Abstract
Whereas many experts believe that fears about AI are wildly overblown, catastrophic predictions are commonplace. Concerns may be particularly prevalent in US creative industries following high-profile labor negotiations around AI-related issues. Although many experts believe AI is most likely to augment rather than eliminate jobs in these fields, this is not true (or equally true) for all kinds of creative work. At the same time, perceptions of its potential effects can vary dramatically, even among people in similar job roles, depending on how the future of AI is imagined in relation to their labor and economic interests. Drawing on interviews with 50 creative workers, we ask, how does the anticipation of – or actual experience with – AI exacerbate perceived differences in interests among occupationally heterogeneous creative workers? We find that these workers fall into three ideal types: the “Hardly Worried” view themselves as uniquely insulated from the detrimental consequences of AI, while the “Heartbroken” respondents were concerned that they are about to lose beloved career options. The third group, “Hired Guns,” were often aware of the risks of AI, and how it could be used, but viewed the training or implementation of AI as an opportunity to make fast and easy money or as providing them with new career opportunities. As a result, we argue that efforts to organize these workers, or to rein in how AI can be used, may require addressing larger issues of discordant AI perceptions and financial inequity.
It has been said that artificial intelligence (AI) is coming for professional and creative jobs (Ozgul et al., 2024). While it may be difficult to train AI to drive a bus in a congested suburb, or to address all of the challenges associated with delivering a meal to the resident of a large apartment building, recent iterations of AI, such as ChatGPT, have shown impressive ability to write college essays (Waltzer et al., 2024) and news articles (Verma, 2024), create art (Tigre Moura et al., 2023), and even diagnose patients (Goh et al., 2024).
For creative workers in the fields of film or television production, the potential threat of AI for jobs was front and center in the 2023 strikes by the Writers Guild of America (WGA) and Screen Actors Guild-American Federation of Television and Radio Artists (SAG-AFTRA) (Carras, 2023; Kinder, 2024). However, while members of the WGA and SAG-AFTRA have organized against the threat of AI, creatives have experienced ongoing challenges in defining themselves as a cohesive group of workers deserving of labor protections and benefits enjoyed by other professionals (Skaggs & Aparicio, 2023). Although creative workers have begun to organize themselves in both traditional and non-traditional labor organizations, a high level of occupational heterogeneity within the industry has frustrated efforts to build solidarity across boundaries of difference like race and institutional affiliation.
In this paper, we build on the sociological insight that perceptions of interests are embedded in “sociotechnical imaginaries” that make predictions and evaluative judgments about the design, implementation, and consequences of technologies like AI (Jasanoff & Kim, 2015). Drawing on interviews conducted in early to mid-2024, soon after the two strikes concluded, we ask, how does anticipation of—or actual experience with—AI exacerbate perceived differences in interests among occupationally heterogeneous creative workers?
We draw our data from a mixed methods panel study conducted with 50 creative workers, many of whom had direct experience working with Large Language Models or other generative AI tools. We define “creative workers” as individuals who self-identify as being focused on acting, writing, photography, dance or actively contributing to those fields (such as working as lighting techs for theatre performances or tailors for film productions). Originally recruited between April and June 2020 as part of a panel study on the long-term impact of Covid on work, these respondents have been interviewed repeatedly between 2020 and 2024 as part of a project to examine cultural and social changes in the years after the pandemic (Ravenelle, 2023). The findings in this paper draw from Phase IV respondent interviews conducted between March and October 2024, a period when Covid was widely viewed as being endemic, and less relevant, and AI was in ascendency.
Our project is especially novel due to three factors. First, the timing of our interviews with workers in creative industries who may be impacted by AI occurred roughly six months after the 2023 WAG and SAG-AFTRA strikes concluded. Many of our respondents had, or were currently, working in film or television, and were directly impacted by the two strikes and resulting provisions related to AI, which prohibited studios from using AI to create or edit human scripts and required that actors give their consent before studios created a digital replica. The media attention on the strike also served to raise awareness of AI as a whole among our respondents, especially in terms of how AI was growing and how it could be used both currently and in the future. Second, while most creative workers we interviewed used AI casually—such as a musician using AI to create a cover for their album—our sample also includes creative workers who are actively using AI for money-making purposes, either by training AI systems or turning to AI-enabled jobs. We argue that this active use of AI for financial gain suggests that any efforts to organize workers or regulate AU will necessitate a discussion of how the future of AI is imagined in relation to personal economic interests. Finally, we document discordant perceptions of AI's effects within an industry that is widely anticipated to face significant challenges and transformations in the wake of AI.
AI's Impacts on Creative Workers
Forecasting the effects of AI on workers is a difficult task that has divided experts. Whereas prominent technology executives like Sam Altman and Chris Hughes have repeatedly made predictions that could be characterized either as utopian or dystopian (Alderman & Satariano, 2024; Robson, 2025), other business leaders and scholars anticipate significant but relatively measured job losses alongside productivity growth (Placani, 2024; Watson, 2019). It is not at all clear from contemporary discourses about AI how this technology will affect workers in general—or even, in many cases, those holding particular occupations. “Automation anxiety” (Bassett & Roberts, 2023) may be particularly prevalent in US creative industries on the heels of high-profile labor negotiations around AI-related issues, and although many experts believe AI is most likely to augment rather than eliminate jobs in these fields, this is not true (or equally true) for all kinds of work (Anantrasirichai & Bull, 2022).
