OIT - Generative AI and the media and culture industry

 ILO Brief 1 Generative AI and the media and culture industry

 Research Brief 2025 Generative AI and the media and culture industry Janine Berg; Pawel Gmyrek; Margherita Licata; Tinovimbanashe Gwenyaya; Martín Sánchez Freytes Scaiano

 Technological Transformation: AI, particularly Generative AI (GenAI), is transforming how tasks are performed in journalism, music, film production, and other creative domains. Unlike previous technologies, GenAI influences not just production but also creative and decision-making processes, challenging traditional notions of human agency and oversight in creativity.  Occupational Impact: Media and culture jobs show varying exposure to GenAI. Physical and performancebased roles (e.g., dancers, choreographers) remain largely human-centred, while roles like those of journalists, writers, and translators face high exposure. GenAI augments some roles while automating others, potentially leading to evolving job profiles emphasizing oversight of AI-generated content rather than direct creation.  Employment and Skills: Traditional skills like critical thinking, creativity and ethics are becoming more essential, complemented by new competencies in AI tool management and digital literacy. Nevertheless, the increased use of AI in the sector may result in job displacement, revenue losses, and reduced demand for traditional roles, necessitating strategic adaptation.  Policy and Governance: The integration of AI calls for frameworks to address ethical concerns, workforce transitions, and labour protection issues, including fair remuneration and compensation models. Social dialogue, including collective bargaining, can play a crucial role in addressing concerns around the "3Cs" (compensation, control and consent) as well as improving working conditions. Collaborative efforts involving governments, industry stakeholders, employers’ organizations and trade unions can mitigate risks and foster a human-centred AI governance system.  Introduction

(compensation, control and consent) as well as improving working conditions. Collaborative efforts involving governments, industry stakeholders, employers’ organizations and trade unions can mitigate risks and foster a human-centred AI governance system.  Introduction “Media and culture” is a diverse sector that consists of a range of sub-sectors, including: creative, arts and entertainment activities; publishing; motion pictures, video and television program production; sound recording and music publishing; programming and broadcasting; and the activities of libraries, archives, museums and other cultural institutions.1 Various occupations make up the workforce of the different subsectors: the “creatives” encompassing photographers, visual artists or musicians, among others; and “technical” occupations supporting creative work, such as audiovisual workers and sound engineers, in addition to other professionals in the media or cultural space. New professions have also emerged in

1 ILO, Report for the Technical Meeting on the Future of Work in the Arts and Entertainment sector, TMFWAE/2023, 2023. Key points ILO Brief 2 Generative AI and the media and culture industry

response to technological advancements , such as “influencers” who develop user -generated content on major digital platforms. Throughout its history, the media and culture sector has been shaped by technological change. Print, film and photography were born from technological advances, and with each invention the media and culture sector has been reshaped, as have its occupations. Over the past 25 years, some of the more consequential shifts have stemmed from the internet and the emergence of technology platforms, which have transformed not only how consumers engage with content, but also how content is produced, disseminated, and remunerated.2 The global reach of the platforms has meant that independent artists, musicians, and writers have been able to gain exposure without relying on traditional media gatekeepers. At the same time, it has also meant that traditional outlets such as newsrooms in broadcast and print media have had to re-organize internal workflows to integrate technologies, whereas some sectors such as recorded music and

that independent artists, musicians, and writers have been able to gain exposure without relying on traditional media gatekeepers. At the same time, it has also meant that traditional outlets such as newsrooms in broadcast and print media have had to re-organize internal workflows to integrate technologies, whereas some sectors such as recorded music and motion pictures have seen the emergence of new actors (i.e. content aggregators, such as streaming or content -sharing platforms). Where employment has continued, remaining staff are often expected to handle a broader range of tasks, requiring digital skills such as social media management and data analytics. Advances in artificial intelligence (AI), and more recently in generative AI (GenAI), represent the latest chapter in a long line of technological shifts that have shaped the creative process and redefined artistic expression . While previous innovations, such as the invention of photography in the late 1800s, the rise of the hand-held film camera that fueled the French New Wave, or even the shift to digital film and videos (i.e. though streaming), profoundly influenced creativity and displaced earlier forms, these shifts did not fundamentally replace human oversight and originality. AI, by contrast, introduces a new level of autonomy , enabling machines to take on significant portions of creative and decision-making processes, including using creators’ work, likeness or images to produce new content (see box 1). Such advances are poised to affect content creation such as scriptwriting, music composition, video production and journalism;3 but also distribution, by allowing the prediction of viewers’ and listeners’ preferences in streaming platforms; 4 or increasing accessibility and reach through AI-supported translation and voice-over systems.5  Box 1 – Understanding AI and Generative AI in the Entertainment Sector Artificial intelligence has been defined as “the science and engineering of making intelligent machines”.6 Machine learning is a field of AI that uses algorithms to enable systems to learn and make predictions based on data. GenAI is a subset of machine learning that “can create something new”.7 In the creative process, GenAI can generate text, images, music, and videos. It uses complex algorithms and large-scale data sets to produce content that mimics human creation.

