Reflecting on cultural labour in the time of AI

Unproductive labour

The first ‘labour’ perspective in cultural policy discussion emerged in the form of ‘cost disease theory’ in the mid-1960s, the heyday of industrialisation that was also defined by the significant expansion of mass culture and media in Europe and the US. According to Baumol and Bowen’s (1966) Performing Arts: An Economic Dilemma, labour-intensive live performing arts form a productivity-stagnant sector. That is, artists are hardly substituted by machines so the performing arts sector’s possibility of making productivity gains is very limited; and similar circumstances are found in various labour-intensive service industries. Meanwhile, the rising productivity in other sectors which enjoy technological developments leads to wage increases, driving the wage increase in the performing arts sector, too. Hence, performing arts organisations whose income is limited would suffer from cost problems (‘disease’) and need external subsidies.

This theory was critiqued by various scholars (e.g. Peacock, 2000; Towse, 2019): artists’ earning does not notably increase; wider dissemination via broadcasting or recording can offer additional income (Baumol himself noted this. See Baumol, 1985); many commercial theatres survive without public funding; and after all, the lack of consumers is the most fundamental cause of the arts organisations’ financial problems. Nevertheless, the cost disease theory underpinned paternalistic cultural subsidy by persuading the government to support ‘unproductive’ arts sectors vis-à-vis commercial cultural and media industries. Although the key concern was performing arts, its argument looked relevant to other types of ‘labour-intense’ arts sectors. Indeed, it justified the establishment of a federal arts funding agency in the US in the mid-1960s and determining its key objective of the financial stability of arts organisations (Wyszomirski, 2013: 158). In the UK and elsewhere, policymakers and the arts sector used this theory ‘in the pressing for ever more generous subsidies’ (Throsby, 1994: 15) until its demise due to the rise of the creative industries discourse in the late 1990s.

From the perspective of cultural labour, we can notice that the cost disease theory implicitly recognised cultural labour’s complexity and tacit nature by arguing that it is hard to be replaced by machine. Ironically, this underpinned the economic reduction of such labour to a mere matter of low productivity. However, the arrival of generative AI means that many kinds of cultural labour can be automated, leading to potential productivity increases depending on the sector, with live performing arts still remaining automation-proof. This reminds us of Baumol and Bowen’s (1966) observation of the differing effects of technology on different industries and makes us wonder how the impact of generative AI will be manifested in terms of productivity gains – or lack of them – across different sectors within the broader cultural and creative industries. Yet, we should be mindful that such an inquiry would carry an overt danger of economically abstracting human creativity and cultural labour as the cost disease theory did.

Creative labour

In the late 1990s and 2000s, new types of cultural policy emerged to facilitate the ‘creative industries’. As radical reversal of the cost disease theory, the idea of creative industries glorified creativity and thus cultural labour, which is the key factor of creative production. Therefore, it is not surprising that cultural, creative and media workers’ creativity, skill and talent occupy the core concern of creative industry policy. Policymakers came to see them broadly as ‘human capital’, which was claimed to be ‘by far the most important form of capital in modern economies’ (Becker, 2002: 3). If physical capital, which is often concentrated on few people, was the main engine of the industrial society, the growth of post-industrial society is believed to depend on human capital and its wide dissemination (Galor and Moav, 2004). The human capital perspective rightly notes the tacit aspects of creativity – creativity is embodied in cultural workers and is difficult to codify or transfer – and the complexity in the creative process – which is often analogised to a black box. This explains why the UK government’s creative industry policy focused on fostering current and future cultural workers who embody creativity through investment in creative education, training and clusters (Banks and Hesmondhalgh, 2009; Jayne, 2005). Seemingly, the recognition of tacit dimensions of creativity hindered policymakers from developing policies that support specific industries or organisations. Instead, they expected creativity as human capital to be accumulated via education and training and would be widely spread via creative clusters and networks across the country. Aligned with the thinking that creativity can potentially be found in every labour (McGuigan, 2010), the creative industries discourse had some potential to catalyse broader discussion on the significance of labour and strengthen the call for more labour shares of economic gains within the context of the increasing inequality.