The advent of consumer-facing generative AI tools like ChatGPT has raised the possibility that creative work, previously considered resistant to automation (Ford, 2015), may soon become the province of AI as its scriptwriting, video generation, music production, graphic design, and other creative capabilities continue to improve. Nonetheless, AI's impacts on workers will vary by occupational role (Anantrasirichai & Bull, 2022; Kim et al., 2023). For example, automation of tasks can increase productivity without a concomitant increase in wages if new tasks are not created, which is especially likely in service industries (Acemoglu, 2025; Furman & Seamans, 2019; Rashid & Kausik, 2024). Further, automating unskilled labor results in increased demand for skilled labor, exacerbating existing inequalities (Acemoglu & Restrepo, 2024).
On the other hand, some scholars foresee that AI will continue to augment human productivity and capabilities rather than replacing people in the creative labor process. Many artists currently use generative AI and other AI tools to explore ideas and “overcome capability gaps that have traditionally limited creative expression” (Arora, 2025, p. 19). Similarly, applications of AI in the news media industry currently involve uses like automated transcription and data compilation that have improved efficiency without necessarily resulting in job losses (Simon, 2024). Given the rapid development of AI technologies, it is not yet certain whether current use cases will change significantly in a manner that would displace creative workers.
Heterogeneous Consequences
For knowledge-based work, the effects of AI are also “jagged,” depending granularly on the specific capacities required by individual tasks, as generative AI can vastly increase productivity in some cases while having little effect or even detrimental effects on other, seemingly similar tasks (Dell’Acqua et al., 2023). Nonetheless, tasks requiring creative thinking are more likely to be automated by employers compared with repetitive tasks that are already inexpensive to perform (Kanbach et al., 2024, p. 1201). Preliminary evidence also shows that freelancers who provide creative services based on writing or image creation have experienced substantial reduction in work and earnings following the release of generative AI (Hui et al., 2024).
Likewise, the diversity of job types and tasks within the creative sector, including “curating, managing, producing, and otherwise supporting the creation or performance of art” (Skaggs & Aparicio, 2023, p. 437) means that creative workers will experience wide-ranging consequences from further automation. For instance, AI may increase productivity of creative knowledge-work while generating new computational tasks for others to perform (Amankwah-Amoah et al., 2024). On the other hand, some evidence suggests that increased demand for computational expertise comes at the expense of displacing “ideational” labor (Erickson, 2024). For instance, in the interviews with gig workers completed to date as part of this project, we find that some of the computational labor needed to train AI is facilitated by gig workers who do not receive salaries, benefits, or job protections—potentially using workers in “bad jobs” to train technologies that could replace “good jobs” (Kalleberg, 2011). This often-invisible precarious work is essential to the functioning of AI tools, even as they are marketed as fully autonomous (Muldoon et al., 2024). These gig workers perform a variety of tasks, including training AI, selecting or cleaning data used as model inputs, and even “impersonating” AI by simply performing tasks that are presented to consumers as being automated (Tubaro et al., 2020).
Heterogeneous Consequences, Mutual Interests?
However, AI's consequences depend not only on the technical implementation of AI, but also on workers’ ability to collectively resist potential job losses and to demand wage increases in line with productivity growth (Acemoglu & Johnson, 2024, p. 597). Although the heterogeneous nature of work within creative and knowledge industries has historically frustrated efforts at unionization (Skaggs & Aparicio, 2023), the high-profile protections against AI won by SAG-AFTRA and WGA (US unions representing performers and writers respectively) have renewed interest in creative workers’ ongoing efforts to organize themselves, now against the uncertain threat of AI (Scherer, 2024).
The success of these efforts hinges in part on whether creative workers in disparate fields can construct a shared identity around perceived similarities between their own interests and those of others whose roles may be differently impacted by AI (Fantasia, 1989, 1995). Whereas many creative workers currently believe AI will result in job losses, a view that motivated unions’ demands during the recent labor actions in the US (Bender, 2024), the rapid adoption of generative AI may already be splintering perceptions as creative workers begin to reap differing benefits and consequences from its integration into labor processes. These uneven effects, coupled with increasing precarity in the broader economy, create barriers to collective identification and solidarity among workers who personally bear responsibility to remain productive in the workplace and competitive in labor markets (Morgan & Pulignano, 2020). The perception of common interests often depends on sustaining a shared definition of potential threat that binds workers into an imagined “community of fate” whose fortunes are seen to rise and fall together (Baehr, 2016). This shared conception of threat further enables common attributions about the causes of threat that can mobilize collective action (Kelly, 1999).
