes, with precarity being the central concept. Writers' attitudes must be understood in the context of their reasons for writing and the precarity which is worsened by genAI. Writers view genAI as unhelpful for their goals and as ethically unacceptable, and they engage in collective strategies to discourage its use.

Our study demonstrates the importance of understanding AI in its social context. Backlash against genAI among writers should be viewed as a conscious strategy of solidarity and self-protection engaged in by a vulnerable population.

2 RelatedWork

Even before the popularization of genAI, the field of computational creativity studied systems with both creative humans and computers in the loop (e.g., [Colton et al., 2016]). Since 2022, qualitative studies of professional artists working with genAI include Kaila et al. [2024], who studied artists in varied fields; Vimpari et al. [2023], who interviewed game industry professionals; and Uusitalo et al. [2024], who studied UX and industrial designers.

Ippolito et al. [2022] studied 13 professional fiction writers interacting with a story generation system. However, the study occurred prior to the popularization of LLMs; the broader social impacts of LLMs are beyond its scope.

Jiang et al. [2023] summarize genAI's harms to visual artists. Several of its themes agree with ours, including anger around copyright infringement and a philosophical argument - akin to our writers' concept of "soul," described below - against viewing genAI output as "art."

Non-scholarly surveys have also gathered fiction writers' opinions about genAI. These include established writers' groups, such as The Authors Guild [2023b] - as well as Humanity in Fiction [2023], which was created in response to the rise of LLMs. These surveys revealed concern about copyright and about broad effects of genAI. Our study adds qualitative data, emphasizing why fiction writers have their opinions and what they are doing about it. We also emphasize how marginalized writers are disproportionately affected. We aim to make these ideas more visible to computer scientists.

3 Methods and Data Collection

This research study was approved by our institutions' Research Ethics Boards, and the survey was conducted in May 2024. We recruited professional fiction writers using snowball sampling, advertising in writers' Slacks and Discords and on the Bluesky account of our first author. We define a "professional fiction writer" as anyone who has ever received money for a work of prose fiction.

Writers completed a questionnaire with quantitative and qualitative questions. They could give or withhold consent to be quoted directly. We received 137 full or partial responses. Each writer was paid CAD $20.

3.1 Demographics and Representativeness of the Data

Table ?? breaks down our writers by demographics. We compared these data to the 2023 Author's Guild survey on writer income [The Authors Guild, 2023a]. Our sample of writers is mostly consistent with the general population of writers. Both populations are predominantly white, and the yearly income from writing is similar. However, 63% of our writers are queer, and 62% are "neurodivergent or disabled." We believe this is an artifact of our sampling, as our first author is queer and disabled. We did not ask writers' nationality.

We realized that queer disabled fiction writers are an interesting population due to their intersectionality [Crenshaw, 2013]. The qualitative responses of the remaining writers (non-queer, non-disabled, or neither queer nor disabled) are set aside as a control group. We have 54 queer disabled writers and 75 other writers. (The remaining 8 writers did not answer enough questions and were discarded.)

3.2 Qualitative Analysis

We undertook qualitative analysis using grounded theory [Corbin and Strauss, 2015]. All three authors briefly reviewed all the qualitative responses. Then we selected writers to analyze through open coding. We approximated theoretical sampling by selecting writers based on developing theory. For example, did writers' experiences look different based on experience, race, gender, or other factors? When writers' quantitative responses differed - for instance, if they showed more optimism than others, or had used genAI when most had not - what did they have to say about those differences?

Identifying data were removed and the writers are referred to with alphanumeric designations. Five non-queer and/or non-disabled writers - B, F, G, H, and I - were analyzed before our decision to focus solely on queer disabled writers, and we do not present the concepts we derived from them here.

We repeated this process for 16 queer and disabled writers - A through U, minus the aforementioned five - until we no longer had specific questions that could be easily answered by theoretical sampling. At that point, the central concepts of our developing theory were apparent to us. We did not have saturation, but new insights were increasingly minor.

4 Quantitative Results

Our Likert-scale questions asked how easy it had been in the past five years, and how easy it would be in the next five, to do three things: make money, get publishers' attention, and get readers' attention. These questions were each asked twice - once for the writer personally, and once for writers in general. We also asked how positive or negative the impact of genAI would be in the next five years, how significant or insignificant it would be, and to what extent predicted changes in the publishing industry would be caused by genAI.