However, such a potentiality was not realised as policymakers were preoccupied with advocating the cultural sector’s immediate economic impact. This deeply resonated with the neoliberal and individualised understanding of cultural workers: as human capitalists, they would be entrepreneurial and self-responsible for skills development and maintenance, take risks and generate further capital in the form of IP (McRobbie, 2006). In short, the glorification of cultural workers and their creativity turned to a celebration of their economic power in the form of the creative industries’ revenue and contributions to GDP and regional economic development. Moreover, policymakers ignored the problems of cultural labour, such as prevailing precarity and poor working conditions.

In retrospect, the creative industries discourse marked a ‘creative turn’ in cultural policy, highlighting the importance of cultural workers as bearers of creativity, bringing new investment into the cultural sector and increasing the sector’s visibility. Yet, it treated creative labour as if it is capital, weakening labour as both a concept and a social force (Lee, 2017). The discourse of creative industries was already debunked by the socio-economic reality of cultural workers, which is characterised by precarity. It will likely be further questioned as generative AI shows how technology easily emulates creativity (or so-called human capital) and what real capital can do, such as the massive-scale accumulation and concentration of creative capacity in global AI platforms. At the same time, generative AI may further diminish the value of labour by demonstrating the possibility of creative production without labour and its ability to fill gaps in cultural labour (‘human capital gaps’) (The Economist, 2024). By revisiting the creative industries discourse, we can realise that the policy perspective of creative labour from the late 1990s was deeply situated in the broad political-economic context of the assimilation of labour to capital, and such a context is likely to continue into the time of generative AI.

Precarious labour

Unlike policymakers, critical scholars paid great attention to cultural labour issues since the 2000s. They unglorified creative labour by pointing to precarious work, the lack of social security, the absence of work–life balance, and (self-)exploitation against the backdrop of neoliberal economic restructuring such as outsourcing, the individualisation of risk and the weakening of unions (e.g. Banks, 2017; Gill and Pratt, 2008; Hesmondhalgh and Baker, 2013; McGuigan, 2010; McRobbie, 2016; Reckwitz, 2017). Nevertheless, there were no notable policy changes until the Covid-19 pandemic, during which the hardship of cultural workers became an acute policy issue (de Peuter et al., 2023; Salvador et al., 2022; UK Parliament, 2022). Multiple manifestations of insecure cultural labour, such as the uncertainty in employment, the lack of control over working conditions, low income and the lack of regulations, were criticised (Comunian and England, 2020). In particular, ‘self-employed’ (freelancing) emerged as the most vulnerable category of cultural labour because state protection of workers’ rights is still under the legacy of industrial society; the self-employed are not seen as a ‘worker’ by employment laws; and they do not have workers’ basic rights (Taylor, 2017). Those with minority backgrounds tend to suffer more from such instability, and it was also argued that the structural inequality in the workforce makes creative work a ‘bad’ work (Brook et al., 2020).

Understandably, cultural workers are urgently calling for remedies. For instance, cultural trade unions and campaign groups in the UK demand more social recognition of the value of cultural labour, seeing it as an essential condition for tackling the precarity issue. Their other demands include fair working conditions and remuneration; regulations on freelancers’ rates; freelance workers’ participation in policymaking; minimum income guarantee or universal basic income; better social protection; and easier access to unemployment benefits (Artists’ Union England, 2020; Equity, 2021; FMTW, 2022; Musicians’ Union, 2022; WGGB, 2020). Policymakers belatedly acknowledged the precarity of cultural labour and have started to view it as an issue of employment/labour, social security, welfare policies as well as cultural subsidy.

The discussion of precarity has functioned as a very effective counter-argument to the glorification of creativity in the creative industries discourse, by seeing cultural labour as a matter of making a living rather than a matter of creativity. However, I argue that such socio-economic focus tends to significantly limit our exploration of generative AI’s impacts on cultural workers and their labour. It is because, in the age of AI, artists’ and cultural and media workers’ livelihoods would depend more heavily on how their creative roles, identity and rights are affirmed and their creativity is socially valued. The uncertainty around these would intensify the key indicators of insecure cultural labour – especially the insecurity in income-earning opportunities, retaining an employment niche, gaining/updating skills and having adequate stable income (Standing, 2020: 12).

