Generative AI and Investigative Journalism—An Application of Baumol's Cost Disease Across Mass Communication Sectors

Abstract

Generative artificial intelligence (AI) has transformed various industries, automating tasks and enhancing productivity. Yet, its impact varies across sectors. In mass communication, AI has notably benefited entertainment, public relations, and advertising. However, it poses a lesser advantage for investigative journalism, where labor-intensive research and other highly specialized activities depend on highly educated journalists. This situation parallels Baumol's concept of a cost disease, which applies to labor-intensive “stagnant” service sectors such as healthcare, education, and the performing arts. These sectors rely heavily on human labor and struggle to achieve significant productivity gains from technological advances. This means they may require decreasing wages or increasing subsidies to compete with more productive industries. In mass communication, this phenomenon could be especially pertinent to (investigative) journalism, which directly competes for attention with other communication forms. This could exacerbate pressures on wages and working conditions, hindering the attraction of talent to the field and necessitating higher public spending on this vital journalism form. This commentary presents a cross-industry analysis of the impact of generative AI on communication sectors and derives directions for further research that would enhance our understanding of journalism's economic viability in a digital communication landscape and inform policy makers to address the affiliated societal challenges. It closes with a discussion of Guzman and Lewis’s call for more cross-industry research on AI in communication industries.

Introduction

Generative artificial intelligence (AI) improves productivity in many sectors of the economy, including the communication industries. All communication sectors—journalism, public relations, advertising, and entertainment (Hanitzsch, 2007)—use AI in similar ways (Guzman & Lewis, 2024, p. 4), but their impact on productivity might differ across sectors. This implies that some sectors develop a relative cost disadvantage and consequently suffer from a “cost disease” (Baumol, 1967, 2012; Baumol & Bowen, 1965), associated with increasing economic pressure to reduce the quality and quantity of their offers.

Journalism holds a special role in the world of mass communication due to its particular societal importance in democracies. It fulfills a variety of relevant functions, including the provision of information, participation in the formation of public opinion, the exercise of control and criticism, and the promotion of education (Blumler & Gurevitch, 1990; McQuail, 1992). Investigative journalism plays a particularly crucial role in the area of control and criticism and is often referred to as democracy's watchdog (Houston, 2010).

The discussion about generative AI's impact on journalism has so far mostly revolved around four lines of argument: first, generative AI has been heralded for its potential productivity increases that allows journalists to carry out their work more efficiently (Guzman & Lewis, 2024, p. 4). This could counter some of the economic struggles digitization has imposed upon journalism, in particular the breakdown of the advertising model, and thus enabling journalism to better serve democratic societies. On the other hand, generative AI has spurred the fear of job loss, as an increasing degree of automation in the field could reduce the need for human labor. At the same time, many experts have stated that the core journalistic functions will not be replaced by machines, as they involve nonrepetitive and highly specific tasks. Lastly, ethical concerns about the implementation of generative AI in journalism are discussed, especially as “communicative AI” (Guzman & Lewis, 2020) increasingly takes over the role of the communicator (Hepp et al., 2023), which potentially compromises journalistic standards of correctness and truthfulness.

In view of these developments, the question arises as to whether the use of generative AI can lead to improvements in investigative journalism in terms of efficiency and journalistic quality. The scope of all of the mentioned valuable discussions mainly remains within the journalism sector. This commentary lays out an analysis across communication industries. I argue that generative AI will have higher productivity effects on public relations (PR), advertising, and entertainment than it has on journalism, and particularly investigative journalism. The main reasons for this are that investigative journalism (1) involves a higher share of nonrepetitive and highly specific tasks, (2) creates genuinely new information, and (3) relies more heavily on truthful and correct information. This makes investigative journalism more heavily dependent on highly skilled human labor (Hamilton, 2016; Houston, 2010) than other fields of communication.

Journalism therefore represents a more stagnant sector in terms of productivity increases induced by generative AI and falls into a relative cost disadvantage against other sectors. This assessment of a “cost disease” implies increasing economic pressures for stagnant sectors, ultimately further threatening the economic viability of (investigatve) journalism.

After a brief theoretical explanation of the concept of Baumol's cost disease, I lay out the argument in more detail and derive implications for media practitioners and media policy of how to deal with this assessment.