For every question, writers were more pessimistic than optimistic. After a Mann-Whitney U-test, four questions yielded statistically significant differences between the 54 members of the queer disabled group and the 75 members of the control group. Queer disabled writers said it had been harder for writers in general to make money in the past five years (p < 0.049); it would be harder for writers to make money in the next five years (p < 0.045); it had been harder for writers to get a publisher's attention in the past five years (p < 0.033); and the impact of genAI in the next five years would be more negative (p < 0.0035).

Despite being of similar ages and having similar years of experience (Table ??), our queer disabled writers make much less income from writing than our control group. These writers are both subject to worse career outcomes, and perceive and expect worse outcomes for their peers. They are also much more pessimistic about genAI.

Only two queer disabled writers out of 54 - M and N - expressed any optimism on questions about writers in general.

Despite dramatically lower income, queer and disabled writers are as likely as the control group to say that their own career has become easier in the past five years, or that it will become easier in the next five. Most of them did not mention homophobia, ableism, or other direct bias in the output of genAI as a reason for pessimisim. Instead, their pessimism concerns the industry as a whole.

5 Preliminary Qualitative Results

Our grounded theory's central concept is "precarity." Our writers consistently told us about precarious conditions that predated genAI. This information put the writers' opinions into context. Writers' motives are not primarily financial but have to do with intentional, personal expression and communication. Even though their output is fiction, they are expressing something unique to themselves. Because genAI does not have personal experiences, writers do not see it as useful for expressing themselves. Also, because a writer's alreadyprecarious income is tied to copyright, writers are sensitive to copyrighted work used for training; they view this as plagiarism. To these precarious writers, genAI is neither useful nor ethical. As a result, they employ boundary-setting and social advocacy to exclude genAI from their industry.

6 Existing Context

6.1 For Love andMoney: Motivations ForWriting

Our writers write because they love writing. This love has three components: emotional attachment to the activity of writing, a sense of purpose in communicating something personal, and a desire to connect with a reader. In all three respects, there is careful intentionality in deciding what to communicate. The joy of writing comes from intentional, often effortful, decision-making.

While money is not their prime motivator, to consistently engage in this activity, writers need stability. They pay great attention to the financial state of their careers and its milestones.

6.2 Pre-Existing Precarity

Although writers desire stability, it is challenging to achieve. The fiction-writing industry is significantly precarious. Our writers describe the precarity of both major branches of the industry - traditional publishing and self-publishing - each with its own mechanisms for creating and selling books.

In traditional publishing, writers prepare manuscripts which an agent sends to publishers. If a publisher purchases rights to the manuscript, the publisher edits, designs a cover, formats, distributes, and markets the book. Copyright remains with the writer, but publishers purchase publication rights, and often subsidiary rights such as for audiobooks and translations. In exchange, the writer receives an advance on royalties. All other rights remain with the writer and can be sold separately. (But writers do not have full control over what they sell; publishers can, for example, refuse to purchase publication rights unless subsidiary rights are included.)

In self-publishing, a writer assumes all a traditional publisher's responsibilities and produces and markets the book themselves, via online retailers such as Amazon, which sold over 90% of self-published books as of 2019 [Curcic, 2023]; consequently, self-published writers rely on Amazon's business model. Self-published writers retain full creative control and receive a larger share of royalties, but lack the reach of a traditional publisher; for instance, self-published books do not appear in brick-and-mortar bookstores.

Both branches of the fiction-writing industry are precarious. In traditional publishing, the number of publishers is shrinking, and larger advances are offered to fewer authors. 24% of writing income is distributed to just the top 1% of writers [Authors' Licensing and Collecting Society, 2022].

Generative AI is one element of the lengths corporations are going to to consolidate power and franchise. Both are alarming, but one is the cause. (Writer C)

Traditional publishers tend to be understaffed and consequently neglectful, so traditionally-published writers increasingly take on similar duties to self-published writers.

Everyone I know is complaining about publishers expecting authors to do all of their own marketing themselves, despite this being one of the main reasons to deal with a publisher. The common complaint being that in order to be an author, you're expected to be a social media influencer as well. (Writer R)

Writers also perceive publishers as fickle, preferring sure hits over artistically risky books and dropping underperforming writers quickly. This creates a double bind: writers do not feel supported, but are punished when books do not sell.