a subset of machine learning that “can create something new”.7 In the creative process, GenAI can generate text, images, music, and videos. It uses complex algorithms and large-scale data sets to produce content that mimics human creation. GenAI uses Large Language Models (LLMs), which are trained on diverse data sets including books, articles, films, scripts, music scores, and other cultural artifacts. By scraping this diverse cultural data, LLMs learn the nuances of language, style, and context, enabling them to generate content that resonates with human experiences and cultural norms. The exact sources of data used by LLMs are extensive and varied, often incorporating publicly available information and specific datasets curated for training purposes. This brief explores some of the possible effects of AI, particularly GenAI, on the media and culture industries and includes an assessment of the potential exposure of certain occupations within media and culture to generative AI technology, that is to say how jobs in the media and culture industries might be affected by the growing use of AI tools that can perform creative tasks. It includes case studies of the uses of GenAI technology in particular sub-sectors and probes issues around

2 ILO, The Future of Work in the Arts and Entertainment Sector – Report. 3 Trootech, “The Impact of Gen-AI on Media and Entertainment Industry”, 2024 4 Zeta, “The Impact of AI on Digital Content and Streaming Platforms ”, 2024 5 EBU, "You can’t look at GenAI too fearfully. You have to find as many opportunities as you can ", 2024 6 John McCarthy, (2007), “What Is Artificial Intelligence?” mimeo, Stanford University. 7 Bernard Marr, “The Difference Between Generative AI And Traditional AI: An Easy Explanation For Anyone ”, Forbes, 24 July 2023. ILO Brief 3 Generative AI and the media and culture industry

employment, skills, productivity, working conditions and remuneration. It also discusses polic y measures to mitigate possible harms, such as worker displacement, income loss, and the protection of creatives’ intellectual property, while also highlighting the opportunities AI presents.

Generative AI and the media and culture industry

employment, skills, productivity, working conditions and remuneration. It also discusses polic y measures to mitigate possible harms, such as worker displacement, income loss, and the protection of creatives’ intellectual property, while also highlighting the opportunities AI presents.  Employment in media-related occupations Media and culture jobs account for 0.96 percent of total employment around the globe, which corresponds to 32.7 million jobs worldwide. To derive this estimate, we identify the principal occupations in “media and culture”, as defined by the International Standard Classification System (ICSO) (Table 1). ISCO is a statistical framework that classifies occupations according to the tasks and duties undertaken in the job, allowing comparisons across countries and over time.  Table 1. Media-related occupations in ISCO-08 classification (4-digit level) Title ISCO-08, 4-digit code Graphic and multimedia designers 2166 Advertising and marketing professionals 2431 Public relations professionals 2432 Web and multimedia developers 2513 Authors and related writers 2641 Journalists 2642 Translators, interpreters and other linguists 2643 Visual artists 2651 Musicians, singers and composers 2652 Dancers and choreographers 2653 Film, stage and related directors and producers 2654 Actors 2655 Announcers on radio, television and other media 2656 Creative and performing artists not elsewhere classified 2659 Photographers 3431 Gallery, Museum and Library Technicians 3433 Other artistic and cultural associate professionals 3435 Broadcasting and audio-visual technicians 3521 Telecommunications engineering technicians 3522 Library Clerks 4411

Based on the occupational codes listed in Table 1, we rely on labour force survey data from ILO’s harmonized microdata repository and use the ILO global employment estimation model to derive employment estimates at the global level, by ILO Brief 4 Generative AI and the media and culture industry

Based on the occupational codes listed in Table 1, we rely on labour force survey data from ILO’s harmonized microdata repository and use the ILO global employment estimation model to derive employment estimates at the global level, by ILO Brief 4 Generative AI and the media and culture industry

country income groups and by region. 8 As Figure 1 shows, the share of jobs in media and culture increases according to countries’ per capita income, reflecting the greater economic diversification of higher-income countries. In high-income countries, media and culture jobs represent 1.76 per cent of total employment (10.7 million jobs), compared to just 0.17 per cent of employment (0.4 million jobs) in low-income countries.  Figure 1. Media related jobs: totals and as share of total employment by income group, region and sex

 Occupational exposure to generative AI The ability of GenAI to produce written, audio and video content in response to simple prompts has raised growing concerns over employment within the media and culture sector. To assess the potential exposure of GenAI on the selected media-related occupations presented in Table 1, we apply the methodology developed in the ILO Working Paper 96. 9 The methodology uses AI and human verification to assign a score between 0 and 1, reflecting the potential for automation by generative AI technology among the tasks of individual occupations. Table 2 presents, as an example, the international definition o f a journalist (ISCO -08 code 2642), as well as the typical tasks associated with the profession. The tasks associated with each occupation form the basis of the prediction of potential automation scores, with the procedure repeated for each media-related occupation.