Imagining Common Futures of AI
For creative workers interpreting the types and degrees of threat from AI, a primary barrier to perceiving common interests is the high degree of uncertainty about the future course of AI development and its consequences. Predicting future states of the economy is a fixture of capitalist societies that by necessity depends on often-unreliable information or even fantastical beliefs in the face of the unknown (Beckert, 2013, 2016). Widespread disagreements among experts and political elites alike have contributed to a void of certainty about AI that much of the public has filled with expectations based on alternative information sources like science-fiction and personal experiences with technological automation (Kowalski, 2026; Sartori & Bocca, 2023).
The varied understandings offered by technical experts, businesses, news media, politicians, and members of the public coalesce in what Jasanoff and Kim (2015, p. 4) call “sociotechnical imaginaries,” or “collectively held, institutionally stabilized, and publicly performed visions of desirable futures, animated by shared understandings of forms of social life and social order attainable through, and supportive of, advances in science and technology.” Sociotechnical imaginaries include both the discursive construction of future possibilities that legitimates particular avenues of technological development, along with the practices that this “rhetorical construction of future worlds” may facilitate, including designing and implementing technologies, as well as shaping the public's response (Selin, 2008, p. 1879).
Although “AI imaginaries” that predict the future impacts of AI are highly fragmented depending on the material and political situation of interested parties (Richter et al., 2023), uncertainty can also enable new and unexpected forms of solidarity to emerge (Morgan & Pulignano, 2020; Suckert, 2022). The futures envisioned within sociotechnical imaginaries are ongoing sites of symbolic and material contestation (Mische, 2009; Neresini et al., 2020; Schiølin, 2020), and debates about the future of AI are only beginning to take shape in contexts ranging from local communities weighing AI infrastructure projects to geopolitical power struggles between nations (Richter et al., 2025).
The fragmented and unsettled nature of AI imaginaries can empower workers and ordinary people to influentially shape the course of this technology (Grohmann et al., 2025). For example, early evidence suggests that consideration of AI can lead broad swaths of the public to question the connection between productivity and deservingness at the heart of neoliberal political culture (Kowalski, 2026). Currently, much of the existing work on AI imaginaries has focused on business leaders and political figures (Richter et al., 2023). What kinds of “shifting solidarities” (Morgan & Pulignano, 2020) are now underway among creative workers, public stakeholders who are often portrayed as likely casualties of AI advancements?
Methodology
The findings in this paper draw from respondent interviews with 50 creative workers conducted between March and October 2024. Our respondents are part of a larger mixed methods panel study of 199 respondents, including gig workers, creative workers, and low-wage workers that began in April 2020 (Ravenelle, 2023). The goal of this larger project was to study the longer-term impact of the pandemic, and related social cultural phenomena, on workers in the New York metropolitan area. As a result, for roughly six years we have been documenting the rise and subsequent decline of the Black Lives Matter movement, changing concerns about covid and immigration, and now, the popularization of AI. By following these respondents over time, and utilizing a respondent-driven interview strategy, we are able to document when issues rise to the forefront and how responses to various cultural phenomena change over time.
Since AI was not mentioned by respondents until the 2024 interviews, the data for this paper comes entirely from the respondents’ 2024 interviews. During 2020, respondents were recruited via Facebook groups for unemployed workers, online ads (e.g., Craigslist), and snowball sampling. Eligibility was limited to workers in the New York metropolitan area (broadly defined as Westchester, NY to Newark, NJ) who lost work or experienced an income disruption due to the pandemic and who were in creative fields. 1
Attrition in our sample has been low. Of the 60 creative freelancers first interviewed in 2020, 50 also completed an interview in 2024, an 18% attrition rate, and there are no major differences between respondents who remained in the study and those who left. Respondents in the 2024 interviews included 13 actors, eight dancers, six people involved with music and music production, five filmmakers, 4 photographers, three models, two writers, two lighting designers, a costume designer, voiceover artist, comic, circus arts performer, stylist, fashion tailor, and operations manager at a media company. This range of occupations tracks closely with the wide variety of creative workers found in New York (Office of the New York City Comptroller, 2024). For the 2024 interviews, respondents were offered a $100 incentive, up from $25 for the first interview, and $50 for a second interview, in order to reduce attrition.
In each phase of the panel study, participants were sent a short survey and informed consent to complete before participating in a respondent-directed telephone interview (Weiss, 1994) scheduled by the respondent at their convenience. All interviews are recorded, transcribed, and coded using flexible coding (Deterding & Waters, 2021). Flexible coding, where responses are indexed to the interview questions, reduces issues with intercoder reliability. After index coding, conceptual codes are developed and applied after discussion with the research team. All respondents were assigned pseudonyms, although names are also chosen to approximate racial-ethnic and gender differences in name popularity.