I expect that I will have a book that flops. The question is how willing will publishing be to trust that the book after the flop will hit. Authors, brilliant authors, get abandoned if they waver in profit. Which is why I keep my eye on other models, so I can pivot fast if I hit a snag. (Writer K)

Advances for traditionally-published books are determined individually, but are typically low.

In contrast, self-published writers compete for reader attention with an unlimited number of other self-published writers. Their primary issue is standing out in a crowded field. There is pressure to write many books very quickly. Meanwhile, constant changes in the platforms that self-published writers use to market their books mean constant new difficulties.

I do think that making money will become harder for authors in general. The demise of twitter has been a large blow to independent publishers. It was one of their main promotional avenues. With the changes to that platformmany users have left and it is much more difficult to promote books and work on the site. Amazon getting rid of its magazine subscription service also dealt a blow to magazine subscription revenue. A magazine like The Dark Magazine, had to drop it's [sic] per word rate from 6 cents down to 5 and is now publishing reprints where before they only published originals . . . I expect private equity to take over a lot of the resources we use and this will make it harder to gain readers attention because tools that were once free will become subscription based. Pay for visibility style social media seems to be the way everything is going. (Writer T)

Because writers "write for love," there is a sense of vocational awe [Ettarh, 2018] and writers are not treated by readers or publishers as requiring living wages or job security. The median income of a full-time writer is below minimum wage. This dearth of income is not business as usual; median income for full-time writers declined by 60% between 2006 and 2022 [The Authors Guild, 2023a]. The marginalized writers in our survey are keenly aware of this drop and are worried about their futures.

7 Experiences and Views of GenAI

7.1 Effects on the Industry

Publishing was already precarious before genAI, and genAI has worsened it further.

The effect of genAI our writers discussed most was the influx of AI-generated books into self-publishing. Competition among self-published writers for reader attention was already a challenge, but genAI has made this dramatically worse.

Generative AI is already flooding the self-pub market, making it very hard for readers to discern what is real fiction and what is machine-generated . . . I suspect that authors without existing name recognition, or without a recognised publisher behind them, will struggle to get found amidst the swamp

. . . Readers will either have a far harder time finding decent books amidst the AI dross, or will give up completely on trying to find anything that doesn't come from a traditional publisher. (Writer J)

This influx has also occurred for short-story magazines. Clarkesworld made headlines when it closed to submissions in early 2023 due to an influx of AI-generated stories [Clarke, 2023]. While Clarkesworld eventually improved its procedures and reopened, AI submissions continue to cause stress for its editors a year and a half later [Clarke, 2024]. Other well-known magazines have cited AI in closing [Dawson, 2024], or continue to struggle with waves of AI-generated submissions [Tobler, 2024; Uncanny Magazine, 2024].

This is not a matter of human writers being out-competed by the quality work of genAI models. Rather, editors report that these models produce "among the worst submissions we've ever received," [Clarke, 2023]. The problem is that the editor's inbox is flooded with, essentially, spam.

In traditional publishing, since submissions require an agent, writers' problems come from lack of control. Writers worry that their publishers will put AI-generated cover art on their books, sell their books as training data without their consent, or use AI to evaluate submitted manuscripts.

I think trad publishers are going to try to work generative AI into all of the processes in an effort to reduce costs. Many trad publishers are requiring that you sign an AI generated release in book contracts. These releases are non negotiable. Angry Robot announced recently that they wanted to use an AI service to help read, rate and sort their submission slush pile. They reversed the decision for this submission round due to push back from authors but did not comment on future submission periods. I believe that many publishing houses will look at implementing these systems without telling authors. These type of programs are detrimental to authors who write "outside the box." AI programs are only able to rate and grade work using the data scraped from other authors work. The biases of the creators of the programwill lead to these programs. (Writer T)

7.2 Views of AI

Ethical Concerns

LLMs are trained, in part, on data scraped from the Internet, including copyrighted fiction posted online, and these texts can be memorized and repeated verbatim by the model [Carlini et al., 2021]. LLMs are also trained on datasets of pirated books, such as Books3 [Reisner, 2023]. In some cases, models trained on this data create"knock-offs" of specific humans' work, which are put into digital marketplaces to directly compete with the original [Tapper, 2023].

Class-action lawsuits have been filed against AI companies for their unauthorized use of copyrighted works (see Knibbs [2024] for a summary.) The queer disabled writers in our study side with of the plaintiffs in these lawsuits and consider the use of AI trained this way to be plagiarism.