8 Data provided by David Bescond (ILO/STATISTICS). 9 Pawel Gmyrek, Janine Berg, and David Bescond, Generative AI and Jobs: A Global Analysis of Potential Effects on Job Quantity and Quality , ILO Working Paper

96 (Geneva: ILO, 2023), https://doi.org/10.54394/FHEM8239 . ILO Brief 5 Generative AI and the media and culture industry

 Table 2. Sample of tasks and definitions from ISCO and GPT-4

Figure 2 presents the distribution of exposure scores assigned to individual tasks for each occupation. Low scores are assigned to tasks that do not have the potential to be performed by GenAI, while high scores are assigned to tasks that have the potential to be performed by GenAI technology. On the horizontal scale ranging from 0 to 1, scores between the range of 0.5 and 0.75 are considered to have a medium level of exposure, and scores above 0.75 as high exposure. We stress that this represents the top threshold of the theoretical concept of exposure. In practice, the actual automation of tasks with high scores is likely to be significantly lower, due to constraints related to the technical feasibility of the deployment of technology, the costs of deployment and other limiting factors. ISCO-08 code: 2642 Definition in ISCO -08: Journalists research, investigate, interpret and communicate news and public affairs through newspapers, television, radio and other media. Tasks in ISCO-08 Collecting local, national and international news through interviews, investigation and observation, attending public events, seeking out records, reviewing written work, attending film and stage performances; Selecting material for publication, checking style, grammar, accuracy and legality of content and arranging for any necessary revisions; Writing editorials and commentaries on topics of current interest to stimulate public interest and express the views of a publication or broadcasting station; Interviewing politicians and other public figures at press conferences and on other occasions, including individual interviews recorded for radio, television or webcast media; Collecting, reporting and commenting on news and current affairs for publication in newspapers and periodicals, or for broadcasting by radio, television or webcast media; Liaising with production staff in checking final proof copies immediately prior to printing;

television or webcast media; Collecting, reporting and commenting on news and current affairs for publication in newspapers and periodicals, or for broadcasting by radio, television or webcast media; Liaising with production staff in checking final proof copies immediately prior to printing; Writing critical reviews of literary, musical and other artistic works based on knowledge, judgement and experience for newspapers, television, radio and other media; Researching and reporting on developments in specialized fields such as medicine, science and technology; Receiving, analysing and verifying news and other copy for accuracy; Selecting, assembling and preparing publicity material about business or other organizations for issue through press, radio, television and other media. ILO Brief 6 Generative AI and the media and culture industry

 Figure 2. Task-level scores by ISCO 4-digit, sorted by mean at job level Note: Levels of exposure to potential automation by GenAI with capabilities similar to GPT-4 on 0-1 scale. “Medium exposure” for 0.5-0.75 scores and “high exposure” for scores greater than 0.75. See ILO Working Paper 96 for details.

For example, full automation of the journalistic task of “receiving, analysing and verifying news and other copy for accuracy” would require major trust in the consistent reliability of such an automation process. While current AI technology can greatly en hance certain aspects involved in such type of verification, the associated risks, such as potential inaccuracies and biases, would likely discourage most media organisations from fully removing the human element from this process. In addition, GenAI also brings new challenges in content verification, particularly due to the possible proliferation of false information generated through different automated or semi -automated processes (see box 3). As such, it is more accurate to consider that this technical a rea involves a bundle of smaller tasks, which will evolve along with the occupation’s exposure to the AI systems. While some tasks may eventually become automatable, new tasks are likely to emerge as the content verification process evolves and adjusts to new technological advancements and