Of the 50 participants interviewed in 2024, 36 (72%) were female, 24% were male, and 4% identified as gender-nonconforming or non-binary. 34 respondents (68%) identified as white, with 6% identifying as Black or African American, 8% as Hispanic, 8% as Asian, and 10% as multiple races. The interviewees ranged in age from 24 to 60-years-old with an average age of 36.32. Our respondents are highly educated, even more so than the 40.2% college education rate typically found in the New York area (U.S. Census Bureau, 2020). 16% had some college experience, and 6% had an Associate's degree, while 42% held a Bachelor's degree, and 34% held a graduate degree or had some graduate school experience. Based on their reported weekly income, nearly half (44%) of interviewees made less than $40,000 per year, with 12% earning less than $20,000 per year. 2
Findings
We find that respondents differed in their views on if—and how—AI was a threat to their occupation-based interests both within, and in relation to, the creative industry. These differences were often linked to their socioeconomic status. In the following section, we outline the three broad responses to the threat of AI, Hardly Worried, Heartbroken, and Hired Guns, category names that draw directly from respondents’ responses (Ravenelle et al., 2021). These three responses are also outlined in Table 1: Three Broad Types of Responses to AI.
Three Broad Types of Responses to AI.
Hardly Worried: “You Need Real Humans Doing Stuff”
The workers we describe as Hardly Worried often brushed off the potential risks to their livelihood that AI could present. A common perception was that these workers offered a human “touch” or “connection” in a way that machines or artificial intelligence could not. For instance, Michelle, 40, a racially-mixed lighting designer, noted that although her industry was talking about AI, she didn’t view it as a threat. “I think it, like anything, is a really great tool to use as an art form or to help you with your art,” she said “I don't think it will replace a human connection, if nothing else, for the novelty of human connection.”
Likewise, Heath, a digital marketing professional who was working at a grocery store to help pay his bills while he looked for work in audio production and podcasting, noted that while AI had some uses, he viewed it as “too flawed.” Instead of offering a replacement for human work, Heath, a 48-year-old white man, described AI as “it's good as maybe leveraging it from a jumping-off point base for research, but at the end of the day, you need real humans doing stuff.”
Much like the elite gig workers who don’t worry about competing with workers overseas because their skills and deliverables were more complex and less “cookie cutter” (Ravenelle & Kowalski, 2022), creative freelancers who were not worried about AI felt that their work required more skill and was more nuanced. As Kurt, a 37 year-old white man who worked as a filmmaker and director, explained: “I'm really not worried about myself being out of a job, because I think that what I do is complex enough and requires enough human interaction that there's no way it can be replicated.”
Respondents were especially nonplussed about the potential impact on their jobs because earlier experience had taught them that technological solutions still required human interaction. Working as an assistant editor for an online magazine, Raven, a 35-year-old Black/Hispanic woman, didn’t understand why her manager hyped up ChatGPT. She noted that the AI system didn’t provide sources for its information and that a lot of the content was wrong. “It was very strange to me that they even considered using it,” she said, before noting that several months later her manager was seeking assurances that no one was using AI in their work.
“I was actually happy about that because I didn't want to use it. I was very annoyed by it,” she said. “And also, you need the human cultural awareness and stuff to discern what's actually an error versus something that's a phrase or a metaphor, knowledge you have for it. They don't have that human cultural knowledge. It's very easy for me to see that you can't completely turn off writing and editing to turn it over to AI, because they keep making mistakes.”
While AI couldn’t be used in her writing field, Raven's company also employed a number of graphic designers and their challenges in using AI tools further cemented her lack of concern. “We have a lot of graphic designers. They can't use it. If you notice, they'll mess up fingers, and things look crazy with AI, you know what I mean? It's not fail-proof at all. I do think it's helpful though, to generate ideas and generate spark and outline or spark ideas. I just don't think it's good for executing ideas.”
Not only did the Hardly Worried workers think that AI would make their life easier, but they also believed that it wasn’t cost effective to replace them. As 30-year-old Jacob, a white man who worked as a videographer and photographer, explained, “Yeah, I think it'll augment my work and it'll make certain things easier… I think the financial incentives to take video editors out of work are not really there. In the scheme of things, it's a relatively low-cost task, and I don't see OpenAI or Microsoft or whoever investing all of their resources in doing that when there are so many other higher cost or higher potential revenue tasks that could be replaced.”
When respondents did mention the possibility of AI affecting them, they often brushed it off. Vanessa, a 32-year-old Asian woman who worked as a sound producer, had recently joined the SAG-AFTRA union and was working full-time on a TV series. Asked if she thought AI could replace her, she was not especially concerned. “I think in terms of a body of evidence, I've proven to myself, I can very easily adapt and change doesn't scare me at all. So if need be, if I do have to change or adapt or build a new skill set because of a forced break, I'm totally open to it, and I would even welcome it.” Despite being a union member, Vanessa expresses her confidence by emphasizing the need for personal adaptation, underscoring the challenges unions face in convincing even their own membership that jobs can be secured through collective action.