Reconfiguration of creative roles

As Kaplan (2022) notes, cultural workers will co-exist with generative AI and this makes inevitable to reconfigure the creative roles that cultural workers have played. In this regard, we can identify three different views. First, cultural trade unions and guilds see generative AI potentially replacing human creative workers (UK Parliament, 2023). Writers such as genre-writers and journalists are seen as vulnerable (The Author’s Guild, 2022), and it is observed that ‘AI generated content could flood the market and further erode authors’ earnings’ (ALCS, 2023). Similarly, Equity (2022), the UK actors’ union, witnesses job loss already happening to audio artists because of AI voice services. AI is also likely to compete with cultural workers in digital art, commercial photography and marketing-related writing. The second view is that human creative labour is not replaceable by AI service as human is good at exploring the frontiers of the unknown (‘ill-structured problems’ in Herbert A. Simon’s words) whilst AI is bounded by existing data. While this belief is being tested, it implies that facing AI challenges, cultural workers will be motivated – or pressurised – to be more unconventional and innovative and also become keen to express ‘human-ness’ in their works. Emphasising the physicality of their works and unmediated interaction with audience – artistic labour described by the cost disease theory – might be a crucial part of this endeavour.

Meanwhile, the third view is that human and AI’s creative roles are complementary. Some artists actively use AI as a way to explore new kind of creativity and test the idea of sharing creative roles with AI (Ashton and Patel, 2024; Navas, 2023; Ploin et al., 2022). It is also interesting to observe how cultural workers’ creative roles are being reshaped. For example, the agreement between Hollywood studios and the scriptwriters’ union allows scriptwriters to use AI as a tool. ¹ Indeed, some Hollywood scriptwriters use AI to ‘come up with ideas, or spin out potential plotlines, or to develop characters’ (NPR, 2023). Similarly, British TV producers regard AIs such as ChatGPT and Perplexity as brainstorming tools that are good at suggesting format ideas, titles, talents and marketing strategies (The Big Creative UK Summit, 2024). These cultural workers try to become ‘meta-creative’ (Bruch, 1988) ² : that is, being creative about managing creative process, evaluating AI outputs and combining human and AI creativities. However, we have yet to understand the nature of meta-creativity, how it changes the existing cultural roles and working conditions, and how it will affect cultural workers’ creative capacity from the long-term perspective.

In short, generative AI contests the human monopoly of creativity, dis-embodying it from human cultural workers and causing a lot of uncertainty about their creative roles. The use of AI may make some creative roles redundant (and some roles more productive). Some cultural workers may seek more innovation to compete with AI, whilst others use AI for innovation and seek meta-creativity. Such threats, opportunities and needs are likely felt unevenly across different sectors and across different roles. This puts pressure on cultural policy to draw a holistic picture of AI impact, reconfiguration of creative roles and changes to creative production process in order to identify areas for state regulation and support.

Lack of control over copyrighted works

Another dimension of creative precarity faced by cultural workers is their lack of control over generative AI’s access to copyrighted cultural content. The increasing concern with their ‘rights’ as cultural creators is aptly demonstrated by the series of copyright lawsuits in the very recent years. An example is the lawsuit filed by three artists against three AI companies for the latter’s use of the artists’ work in AI datamining and training without consent, compensation and credit (The New Yorker, 2023). Similarly, the New York Times brought Open AI and Microsoft to the court for their unauthorised use of its articles for AI datamining (The Guardian, 2023). The concern is echoed by an AI Open Letter (2023) signed by 2,400+ journalists, writers and artists globally who see AI datamining as ‘effectively the greatest art heist in history’. ³ Although there is a view that the simple equation of datamining with theft is problematic because all cultural creators rely on the previous ideas, genres and concepts, ⁴ this view is gaining less sympathy.

The current lawsuits indicate cultural workers’ anxiety about the degradation of their activities to data production for AI and their attempt to slow AI’s advancement. Policymakers are called to urgently respond to the rising anxiety and worries held by cultural creators and media companies by finding ways to regulate AI’s use of copyrighted cultural content. Such regulation is likely to result in the institutionalisation of AI’s fee-paying access to copyrighted materials, expanding the cultural and media industries’ copyright business model. However, we should be reminded that the actual economic benefit of copyright for cultural workers is always questionable. Moreover, strong copyright protection and AI companies’ remuneration for copyright owners may reinforce the integration of cultural production into data capitalism (Sadowski, 2019) by equating cultural labour, copyright generation and data production. This means that potential policy response should be more than a simple expansion of the copyright business model and consider indirect ways for AI companies to support cultural producers.