By analyzing the productivity effects of generative AI across communication sectors, this paper follows Guzman and Lewis’s (2024) call for “a more cross-industry approach to scholarship (to) develop a more encompassing picture about AI's impact on media work and media consumption” (p. 1). The commentary closes with a short discussion of the potential of cross-industry analyses.

Baumol's cost disease

Baumol's cost disease is a concept that examines the uneven distribution of productivity increases in different sectors of the economy and their impact on the cost structure and competitiveness of industries (Baumol, 1967, 2012; Baumol & Bowen, 1965). This phenomenon has far-reaching implications for the economy and in particular for sectors that are heavily dependent on human labor.

Baumol's cost disease is based on a number of assumptions and mechanisms that explain the development and change of economic sectors over time. A central assumption of Baumol's cost disease is that sectors compete for labor by offering competitive wages and working conditions. This competition for skilled labor leads to an adjustment of wages across sectors. Another assumption is that wages usually follow productivity increases. When productivity increases in one industry, workers become more efficient, which forms the basis for wage increases.

Baumol's cost disease is triggered by the uneven distribution of productivity increases between economic sectors. Some sectors, such as industrial goods or information technology, experience rapid and significant increases in productivity. In other sectors, such as education, healthcare, or the performing arts, such increases are limited.

Despite the different productivity trends in the sectors, wages tend to be adjusted evenly across all industries. Employees in sectors with low productivity increase demand and receive similar wage increases as employees in more productive sectors or they leave the sector.

As wages rise in all sectors, this leads to relative cost increases in sectors with lower productivity growth. The increased wage costs account for a larger proportion of total costs over time, without productivity increasing to the same extent.

Sectors affected by Baumol's cost disease have limited options to react. Increasing costs, for example by raising wages, is one possibility, which leads to higher prices for their services or products. This in turn can affect competitiveness.

Another reaction to Baumol's cost disease is to reduce the supply in the affected sectors. This can mean a reduction in the quality or quantity of services or products offered. This can lead to consumers being less satisfied and the sectors losing competitiveness.

In many cases, affected sectors try to compensate for increased wage costs by worsening the working conditions of employees. This can lead to dissatisfaction among the workforce and reduce the attractiveness of these industries as a place to work.

Baumol's cost disease has been well documented empirically in various studies and manifests itself particularly in sectors characterized by nonroutine human interactions or activities, as they generally experience lower productivity growth. Some of the industries that are typically affected by this phenomenon are healthcare, education, and culture/performing arts. These sectors are highly dependent on skilled labor and human expertise, which makes it difficult to increase productivity to the same extent as in other sectors.

Most of the sectors severely affected by Baumol's cost disease fall into the area of “public goods” or “public services.” This raises the question of whether journalism, especially investigative journalism, could also be affected by similar problems. Investigative journalism has striking similarities with the industries mentioned above. It requires highly skilled journalists carrying out very specific and nonroutine tasks. Productivity gains in the journalism industry have historically been limited, as the quality and originality of reporting are largely dependent on human involvement.

Journalism's diagnosis

Productivity gains achieved through the use of AI are of crucial importance in mass communication. AI technologies enable automation, personalization, and more efficient processes in various aspects of mass communication. This includes automated data analysis, content creation, content delivery, and curation as well as service provision. However, these productivity gains are unlikely to be equally high in all areas of mass communication.

To understand the impact of Baumol's cost disease on mass communication, it is useful to distinguish between four basic types of mass communication—advertising, public relations, entertainment, and journalism (Christmas et al., 2020; Hanitzsch, 2007; Splichal & Dahlgren, 2016).

Advertising includes the creation and distribution of advertising messages and marketing campaigns. Here, productivity gains can be significant through data-driven targeting algorithms and automation of content creation.

PR involves maintaining relationships between organizations and the public. Here too, AI tools can help to streamline the process, for example through media monitoring and social media analysis, but generative AI can also automate content creation to a significant extent.

Entertainment includes media content such as films, music, and entertainment programs. Here, too, productivity increases can be realized through AI in production and distribution.