When Hardly Worried workers did express concern about AI coming for jobs, it wasn’t their job, but actually the work of other people. Claudia, a 37-year-old white woman had been a filmmaker, actor, and director before the pandemic and she continued to dabble in tv production after mostly transitioning into selling real estate. Claudia, who now had a stable source of income, was not worried about the impact of AI on her career, but she was also quick to note that there were many jobs that could easily be filled by AI. “I think if anything, the AI will come for the coordinators. I think that a lot of the scheduling is going to go to the AI,” she said, later adding, “Sure, you're going to probably see they're going to come for costume illustrators, that's for sure. That's a whole job on its own, people who all they do is make costume sketches. And I can guarantee you that AI is going to come for them.”
The perception of AI as coming for other people's jobs wasn’t just about roles and industries, but also included a race component, such as people of color no longer being hired for modeling jobs. Aanya, a 38-year-old Asian woman photographer, noted that she had to “think really hard about working with AI rather than fighting it.” Much of her work involved taking pictures of products and while AI tools could create an image of a person in a blue ski jacket, it would often be a generic jacket and not necessarily the specific brand of ski jacket that a client was looking to sell. At the same time, in some ways, anonymous models were more replaceable than products.
“What I've heard is it's going to screw up people who are people of color. Instead of agencies trying to go out and hire diverse casting, people of color, different body types, blah, blah, blah, the fear is that they're going to shoot it on one kind of body type and then just have AI generate alternate bodies and heads wearing the same product. That really sucks for the models,” she said. For all of the talk about human uniqueness, when it comes to people being used as tools to showcase clothes, the anonymity of unknown models may make it easier to replace them with AI generated “people.” Aanya therefore constructs her own interests around the possibility of creative adaptation—“working with AI”—while casting the models as helpless against AI's onslaught. Ultimately, Aanya expresses sympathy rather than solidarity, noting that this situation “really sucks for the models,” in contrast with her own circumstances.
Heartbroken: “It's Very, Very, Very, Very sad to Wipe out That Source of Income”
Whereas the Hardly Worried didn’t view AI as particularly disruptive to their work, respondents who fell under the Heartbroken category felt that AI had taken away a crucial component of humanity, or was reducing their career options. Much like the Hardly Worried, these workers also viewed AI as impacting other workers, but also included themselves among the workers who would lose work.
For instance, Eric, a 37-year-old white man
While Eric was concerned about the potential impact on his back-up career, other workers who had recently trained for a new career that was currently being impacted by AI were even more worried. A former TV and film extra who also worked as a DJ and had dreams of a writing career, Matthew, a 28-year-old white man, had recently completed a certificate in digital art and game production. He viewed the certificate as evidence of a “marketable skill” but was concerned about the advancements in AI art. “Some companies are totally against all the AI art, so I just have to hopefully get a job with those. And a lot of the consumers are against AI art, as well. So there's that. I'm competing against AI and trying to find a company who doesn't want to use AI,” he said. “I just have to learn how to harness it…. At the same time, I feel dirty as hell, even trying to learn that. I feel like I'm cheating… It just takes away some of the art… doesn't feel as human to me.”
Matthew is forced to walk a tightrope between “competing against AI,” and possibly finding a company that doesn’t want to use the technology, and “learn[ing] how to harness it.” Matthew's nominal decision in this matter is emblematic of the false freedom of choice that Bauman (2000) identifies with generalized precarity: individuals now bear greater personal responsibility for their actions within increasingly dire socioeconomic constraints. As part of his career development, Matthew had recently attended a conference for game developers and in the expo hall he was struck by how many tools were available for outsourcing developers’ work to AI, and how that affected employment stability.
“It makes the contract jobs a lot shorter. Instead of having a solid six months to work on 15 different things, it'll probably cut it down to a month with AI or two… I'm not an industry expert, but that's kind of what I saw when I was in California and it was terrifying. I went back to school and got something, and it's sad,” he said. “I'm broken-hearted. It's like I was very proud to have finished something, but at the same time, I'm like, for what, now that AI is going to take a lot of chances away from me?”
Matthew wasn’t alone in expressing the juxtaposition between feeling as though one needs to use AI, but also being scared of what it can accomplish. As a film editor and musician, Melissa, a 29-year-old white woman, was used to creating visual art. She’d recently used AI to generate art for an album cover for a band that she was working with. “I am not a painter, but I could quickly create a no copyright painting that fit perfectly with what I needed to create for this CD cover art,” she said. But the ease with which this “not a painter” could create art also gave her pause in terms of her career as a cinematographer, where her skills could also be replicated by untrained cinematographers with AI tools. “It's everything I trained in, and now people are just going, ‘oh, I want the person in front to be in focus’ and just tap the screen,” she said. “That scares me a little bit.”
The Hired Guns: “This is the Best Opportunity Right now for me That I can Think of”
While the Hardly Worried and the Heartbroken respondents were split in terms of their perception of the potential impact of AI on their careers, the Hired Guns were very clear that while AI may take away jobs, it was also offering them a financial opportunity in the process—and that they needed the money. The purported autonomy of AI tools depends on extensive human labor for training, verification, and guidance (Steinhoff, 2023). This labor is often performed in short-term gigs that offer low pay and minimal or no stability, creating an “increasingly precarious globalised workforce of underpaid pieceworkers” (Le Ludec et al., 2023:2). In the United States, however, many workers felt that AI-related gigs offered better compensation compared with other opportunities.