Copyright exists for a reason and if you want to use copyrighted works, you have to pay for them . . . Language models have the effective capacity to recreate their inputs and therefore are effectively functioning as plagiarism engines when trained over copyrighted text.

Money should flow from reader through publisher to author in exchange for the author and publisher providing the reader something to read. Plagiarism subverts that by cutting the original author and publisher out of the loop so they cannot recoup the costs of producing their content. (Writer N)

This plagiarism is viewed not only as a loss of income but as a moral and personal violation:

A machine can NEVER do what we do. We pour our hearts into the things that we write, and learning that that work has been stolen to feed a machine that regurgitates words based on formulas is soulcrushing. (Writer A)

The typical defense to accusations of plagiarism is to cite "fair use," akin to sampling and remix culture in music [Lemley and Casey, 2020]. However, our queer disabled writers do not view it this way and are not excited about having their or others' writing remixed by LLMs.

Many writers mentioned other ethical concerns, such as the energy and climate costs of genAI, the proliferation of misinformation, and the disruption of online services such as news and search; these concerns have been extensively discussed elsewhere (e.g., [Bender et al., 2021; Jiang et al., 2023])

"Soullessness"

Fiction writers "write for love"; intentionality and conscious communication are central to what makes a work of fiction valuable in a writer's eyes. Our theoretical concept of value derived from conscious, intentional communication is "soul." We do not make any ontological claims here about actual souls, but simply adopt a term the writers themselves used.

To have soul means that a writer consciously uses fiction to communicate something they think and feel. They make intentional effort to express these thoughts and feelings in a form to which a reader can respond. To a writer with soul, each work of fiction has intentional meaning and this meaning is meant to be transmitted.

Writing is thinking, and art is about humans making things special and sharing them. Much of the joy of writing and art is about humans trying to understand their worlds, learning what's special about them, and deciding to share their discoveries. While there's great value in generative AI, it is only replicating a small part of the human experience, and we should make sure that it does not disrupt the cycle of understanding, making special, and sharing that makes us human. (Writer N)

By contrast, our queer disabled writers view genAI's writing as "soulless," coming from a non-conscious, mechanical process, and therefore devoid of intentionality. Rather than creating art, even bad or unskilled art, the AI is seen as producing something fundamentally counterfeit.

"Soul" resembles the argument of Jiang et al. [2023] that, in philosophies of art from around the world, art is a specifically human process involving conscious intent to communicate an individual's affective experience to an audience.

The idea of co-creativity, or of worthwhile art emerging from the interaction of a "souled" human with a "soulless" device, was not mentioned by our writers. This contrasts studies of genAI use in other fields. Many of the artists in Kalai et al. [2024], for instance, embrace becoming "gardeners" or "curators" of AI works rather than direct creators. Perhaps because of the solitary character of writing, and perhaps because they are less likely to see their own lives reflected in "mainstream" literature, queer disabled writers do not want to be gardeners. Several expressed that even if genAI did allow them to create books faster, and without significant ethical issues, they would still not use it due to its lack of a soul.

It cannot do any part of my job better than I can do it myself. And even if it could, I would not buy a machine that would eat my ice cream for me. (Writer K)

Most queer disabled writers in our data set had never used genAI, and most would not do so under any circumstances. A few did describe experiments with genAI. Their initial experiences involved simple curiosity, leading to disappointment.

I played around with using ChatGPT when it came out to see what it could do with poetry, dialogue, various writing things. I was disappointed in the results. I am told ChatGPT has improved, though it's poetry doesn't seem to have improved... (Writer D)

For other writers with more technical expertise, the AI did prove useful as a brainstorming tool, as a tool for preliminary research, or a producer of raw material that the writer could modify and combine with other inspirations as they saw fit.

I experimented with generative AI to create a movie script, but abandoned it. I have used it to create illustrations . . . but the ability of the systems to take art direction was poor.

I have used it to create marketing materials for books, but while the content of the suggestions was good, the actual text was unusable. (Writer N)

Only three queer disabled writers (D, M, and N) indicated that they have ever used genAI. Due to theoretical sampling, we have already analyzed all three of these writers' responses. Because the number is so small, any firm conclusions about fiction writers' direct experiences using genAI will need to wait for a full analysis of our data.