Journalism also uses generative AI in very similar ways as the above-mentioned fields, for example in reporting on financial data and sports results. However, three aspects limit the applicability of generative AI in journalism: the nonrepetitiveness of core tasks, the so-called “truth constraint,” and the inability of AI models to produce genuinely new information.

The greater the required creativity, complexity, and specificity of a task, the more highly educated and specialized labor is needed to complete the task, and the less likely this task can be taken over by machines.

Many tasks involved in investigative journalism are highly specific, that is, different every time, making it harder to have machines reliably carry out the task. It involves independent research and detailed reporting on issues of public importance, often against the wishes of the actors involved to keep this information secret (Weinberg, 1996). This implies highly specific investigations, lengthy research processes, and high uncertainty of results up to the danger of costly legal disputes.

Additionally, the produced information and content must adhere to high standards of correctness and truthfulness, a defining criterion for journalism against other forms of mass communication (Hanitzsch, 2007).

In addition, generative AI models are designed to learn the underlying patterns and structures of data and generate new data points that could plausibly be part of the original dataset (Pinaya et al., 2023). This opens up opportunities for simpler forms of journalism by offering the possibility of making the research and reporting process more efficient and cost-effective (Hamilton, 2016). Investigative journalism, however, brings issues and information to light that were previously unknown to the public. This means that new information is produced that generative AI applications, which are usually trained with existing data and therefore with known data from the past, can neither produce nor verify.

So, while advertising, PR, and entertainment can benefit greatly from the efficiency gains of AI, investigative journalism is less suited to AI-initiated productivity gains due to its reliance on nonroutine human research and analysis.

The biggest potential of AI applications in investigative journalism is usually ascribed to data analysis (Broussard, 2015; Rinehart & Kung, 2022; Stray, 2019; Wilczek et al., 2024). While great efficiency increases can be expected in detecting newsworthy patterns in readily available large data sets, investigative reporting typically revolves around “substantive issues that someone wants to keep secret” (Hamilton, 2016, p. 10). Consequently, significant human effort is usually necessary to request, negotiate, scrape, or purchase such datasets (Stray, 2019), limiting its impact on productivity in investigative journalism as opposed to other forms of journalism and communication.

Therefore, the challenge is how investigative journalism can be financially sustained when wage levels in other fields of mass communication are rising, but investigative journalism has lower productivity gains and thus struggles to fund similar wages.

This mechanism is exacerbated by the fact that mass communication today is characterized by intense competition. This is the result of various factors, including the convergence of communication channels and end devices. Traditionally separate forms of mass communication such as print media, broadcasting, online platforms, and social media are now competing directly with each other on the smartphone for audience attention.

Another key factor in mass communication is the relative ease with which workers can switch between different forms of communication. Job profiles in journalism, advertising, and PR have significant overlaps in the necessary skill sets. Employees in these sectors can therefore transfer between the different forms of communication quite easily—a phenomenon that has been observed in the recent past to the detriment of journalism and in favor of advertising and PR. This flexibility and mobility of the workforce between the different forms of mass communication further increases competition and the dynamics of job changes.

Treatment options

The mechanism summarized once again: AI applications, especially generative AI, will make investigative journalism more productive. However, contrary to the intuitive conclusion that this must be good news for the economic viability of journalism, the opposite could be the case. Namely, if productivity increases are greater in other industries, and especially in other mass communication industries, journalism will then have a relative cost disadvantage despite an absolute increase in productivity, as wages and working conditions will improve more or less simultaneously in all sectors. However, advertising, PR, and entertainment will be better able to compensate for this with higher productivity increases than in journalism.

The consequences are troublesome: if investigative journalism wants to remain competitive, three potential strategies are available. Firstly, costs can be cut and working conditions and wages reduced. This is a strategy that tends to be particularly popular in industries with a highly intrinsically motivated workforce and has already been observed in journalism in recent decades. A second option is to reduce the quality and quantity on offer. These two strategies usually go hand in hand—those who reduce costs often have to accept a reduction in quality, at least in the medium term. Finally, prices can also be increased to compensate for the relatively higher costs compared to other sectors.