Creative identity separable from cultural labour

Generative AI allows not only the dis-embodiment of creativity from cultural workers but also the separation of their identity from their labour. Thus, who will control their creative identity becomes a critical issue. AI can imitate human writers’ distinct writing style (The Author’s Guild, 2022) and easily digitalise and appropriate actors’ voice and appearance. It is widely known that film actors and media celebrities – as well as politicians – are exposed to the serious risk of deepfake. Furthermore, the unprecedented separation of personal and creative identity from cultural labour leads to job losses and human creators’ (and performers’) lack of control over their own identity. According to Equity (2022: 7), 61% of surveyed members now find their job consisting of voice synthesis/capture, avatar creation and performance capture, but their biometric data, such as voice and movements, tend to be owned by the commissioning AI firms forever. Similarly, Hollywood actors recently rejected the studios’ suggestion that the latter would pay background actors for a daily rate currently under US$200, scan their body and use the image for the rest of the film/drama production, own the scanned image for perpetuity and use it whenever they want without actors’ consent or further compensation for them (CNN, 2023).

Hence, actors and performers are urgently demanding policy intervention, such as the introduction of ‘personality rights’ to protect their image, likeness and voice and fight deepfake (Equity, 2022). Such policy intervention is morally grounded and is urgently needed. Yet, the paradox is that the proposed personality rights can also reinforce the identity-labour separation and institutionalise the trade of personality, including artists’ biological features. For some actors, pop singers and media celebrities as well as cultural companies, the labour-identity separation means a new avenue for income generation: for example, the right to voice of James Earl Jones (voice of Darth Vader) has been sold to Disney (The Economist, 2023) and there are similar cases. Mirroring the unequal distribution of benefits from copyright ownership, the economic benefit of personality rights is likely to be unevenly distributed.

The identity-labour separation challenges the existing human capital perspective that assumes the embodiedness of creativity, skills and talent (and creative identity as a reflection of these) in cultural workers. At the same time, it can also strengthen the view that treats creativity as ‘capital’: creative identity can now become a tradable product and then part of the capital asset of the buyer, further blurring the distinction between labour and capital. Seemingly, such phenomena will diminish the role of labour over time while increasingly commodifying creative identity of popular performers or media celebrities.

Uncertainty in audience responses

Another source of creative precarity is the uncertainty over how audiences perceive AI-generated creative content vis-à-vis human-made one and if and why they value the latter more. According to the existing research (e.g. Bellaiche et al., 2023; Fortuna and Modliński, 2021), people attach more value to cultural content that ‘they think’ are human-made and view AI-made works as less likeable and ‘less worthy and less profound’ (Bellaiche et al., 2023: 7). Many people seem to agree with the view that AI-creation might be regarded as ‘art-looking art’ without ‘the ingenuity, the personal vision, the individual sensibility’ ⁵ so the use of a watermark for AI-generated content may significantly influence their response. A pivotal question for both researchers and policymakers is why people value human-made cultural content more and if this will continue even if the use of AI is more popularised in the future. According to Bellaiche et al. (2023), people currently hold a nuanced understanding of creativity, in which effort, the embodiment of creativity and time spent on creation are seen as key criteria. Another useful finding is that people tend to more positively consider robot-created art when they observe the robot’s art-production process (Chamberlain et al., 2018). These are interesting as they imply that people’s perception of creativity appears closely associated with ‘labour’ elements in cultural creation. This encourages us to explore how audience connect other key aspects of cultural labour, for example, autonomy and affect (Banks, 2007; Gill and Pratt, 2008; Hesmondhalgh and Baker, 2013; McKinlay and Smith, 2009), to their evaluation of human-made cultural content.

However, it is difficult to know how such ‘human-ness’ or human elements in cultural labour can actually be felt by audience when not only lay people but also cultural experts hardly distinguish human creation from the AI outputs if they are presented digitally. This raises serious problems of provenance and attribution. Furthermore, the use of AI occurs on a spectrum – AI can function as a brainstorming tool for TV production, can compose a piece of music based on the composer’s intervention at multiple stages, or can generate a poster by a simple prompt – but the audience cannot tell these differences. Equally intriguing is that generative AI produces cultural content via ‘sub-symbolic’ process (i.e. not relying on symbols shared by people in society, See Michell, 2019), and its content is a statistically generated pattern (Pasquinelli and Joler, 2020). Yet, the audience consumes the content via ‘symbolic’ process by finding and generating meanings. This fundamentally challenges our perception of communication and the view that the essence of cultural work is the communication of meanings (Williams, 1965: Chapter 1; Hesmondhalgh and Baker, 2013).