7.3 Mental Health Effects onWriters

Because genAI has deepened the precarity of an already precarious industry, and because of our writers' ethical opposition to AI, many have experienced strong emotional reactions. A large number said that the rise of genAI had been harmful to their mental health. While the specific difficulties can vary - various writers described anxiety, stress, resentment, anger, sadness, depression, and hypervigilance, with few listing more than one of these - almost all told the same underlying story. It was already difficult to maintain good mental health as a disabled writer in the publishing industry, and the rise of genAI made this worse.

I was already somewhat stressed and anxious about the prospect of building myself a working career as a writer, and the rise of generative AI has simply magnified that anxiety. It is becoming progressively more difficult to imagine myself being truly successful as a writer and being able to rely on writing as my primary source of income, which I would really love to do, as I am disabled and struggle to find other work. (Writer A)

The ubiquity of genAI - advertised widely, added to existing software, and constantly discussed - makes it hard to escape these concerns. Coping strategies, such as making time to focus on something else, become more difficult when the problem is hypervisible and ubiquitous.

there's no escape now. it's in every browser, every search engine, it's in Word, which I have to use, it's in my google docs, which I no longer use for betat [sic] reading and critique. And It's not like I can simply print my manuscript and do everything on paper from here on in, that's ridiculous. (Writer K)

These impacts on mental health should give us pause. A widespread increase in the severity of mental illness among writers who were already disabled means a high likelihood of the unique viewpoints of this already marginalized population being silenced or marginalized even further.

8 Strategies of Response

Given that fiction writing is precarious, and that genAI has worsened this precarity, it is not surprising that fiction writers have developed individual and collective strategies to protect themselves. Writers pursue "momentum," which increases their resilience by building on past success. They set boundaries to avoid being involved with genAI. They have also work to create social pressure to discourage the use of genAI. These actions include solidarity with other precarious creative humans, such as the visual artists who create cover art and the narrators who create audiobooks.

8.1 Momentum: Building Personal Success

"Momentum" is the tendency for previous successes to enable future success. The more that a writer is recognized by publishers and readers, the easier it will be to sell future work. Momentum includes broad measures like sales numbers, but also the idea of building a "niche" or a "brand" - a style or set of themes that readers know to expect. Momentum protects writers individually against precarity.

It's become easier for me because I've built a bit of a brand . . . and it's been helpful in promoting my work to readers. Also, because of my previous sales, community work and award nominations I've had a bit of an easier time convincing people to try my work. (Writer T)

Inexperienced writers are also aware of momentum and hope to create it for themselves.

I am hoping that making money as an author will become easier for me over the next five years, because I hope to publish my first novels, along with more short stories. I also hope to start building a fan base that will recognize my name and more consistently purchase my work. (Writer A)

Momentum is not absolute; it can be lost. Even writers with a great deal of momentum - successful books and award nominations - cautioned us that their situation is precarious.

I can't deny that it's easier for me now. I have the evidence of book sales - my first statement sales numbers are higher with each "tentpole" book I've released . . . The trick is keeping that curve, though, and I don't trust that I can. I have to get readers to trust me all over again every time I release a title, because while I am generally a fantasy writer with strong romance threads with protagonists of all kinds of sexualites [sic] who bangs on about social justice in very obvious ways, there's always something different. (Writer K)

8.2 GenAI Boundaries

Many writers in our data have acted against genAI by setting personal boundaries. Rather than safeguarding the writer's career, genAI boundaries are set in solidarity with other humans whose careers are at risk from genAI. They also protect the writer from the moral risk of becoming involved with a technology that they find unethical.

We find four distinct levels of genAI boundaries. At the first and mildest level, writers seek to prevent AI from being used in relation to their work. This is challenging for traditionally-published writers who lack control over cover art, audio versions, translations, or advertising. One way to gain this control is in the book's contract.

I want terms that all of my production and promotion be done by people who don't use AI in their work. I don't want anything to do with it. And that might mean that my career in traditional publishing is over. (Writer K)

At the second level of genAI boundaries, writers refuse to work with publishers who have used AI, regardless of whether AI will be used on that writer's work.

I've had to start being more watchful . . . you always have to keep an eye out for if publishers are using AI covers, etc, and so be aware of who you don't want to publish with. (Writer J)

At the third level, writers refuse to use software with AI features, such as word processors and search engines - even when this makes the writer's everyday labor harder.