From a societal perspective, all three strategies imply undesirable consequences, as they all ultimately lead to less and/or lower quality journalistic activity and/or less reach. This impedes the functions of the mass media in democratic societies, as there would be less control and criticism and society would be less well informed. Consequences are harsh, as we see in the so-called news desserts in the United States, the United Kingdom, Australia, and other countries around the globe, where the political systems and civic societies take measurable downturns in performance, such as decreases in political participation and accountability (Matherly & Greenwood, 2021; Rubado & Jennings, 2020).

However, experiences from other industries can inform societal reactions to this phenomenon. After all, investigative journalism is not the first industry to be affected by Baumol's cost disease. Health, education, and cultural sectors have received an increasing proportion of gross domestic products in order to keep these sectors viable. This is not as problematic as it might seem at first glance. “Stagnant-sector services will never become unaffordable to society. This is because the economy's constantly growing productivity simultaneously increases the community's overall purchasing power” (Baumol, 2012). As the increase in overall economic value creation is always greater than the relative cost disadvantage of the stagnant sectors, investigative journalism could be subject to increased financial support.

In concrete terms, this could mean subsidizing investigative journalism to maintain its quantifiable benefits for the economy and society. Since one of journalism's main functions is to control the powerful, support needs to be independent from possible state influences. There are plenty of suggestions and empirical evidence as to how this could be done. Tax cuts for the news industry (in particular for nonprofit organizations), subsidizing journalistic jobs (like in Luxembourg and California), and price subsidies by issuing vouchers to consumers to spend on journalistic products or to donate to journalistic institutions, are but a few suggestions. All of these instruments would directly benefit journalistic activity, but without the threat of the government interfering with the content.

Avenues for future research

The issue of Baumol's cost disease in investigative journalism calls for a specific research agenda. First, future research could investigate the specific ways in which generative AI technologies are being implemented in investigative journalism processes. First attempts, describing the implementation of AI in journalism can already be observed (Beckett & Yaseen, 2023; Diakopoulos et al., 2024; Schützeneder et al., 2024; Simon, 2024; Wilczek et al., 2024), but so far do not distinguish between different types of journalism. Specifically, the research could employ case studies of news organizations or investigative journalism teams that have adopted AI tools and analyze the effects on productivity, cost structure, employment, and quality of reporting.

Research should also delve into empirical studies across communication industries—as urgently suggested by Guzman and Lewis (2024)—to assess the extent of productivity growth in (investigative) journalism compared to other mass communication sectors to quantify relative cost differences.

A third avenue would be the investigation of policy interventions aimed at supporting investigative journalism while preserving journalistic independence, such as the ones mentioned in this article. This could include analyzing existing policies in different countries, such as tax incentives for news organizations or public funding mechanisms, and assessing their effectiveness in sustaining quality investigative reporting. It could also include empirical studies on the expected effects of not yet excisting (hypothetical) support mechanisms, similar to approaches that measure the effects of cross-publisher platforms on market revenue and subscription numbers (Erbrich et al., 2024).

By exploring these avenues for research, academic scholarship can contribute to a deeper understanding of the challenges and opportunities stemming from Baumol's cost disease in investigative journalism and develop strategies to ensure its continued relevance and impact in democratic societies.

The potential of cross-communication industry research on generative AI

Guzman and Lewis (2024) convincingly state that “The allure of AI is the ability for practitioners to carry out the work of journalism, advertising, or public relations more efficiently” (p. 4) and “media professionals across these fields are adopting similar AI technologies (e.g., machine learning, natural language processing, and recommender systems) for often similar purposes” (p. 1). The authors use examples from branding and ethics to illustrate the usefulness of cross-industry analyses to produce findings that hold true across sectors.

The cross-industry analysis presented in this commentary exploits the idea that the extent of those efficiency increases likely differ significantly across communications industries. The analysis produces industry-specific findings that would not have been possible to derive from a within-industry analysis. In this respect, this commentary illustrates that the potential of cross-industry analyses is not limited to findings that are generalizable across industries. It rather extends to the identification of industry-specific findings and policy implications.

“The practical, strategic, and ethical challenges posed by AI (…) are broadly shared across industries, their workers, and their audiences.” (Guzman & Lewis, 2024, p. 6). Yet they likely differ in effect strength and cross-industry analyses can help draw a clearer picture of what to treat equally and what unequally.