For instance, Kaitlyn, a 27-year-old white woman, was a former event musician working at an AI music start-up as an Algorithmic Composer. Her job was creating beats, the underlying pulse that provides rhythm for music. “I'm a musician who has a job because of AI,” she said. “My job is making beats for people to get for cheap…” For Kaitlyn, who had worried about paying her rent previously, beat-making was lucrative and offered her a chance to build a considerable savings.
However, most of the workers who had the most experience using AI to earn a living were the respondents who regularly operated under a sense that they must “hustle” for work (Cottom, 2020). Unlike their Hardly Worried and Heartbroken peers, who still lived in their childhood homes, or had multiple roommates or romantic partners who could help with bills, the Hired Guns generally described themselves as actively seeking multiple sources of income.
While these Hired Guns identified their primary occupation as working in the creative fields, they also told stories of being hired for short-term gig work to train AI tools that could be used to replace workers in creative industries. Michael, a 29-year-old white man, was a musician and writer, who often picked up odd jobs on CraigsList and whose band had recently started to gain some interest and additional paying gigs. Still, he had recently interviewed to help train the AI tool owned by one of the five big tech companies. “So they've got a lot of positions for writers to basically evaluate AI answers to prompts. So yeah, I'm waiting. I passed my test yesterday. I've done my homework on them. People go full-time with them. So the work is there.”
Michael described the job as evaluating AI's response to prompts, almost like a teacher providing feedback to a student on their essay. “I don't want the AIs to give people bad advice,” he said. “It's work, and it's stay-at-home. And that's what I've always been after.”
At the same time, Michael described himself as someone who is “do[ing] a lot of writing,” including working as a college essay editor for years. Asked if he's worried that he's helping to train the robot tool that may one day take his job, he's quick to note that it's not his problem. “We are so far removed from sufficient oversight of this program, of this fucking thing. We are completely removed from that. And I think that it would be pretty insane for me, who feels—I can't really go down that road. You understand?” he said. “Why would I make that my fucking problem? There's so many people in different positions. This is the best opportunity right now for me that I can think of. And if you can think of something better, let me know. But for me, and what I like, the kind of person that I am, and also, I have my creative outlet, so it's all just fueling that. So it's not like, ‘Oh, I'm making this my career.’ No, I'm contributing.”
Michael wasn’t the only creative freelancer who was regularly earning money off of the ascension of AI. Ethan, 39, a white male comedic actor, described much of his recent work as being focused on AI. “Whether it's like I'm actually filming stuff that's going to be used, where they're going to use my likeness and stuff, and I'm signing away my rights that you can do whatever you want with it. Or it's like I'll be doing an educational video for college professors about how to incorporate AI into the classroom,” he said. “Every frigging thing that I do these days is about AI.”
But while Ethan may be portrayed on videos as an expert espousing the virtues of AI, he's quick to note that that image is simply acting. He engages in substantial “emotion work” to set aside his personal feelings about AI in order to earn his pay (Hochschild, 1979). “I just want to tell people that AI isn't doing anything. It doesn't know anything… It's just repeating stuff that's been fed from pirated material that it learned from. This is one of the biggest hoaxes, almost, on society. It's like The Wizard of Oz. Don't look behind the curtain because there's nothing there,” he said. “It's sad to me because there's dedicated creative people that have devoted their life to these crafts of creative arts, and it's like they matter as much as a guy putting bolts in a thing on an assembly line.”
Indeed, Ethan's experience of being paid to tape videos proclaiming the benefits of AI, when he personally views it as a “hoax,” stretches the concept of emotional work. However successful Ethan has been setting aside his misgivings to earn pay, confronting them during the interview provokes a perceived loss of “face,” including expressions of shame for contravening his personal sense of morality (Goffman, 1967). “The creative arts are not being given any respect in the AI discussion. It's just, ‘what can we do for the almighty dollar?’ … You're just killing your soul. I don't know, it's a thing, but then again, I'll do all these shoots,” he said. “I basically was at the forefront of doing all these motion capture sessions for these different companies like Facebook and Google when they're developing all this stuff years ago. I'm doing all these motion capture sessions and machine learning, where I'm going in and I'm reciting 500 lines of dialogue to teach the AI. I feel personally responsible for some of this crap.”
Much like Michael, Ethan also uses phrases that invoke the Nuremberg trials and the defense of “just following orders” to discuss his experience of working to support the wholesale roll-out of AI. “And I always said, I feel like the guy, the construction worker laying the bricks for the gas chambers and I go, ‘hey, I just got this lucrative job laying brick.’ And they go, ‘Ethan, don't you know what they're going to do? Once you build that thing? Everyone's going to die.’ And I go, ‘I said, I am laying brick.’ It's like you don't want to know,” Ethan said. “But when they're dangling this money at you, they're like, ‘hey, will you take $700 to come and wear these motion capture things on your body and do a bunch of weird shit?’ I'm like, ‘yeah, I'm in.’ Because if I don't do it… And again, it's like some Nazi shit, because you're like, ‘I was just taking orders.’ If you don't do it, somebody else will… So they're going to film what they want anyway, and you may as well get money. Because then not only if I don't do it, not only is the AI taking over the world, but I'm also dirt poor, so I may as well just go along with it.”