The need to avoid it has added so much work, and so many extra steps, to my work. I can't use programs I used to rely on, or social media platforms that used to promote my work, and tools I still do use because I have no alternatives require continual opting-out . . . (Writer R)

At the fourth level, some writers set genAI boundaries by cutting professional and social ties with anyone who is not also boycotting genAI. (Writer U is the only writer, among those we have qualitatively analyzed so far, who said they had done this; they did not consent to be quoted directly.)

When we have analyzed all 137 writers, we will be able to estimate the relative prevalence of these various levels. It is still possible we may also discover additional levels.

None of these boundary-setting strategies are risk-free and some have obvious costs. Writers engage in them anyway due to the strength of their convictions against genAI and their solidarity with other human artists.

8.3 Public Backlash and "Speaking Out"

The third strategy is public backlash. Groups of creative workers who can unionize, such as the Writer's Guild of America (representing screenwriters), have organized against genAI with some success [Anguiano and Beckett, 2023]. Prose fiction writers are freelancers who cannot collectively bargain, so their collective action involves informal social pressure. Writers in our data discussed public discourse as a collective means of limiting AI-related harms.

Generative AI is causing negative effects already, and I have no doubt it will continue to do so for a while. However, I am hopeful that the public backlash over it (see Angry Robot's recent open subs controversy as one example) will ensure that no reputable publisher will start using it in any real sense. (Writer J)

Several writers mentioned Angry Robot, a midsize science fiction publisher that, in early 2024, proposed to use AI to sort manuscripts during an open submission call [Portsmouth, 2024]. Less than a week later, after significant social media backlash, the publisher backtracked [Coker, 2024])). If not for the backlash, the AI would likely still be in use. In at least three cases, larger publishers - specifically, Tor [Codega, 2022; Stewart, 2024] and Bloomsbury [Weatherbed, 2023] - have faced similar waves of online backlash after using AI-generated art in their cover designs. As Stewart [2024] describes, this backlash was intentionally driven by BookTok influencers preventing AI from "infiltrat[ing] creative spaces." Although the publishers did not retract the covers, Tor claimed it hadn't been aware of the use of AI [Codega, 2022; Stewart, 2024]. Online backlash can affect publishers' choices and cause them to reduce or hide uses of genAI.

Formal groups of writers can also play a role. Groups like The Author's Guild have engaged in data gathering and advocacy geared towards informing policymakers about writers' values and needs related to AI [The Authors Guild, 2023b; International Thriller Writers, 2023] and have joined several class-action lawsuits [Knibbs, 2024]. However, the writers in our study seem less aware of these efforts and more aware of informal efforts on social media.

Various scholars in the social sciences (e.g., [Mesler et al., 2024]) have investigated online call-outs, pile-ons, and "cancelling," primarily in terms of the importance of these behaviors to a participant's political or in-group identity. However, these pile-ons, as in the Angry Robot example, can also have a material effect. Since marginalized writers view genAI as increasing precarity and harming their mental health, preventing the use of genAI is self-protective. Online vitriol against AI should therefore be viewed, not as mere "polarization," identity formation, or virtue signaling, but as intentional strategies to protect writers' interests as well as those of other creative humans with whom writers stand in solidarity.

9 Discussion

GenAI's creators predict that every industry will be transformed, but appear to be uninterested in the needs of workers in these industries. The tasks necessary to protect workers are explicitly or implicitly outsourced to other researchers with less institutional power (e.g., [Eloundou et al., 2023].) Studies like ours are one way of illustrating the actual situations in which workers find themselves due to technological change.

Our writers do not feel that genAI has made any aspect of their work easier. In the publishing industry as experienced by our queer disabled writers, the biggest barriers to full participation are not lack of skill or lack of technology, but precarity, uncertainty about the future, and low income. GenAI appears to have increased these barriers.

Compared to similar qualitative studies [Kaila et al., 2024; Vimpari et al., 2023; Uusitalo et al., 2024], the opinions of our queer disabled writers are unusually negative. There are several possible explanations. First, queer disabled writers' marginalization results in more severe precarity - as well as a lower belief, as mentioned by Writer A above, in their ability to succeed in other industries. Although few writers mention their queerness or disability directly, it is not surprising that a population already experiencing worse outcomes would be more pessimistic about the future. We have yet to analyze qualitative differences between these writers and the more optimistic ones in our control group - although the control group also tends toward pessimism.