such, it is more accurate to consider that this technical a rea involves a bundle of smaller tasks, which will evolve along with the occupation’s exposure to the AI systems. While some tasks may eventually become automatable, new tasks are likely to emerge as the content verification process evolves and adjusts to new technological advancements and challenges. Nevertheless, analysing the distribution of theoretical exposure scores at the level of detailed occupations and their internationally agreed task bundles in the ISCO -08 system, can help indicate the general d irection of some of the changes that the media and culture industry will possibly undergo. In addition to the distribution of task-level scores, we can also consider the overall occupational scores, calculated as the mean of individual task scores under each occupation. Figure 3 presents these occupational means of individual mediarelated jobs in the context of all 436 occupations in the ISCO -08 system, plotted against the standard deviation (SD) of task-level scores within each occupation. Jobs with a high mean score and a low standard deviation fall into the category of high automation potential (red triangles in the plot), as the majority of the occupation’s tasks have high exposure scores. Jobs with a high augmentation potential (blue squares in the plot) are at the other extreme, as they have a low occupationlevel mean score, but a high sta ndard deviation of the task scores, meaning that the tasks are diverse enough to ensure the continued need for human involvement. Media and culture jobs, as defined in Table 1, are marked with green dots. ILO Brief 7 Generative AI and the media and culture industry

 Figure 3. Media-related jobs: exposure to AI, relative to other occupations in ISCO-08

Analysing Figures 2 and 3 jointly, we can observe that the level of exposure to GenAI varies greatly across different occupations within the media and culture sector . The lowest exposure concerns such occupations as dancers, choreographers and creative and performing artists, for whom most task-level scores and occupational mean scores are placed far to the left of the 0.5 threshold, partly due to the physical necessi ties of this work. Several other occupations –

occupations within the media and culture sector . The lowest exposure concerns such occupations as dancers, choreographers and creative and performing artists, for whom most task-level scores and occupational mean scores are placed far to the left of the 0.5 threshold, partly due to the physical necessi ties of this work. Several other occupations – such as broadcasting and audiovisual technical or graphic and multimedia designers – are found in the middle range. For these roles, only some tasks show exposure scores above 0.5, suggesting that while human involvement remains central to most tasks they involve, GenAI tools could potentially “augment” or “complement” these jobs by taking over some of the tasks, either partially or entirely, transforming the way these jobs are performed rather than fully replacing them. On the highest end of that spectrum are the occupations of authors and writers, translators, interpreters and linguists, as well as media announcers and journalists. The distribution of task -level scores (Figure 2) and the occupations ’ average scores (Figure 3) suggests that many of the tasks defined for these jobs in the ISCO-08 documentation have a medium or high level of exposure to GenAI. The media and entertainment industry serves as a useful example of why the feasibility of automating tasks must be assessed within a wider context, beyond just theoretical possibilities. ILO Brief 8 Generative AI and the media and culture industry

Potential AI exposure does not imply immediate AI deployment for full automation. In most of these highly exposed occupations, the human role remains crucial to the job. For example, while media messages can indeed be presented by a machine with a human voice, few people would be willing to watch the evening news read by a digital avatar. In theory, advanced chatbots could also be used to ask interview questions; however, such a product is unlikely to match the interest generated by a skilled human journalist, or of interest enough to attract guests to a programme. Similarly, in the category of Authors and Related Authors, creating written media content can indeed be supported by a machine in many ways, with benefits for speed and even creativity. However, attempting to replace all writing with a bot trained on previous content

generated by a skilled human journalist, or of interest enough to attract guests to a programme. Similarly, in the category of Authors and Related Authors, creating written media content can indeed be supported by a machine in many ways, with benefits for speed and even creativity. However, attempting to replace all writing with a bot trained on previous content would be far from a viable strategy in the media industry. Therefore, the main objective of this analysis is to understand the direction of possible changes and provide insights for consultations and debates that are necessary for the design of evidence-based transition policies, measures and regulations, appropriate to national contexts. As we explore the broader implications of AI exposure across various occupations, it is equally important to delve into specific sectors to understand how AI technologies are being integrated and relevant impacts. Accordingly, case studies are presented to examine the transformative effects of AI in different sub-sectors within this field, and to better appreciate the nuanced benefits and challenges brought about by these advancements.  Case studies on emerging practices and effects of GenAI in specific media and culture sectors The transformative effects of AI, particularly Generative AI (GenAI), on the media and culture sector are not uniform; they vary significantly across sub -sectors and even within the same sub -sector. The use of automation for specific tasks may benefit certain actors while disadvantaging others, creating trade-offs that complicate the identification of general trends. In this section, we present several case studies that delve deeper into the potential transformations, examining both the specific benefits an d challenges introduced by AI and GenAI. These case studies aim to illustrate how AI can positively enhance productivity and creativity in certain areas, while simultaneously posing risks related to job displacement, working conditions and skills gaps, among others.