While Ethan's description of laying the bricks that are then used to destroy lives may seem like hyperbole, he has firsthand experience being told by producers that his work will be used to replace human actors. A recent gig involved participating in an AI recording session conducted by a streaming giant: They just had us in a crowd on a soundstage, and had us changing into all different outfits, and just getting shots of us just looking around, and standing there, and doing little movements, and stuff. And they go, “Yeah, this can just be used in any series, anything we own, anything we produce, we can just key you in, and we can also change your likeness digitally, so it doesn't just look like the same people over and over. And yeah, we can just put you in anything, and we just own it.” So I mean, in one part you're like, “That sucks.” But then the other part you're like, “Oh, all this money and all I had to do was stand there.” But I guess maybe I should have refused or had principles or something. I don't know.
Michael and Ethan are well aware of how their participation may be used to strengthen AI and may impact their fields. Yet their financial situation is such that both men feel the need to participate in the AI-funded work and they don’t appear to be alone in their willingness to contribute to the training and recordings needed for AI to function. And while it's unclear how many people were part of the “crowd” at the streaming service's recording session, it certainly seems that workers who are struggling are especially vulnerable to the offer of hundreds of dollars to “go stand in a room for three hours and just goof around.”
Their responses suggest that any effort to fight the growing usage of AI will need to address not just career interests, but also financial precarity. As long as these companies are offering workers hundreds of dollars, people will feel the need to participate. After all, as Ethan summed up his decision making: “if I don't do it, not only is the AI taking over the world, but I'm also dirt poor, so I may as well just go along with it.”
Fighting Back: “Other People can Take Advantage of my 20 Years of Work”
We need to caution that not all creative freelancers who were participating in the growth of AI were indifferent to the technology. Especially if workers were participating without their knowledge, or without their consent, there was a notable sense of anger. For instance, Jason, a 37-year-old white man and for-hire filmmaker, also felt that his work was too complex to be replaced by AI. However, some of his clients had taken his scripts, uploaded them into ChatGPT and other AI tools to condense and edit the work, and then sent it back to him. “I feel violated because I don't like the idea of my work and my creativity being now part of the model, being trained, so that other people can take advantage of my 20 years of work, NYU degree and all this stuff. And all the books I've read and all the experiences I've had, so that someone else can spit out something that's like my work, for free.”
This sense of having his years of experience being taken advantage of, for free, also motivated Jason to fight back against an effort by Vimeo, a video sharing and hosting platform, when it emailed users about using their video content for training AI. “I emailed everybody I know, I put a post out on social media: ‘We must all respond with the most violent language possible to shut this down.’ Because they're trying to see what they can get away with,” he said. “Because it's profitable for Vimeo to take all of our data that we own, that we pay for, to send over to whoever it is, to make an AI technology that Vimeo profits from and that we don't.”
It's important to note here that even for Jason, who describes himself as “incredibly anti-generative AI,” part of the issue with the potential Vimeo AI deal was that the plan was “to make an AI technology that Vimeo profits from and that we don't.” Even for workers who are against the tools of AI, and who feel that their work is too complex for them to be replaced, are still cognizant that AI can be used to make money off of them—and are concerned about being left out of the profit equation. For Jason, who had seen clients feed his work into AI systems, this further cemented his sense that AI could take advantage of his years of experience and “spit out something that's like my work, for free.”
Discussion
This study responds to calls for sociological attention to economic interests (Martin & Lembo, 2020; Swedberg, 2005). The advent of generative AI will have uneven consequences for workers depending on their occupations and even the specific tasks required of them to perform their jobs (Acemoglu, 2025; Acemoglu & Restrepo, 2024). Workers’ rational assessment of how they are likely to be affected by AI may therefore be expected to depend heavily on the specificities of the work they perform. This reasoning should be particularly observable in how creative workers understand their interests in relation to AI, due to the heterogeneity of job types, precariousness, income levels, and other work characteristics that shape these workers’ conceptions of themselves in relation to other creative workers (Skaggs & Aparicio, 2023).