Also, the concept of "soul" is relevant to fiction writers due to the solitary nature of their creative practice. It may be easier for a fiction writer to "pour their heart out" into a work, and feel a strong sense of personal ownership, compared to a worker in video games or industrial design who primarily works in a team. Workers in industries where individual selfexpression is paramount may have a harder time embracing genAI than workers with a team-based approach. However, studies in fields such as industrial design have shown that concern does exist in these industries over ownership of AI output, lack of control over adoption of the technology, and the possibility of human workers being replaced; this anxiety is highest among the lowest-ranking workers with the least power [Uusitalo et al., 2024]. Worries about precarity and solidarity appear to extend across creative fields.

9.1 Limitations

Our results only reflect a particular set of fiction writers. Very little of our theory, except perhaps the results on mental health, depends on a writer's status as queer or disabled, so it may generalize to writers more broadly. We have yet to analyze the views of the writers in the control group.

Queer identity and disability status are only two of many possible axes of marginalization. While our survey materials asked about several forms of marginalization (e.g. race, immigration status, language, religion) we did not receive enough responses in these categories to give them the intense focus that we could for queerness and disability. In particular, both our data and the general population of fiction writers are disproportionately white. We have attempted to mitigate these biases through theoretical sampling and by deliberately seeking out BIPOC voices within our data (some of whom, e.g., Writers K and T, are extensively quoted here.) This should not be construed as a substitute for a study directly focusing on BIPOC experiences of genAI in publishing. Similarly, our writers are from the Global North; those in the Global South experience genAI differently [Arora, 2024].

Finally, because our data was collected mainly on Bluesky, it is possible that the responses we received may overemphasize the importance of online discourse, or may be more polarized than a fully random sample of writers. Non-academic surveys (e.g., [The Authors Guild, 2023b; Humanity in Fiction, 2023]) also show widespread concern about plagiarism and distrust of LLMs among writers, so we do believe that the overall opinions expressed in this study are representative opinions. Nevertheless, we do not attempt to claim that these are the opinions of every writer - merely that they result from real concerns and deserve our attention.

9.2 FutureWork

The first priority for our future work is to finish analyzing our data. Although the current paper lacks space for it, we would like to apply this analysis of writers' needs to relevant law and policy discussions, such as discussions of how best to compensate writers for the use of their work as training data, and of opt-in vs. opt-out mechanisms. We hope to expand this into a longditudinal study in which changes in writers' experiences and attitudes are tracked over time.

As computational creativity researchers, we find AI's impact on the arts particularly interesting, but this should not imply that creative workers are unique or more worthy of protections than other workers. Any workers who are subject to precarity, as a result of the imposition of genAI tools and processes, should be considered.

Ethical Statement

The purpose of this research is to shed light on experiences, including harms, inflicted on a specific community through technological change. The research is intended to benefit the participants by bringing their concerns to a wider scholarly audience. However, it is also possible for participants to be harmed.

In particular, the qualitative responses we received include deeply personal and emotional statements about contentious sociopolitical issues and many are, by their very nature, identifying. For instance, some writers chose to list awards they had received, or the number and type of books they had published in different years, which, in combination with other information such as their answers to demographic questions, could be used to identify the specific writer by name. Writers also, in some cases, worried that their careers could be in jeopardy due to their opposition to generative AI. For this reason, our full qualitative data set is sensitive data which cannot be shared, even for the purpose of ensuring reproducibility.

It is also possible for participants to experience distress as a result of the survey questions, which ask about major career difficulties and about mental health. Participants were warned about this risk in the consent letter and were able to end their participation at any time.

This research was approved by our institutions' ethics review boards. Informed consent to participate was given by all participants, including optional consent to be directly quoted in the paper, which could be given or withheld. Even when consent was given, we have refrained from including any quotes, such as lists of awards received, which could be used to indirectly identify the participants.

The first author of this paper is, by the definition used in this paper, a professional fiction writer (their fictional work has been published under a pen name). Both the first and second authors identify as queer and disabled.

Acknowledgments

This research was made possible by an internal grant from the Queen's University Fund for Scholarly Research and Creative Work and Professional Development (Adjuncts) to C.E.L, and by Discovery Grants from the Natural Sciences and Engineering Research Council to D.G.B. andM.R.G. We would like to thank Charlie Clarke, Kazjon Grace, Anna Jordanous, and Chris Liscio for insightful and helpful conversations.