AI in Journalism: Shaping the Future of Reporting AI has a wide range of applications in journalism, some of which offer clear advantages, while others raise concerns. With respect to advantages, some of the key applications are in data analysis and fact -checking (see Box 2). AI can quickly process large datasets, identify trends and verify claims much faster than humans. It can support investigative journalism by identifying significant topics, monitoring social media for events such as protests or accidents and isolating factual

respect to advantages, some of the key applications are in data analysis and fact -checking (see Box 2). AI can quickly process large datasets, identify trends and verify claims much faster than humans. It can support investigative journalism by identifying significant topics, monitoring social media for events such as protests or accidents and isolating factual claims from complex texts. This allows journalists to focus on critical reporting with greater precision.10 Another transformative role of AI lies in content creation. Journalists can leverage AI to generate data -driven drafts, drawing from public databases on topics such as economics, sports, or the weather, providing a foundation that reporters can refine and contextualize, thus reducing the time required to produce content. Furthermore, AI facilitates content personalization by analysing user behaviour and preferences. Platforms can deliver tailored articles, podcasts, or local news, creating a more relevant e xperience for audiences. Additionally, AI can streamline workflows by automating tasks like metadata generation, interview transcription, data visualization and content organization.

10 UNESCO, AI and the future of journalism, 2024 and UNESCO, Reporting on artificial intelligence: a handbook for journalism educators , module 7, 2023. ILO Brief 9 Generative AI and the media and culture industry

 Box 2 - Pioneering AI in Journalism: La Nación's Innovation in News Gathering, Data Analysis, and Gender Bias Reduction The Argentine newspaper, La Nación, has been a pioneer in the use of artificial intelligence (AI) for journalism, exploring its application in news and information gathering for both narrative and business purposes. Their experimentation began with a 2016 project that investigated the progress of solar parks in Argentina, using machine learning and computer vision to analyse satellite images. Eleven thousand images were used to train the algorithm, after which 7 million images were processed, covering 2.8 million square kilometres of land. The data indicated the government's renewable energy goals had not been met. La Nación also established an AI laboratory that brought together journalists, data analysts, and developers to expand their working capacity and accelerate learning. Among their projects was an analysis of trap music lyrics using AI,

government's renewable energy goals had not been met. La Nación also established an AI laboratory that brought together journalists, data analysts, and developers to expand their working capacity and accelerate learning. Among their projects was an analysis of trap music lyrics using AI, which allowed them to identify trends and linguistic features of this globally popular music genre. The methods used could also be applied to other types of music or even different texts, such as political speeches. La Nación has also employed AI to reduce gender bias in journalism by developing a tool that measures the proportion of female and male sources cited in articles, as well as the use of images by gender.

Source: Global Investigative Journalism Network, AI Journalism Lessons from a 150-Year-Old Argentinian Newspaper,

2022 At the same time, the integration of AI into journalism presents several challenges that could undermine the quality and integrity of the profession. One major concern is the potential decline in content quality. AI struggles to fully grasp nuance, cultural context, and linguis tic subtleties, which are essential for in -depth and insightful reporting. This limitation risks generating superficial coverage, especially on complex topics requiring critical analysis. Bias is another concern. AI systems learn from large datasets that can reflect societal, political, or cultural biases.11 These biases can unintentionally be amplified by AI, leading to unbalanced reporting and the perpetuation of stereotypes, threatening the credibility of journalism. Additionally, AI -generated content lacks the emotional depth and personal perspective that human journalists bring and may lead to impersonal narratives that may fail to connect with readers on a meaningful level. AI also raises concerns about plagiarism, as its reliance on extensive datasets increases the likelihood of producing text resembling existing work, posing ethical and legal risks (see Box 3). Lastly, an over -reliance on AI could erode journalists’ critical thinking, investigative skills, and creativity. This dependency risks turning journalism into a mechanized process, diminishing its human-driven essence and weakening the profession's role as a cornerstone of truth and accountability. Furthermore, a lack of recognition and fair remuneration can reduce incentives for creating original, high-quality content,

critical thinking, investigative skills, and creativity. This dependency risks turning journalism into a mechanized process, diminishing its human-driven essence and weakening the profession's role as a cornerstone of truth and accountability. Furthermore, a lack of recognition and fair remuneration can reduce incentives for creating original, high-quality content, affecting both journalism and artistic work. This decline in original content could degrade the quality of outputs from AI models, further destabilizin g the information ecosystem and exacerbating challenges faced by news outlets due to the influence of social media and search engines. Other challenges include potential harms to language diversity, cultural expressions, as well as concerns over the economic viability of smaller outlets, particularly in the Global South, as a result of unequal access to AI infrastructure and training.

11 Paweł Gmyrek, Christoph Lutz, and Gemma