This study shows that creative workers express differing understandings of their interests even when they perform similar types of work. These understandings do not differ primarily according to the job characteristics or economic situation of respondents, but instead based on personal assessments of their own capabilities and choices to morally problematize (or not problematize) the economic consequences of AI. This variation reflects the manner in which economic precarity and uncertainty about the future course of AI technology raise the stakes of anticipating potential threats from sources like labor displacement from AI, while simultaneously making a wide range of predictions appear plausible. Understanding how or even whether personal and collective interests are under threat is difficult when a diverse array of “AI imaginaries” exists among varied stakeholders with their own interests and disparate perspectives in relation to AI (Richter et al., 2025). This dissensus is visible among the creative workers in our sample, and presents a potentially significant barrier to the articulation of group interests and identity in the face of AI. On the other hand, these expectations are unlikely to remain static, and heterogeneity also reflects the “shifting solidarities” (Morgan & Pulignano, 2020) that accompany major changes in social circumstances. The imagined futures of AI are so uncertain and fractured because they are still being shaped. Workers in the creative industry and beyond may be uniquely positioned to intervene while the politics of AI are still in their infancy (Grohmann et al., 2025).
This finding highlights an important function of labor organizations: coordinating expectations about the future and perceptions of interest in relation to those expectations. The uncertainty of AI's impacts, coupled with the economic and cultural responsibility for individuals to personally imagine those impacts, means that workers will likely interpret their own circumstances in wide-ranging and potentially idiosyncratic ways that impede conceptions of shared interest. Labor organizing among creative workers and other workers likely to be affected by AI can bridge these heterogeneous, individualized understandings by advancing a clear set of expectations around how workers will be collectively affected by AI.
Conclusion
While the risks of AI may be felt in various industries, the potential threat of AI has been particularly salient in creative fields, and received extensive media attention due to the 2024 strikes by members of the WGA and SAG-AFTRA. Yet, not all creative work may be equally at risk, and, drawing on the concept of “sociotechnical imaginaries” that inform workers’ perceived interests in relation to the future, not all workers may view AI as equally risky. As a result, in this paper, we ask, how does the anticipation of—or actual experience with—AI exacerbate perceived differences in interests among occupationally heterogeneous creative workers?
To answer this question, we draw on interviews with 50 workers in creative industries, many of whom report first-hand experience working with AI tools. Our study is novel for its focus on workers in creative industries, and for its timing, with data collection occurring six months after the 2023 WAG and SAG-AFTRA strikes concluded.
We find that respondents fall into three ideal types in their responses to AI. The Hardly Worried viewed themselves as uniquely insulated from the detrimental consequences of AI, often describing their work as complex or requiring a level of human interaction that AI simply couldn’t offer. For these workers, the risks of AI “coming for” jobs, was limited to the more simplistic jobs of other people, such as clothing catalog models or schedulers. By comparison, the Heartbroken respondents viewed AI as potentially taking work away from themselves and others. For these workers, there was a fear that AI was “going to take a lot of chances away from me” and disrupt hard-earned skills.
The final broad category of workers, the Hired Guns, had first-hand experience being paid to train what they believed were AI systems or tools. When these workers remained nonplussed about the risks of AI, it was often because they remained convinced that any future work loss would not affect them. However, a subset of these workers were aware that they were potentially training their replacements, but brushed aside such concerns with claims that they were “just following orders” or that if they didn’t do the work, someone else simply would. Finally, we note a sense of anger from creative workers who discover that their work is being used to train AI without their permission or appropriate remuneration, taking advantage of years of experience and education for free.
To be fair, not all creative workers view themselves as equally at risk, and the workers who are helping to train AI programs may view themselves as being unable to afford to care about the future impact. Indeed, there were no patterns in working conditions, contracts, or collective bargaining units, further supporting heterogeneity. Workers who have already seen AI take the jobs of friends or colleagues, or who have learned about such experiences from their unions or the media, may be more willing to organize against future AI incursions. Raising awareness of the risks of AI for creative workers—much like the media coverage of the SAG-AFTRA and WAG contract negotiations—may further help to prepare workers and garner support for efforts to regulate or reduce the use of AI.
To reduce the ability of industry or researchers to hide the training of AI tools, non-disclosure agreements for research participants should be banned. While they may be useful for the salaried, employed researchers who are creating the models, it is exceptionally unlikely that the worker whose actions, or likeness, is being fed to the AI tool will learn enough during a few hours of work to inform and unjustly enrich a competitor. Additionally, while deception or limited informed consent may be used by researchers who are studying actions where the Hawthorne effect may be a risk, individuals who are training AI tools should be provided with comprehensive consent forms that outline, in clear language, what types of impact the tools could have, especially in terms of future work. If people are literally training their future replacements, they should be informed of that risk.
Additionally, it is obvious that some workers who are training AI tools are doing so simply for the money. Additional employment opportunities, such as increased funding for the arts, may seem like a pipe dream, especially in the current political environment, but could be a valuable investment in helping people to earn extra funds without having to train their replacements. Additional follow-up research with these workers may also be advisable, especially to see if respondents’ perspectives on training early interactions of AI systems may have changed. It would be interesting to see if the Heartbroken respondents have found new work solutions or if the Hardly Worried participants are now veering into worry. While many experts believe that fears about AI are often overblown, it will be interesting to see how the interest identities of creative workers are impacted in the future, especially if indeed, “AI [ends up] taking over the world.”
Footnotes
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This material is based upon work supported by the National Science Foundation under Grant Number 2029924 and 2241780.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.