Changing or Reinforcing the “Rules of the Game”: A Field Theory Perspective on the Impacts of Automated Journalism on Media Practitioners

ABSTRACT

In recent years, a growing number of media outlets such as The Associated Press, The Washington Post and the BBC have adopted “automated journalism,” a solution understood as the generation of journalistic text with software and algorithms, with no human intervention except for the initial programming. However, to discern common patterns across news organisations, a comprehensive empirical investigation looking into the impacts of automated news on media practitioners is needed. To lay the ground for such an empirical work, this paper raises the necessary questions to be taken into consideration, while reflecting on the potential implications for journalism practice. To do so, we relied on key concepts developed in Bourdieu’s Field theory to assess whether automated journalism is likely to change or reinforce the prevailing journalistic doxa, or the “rules of the game” within the journalistic field. This led us to conceive an analytical framework that builds on these concepts, first to come up with our own reflections on the modifications and reinforcements of the journalistic doxa, then to contribute a table of essential dimensions to consider for an empirical investigation that looks into the impacts of automated journalism on media practitioners.

KEYWORDS:

• Algorithmic journalism
• automated journalism
• automated news
• computational journalism
• field theory
• robot journalism

Introduction

In recent years, a software process known as Natural Language Generation (NLG) that usually involves pre-written templates feeding on structured data, but also advanced machine learning techniques that “learn patterns of language use from large corpora of examples” (Diakopoulos 2019, 101), has increasingly been employed to produce “automated journalism,” a solution generally understood as the computer generation of news articles or text, without any human involvement except for the initial programming (Graefe 2016; Lindén 2017). Automated journalism started gaining traction after The Los Angeles Times automated its Homicide Report blog in 2010 and launched Quakebot, a computer software monitoring seismic activity in California, in 2014. At the same time, The Associated Press partnered with the start-up Automated Insights to automate corporate earnings stories and The Washington Post developed its own programme to report on the results of the Rio Olympics. More recently, The BBC resorted to automated journalism to cover the 2019 general elections in the United Kingdom, both in English and Welsh.

In a systematic literature review of automated journalism scholarship (Danzon-Chambaud 2021), we demonstrated that research in this area can be split into two categories. First, studies on the reach of automated journalism showing that readers perceive the credibility, objectivity and trustworthiness of automated news as being equivalent to human-written news, and that the adoption of automated news could influence a wide variety of domains such as legal proceedings and securities exchange. Second, studies on automated journalism in practice, which stressed the importance of taking a better look at individual and collective impacts the technology has on media practitioners and other associated actors, in order to identify common patterns across media organisations. Such a research agenda could take advantage of Bourdieu’s Field theory, which Anderson (2013) saw as an adequate framework for the study of computational and algorithmic journalism.

Similarly to New Institutionalism, Field theory contributes a “mezzo-level” scope of analysis able to account for the dimensions located “between the individual news organization and the society as a whole” (Benson 2006, 199), while also taking into consideration the internal tensions occurring within this microcosm or “field” (Bourdieu 2005). Although Bourdieu’s model was initially developed to analyse artistic behaviours (Bourdieu 1992) and was later used to examine the influence of televised news (Bourdieu 1996), field theory could prove useful to study the impact of digital media on journalism (Benson and Neveu 2005). The internal tensions described by Bourdieu are even more important to take into account nowadays as digital news are sometimes seen as a threat to the very core of news media’s business models, and developments associated with the increased competition in the digital news sector and the entry of new actors such as bloggers and social media platforms are clearly challenging traditional news organisations (see e.g., Nielsen 2012, 2016). Furthermore, digital developments are also criticised for bringing on contentious issues such as precarious forms of work in content farms and in blog writing (Bakker 2012), the outsourcing of domestic coverage abroad through live streaming means (Örnebring and Ferrer Conill 2016) and the increased use of metrics within news organisations, which gradually led discourses around “innovation” to win over discourses around “quality journalism,” as a switch to audience engagement occurred over time (Costera Meijer 2020). In parallel to these challenges, new questions that deal with personal data exploitation—a multifaceted phenomenon that involves industry, government and academia (van Dijck 2014) and could relate in a sense to historic forms of colonialism (Couldry and Mejias 2019)—should be addressed in light of big data’s growing role in journalism (Lewis and Westlund 2015).

Field theory can therefore provide a timely “vector of power dynamics” to the study of technological innovation, which, according to Anderson (2013, 1013) is “too often understood from within an ‘all boats will rise’ mentality.” In this paper, we will use Bourdieu’s lenses to raise the necessary questions for an empirical investigation of the impacts of automated journalism on media practitioners, which will distinguish common patterns across news organisations. To do so, we will constitute an analytical framework that builds on Field theory concepts and use it to reflect on the potential implications for journalism practice.

Field Theory Concepts

In media studies, the concepts at the core of Bourdieu’s Field theory have been employed to analyse, among others, the journalistic “gut feeling” (Schultz 2007), the internal dynamics within online forms of journalism and the changes it triggers within journalism as a whole (Siapera and Spyridou 2012), how online journalists perceive themselves (Møller Hartley 2013), bloggers’ perceptions of mainstream media (Vos, Craft, and Ashley 2012), the position of sports journalism within the journalistic field (English 2016), media discourses on entrepreneurial journalism (Vos and Singer 2016), the entrance of Buzzfeed and venture-backed news startups into journalism (Tandoc and Jenkins 2017; Usher 2017), data journalism practices in the Arab world (Fahmy and Attia 2020), and, lastly, the growing algorithmic automation of newsrooms (Wu, Tandoc, and Salmon 2019a).

In Field theory, Bourdieu envisions the social world as a myriad of “fields” that are exposed to an evergoing struggle between two major forces: on the one hand, a form of power that arises from economic capital, in other words “money or assets that can be turned into money,” and on the other hand, a form of strength that is made of cultural capital, which we can understand as a set of unique abilities that include “educational credentials, technical expertise, general knowledge, verbal abilities, and artistic sensibilities” (Benson and Neveu 2005, 4). Within each field, the ongoing competition between economic and cultural capitals translates into two opposing poles, which can be found under a different wording depending on the field (Bourdieu 2013), but are known as the heteronomous and autonomous poles in the case of journalism.

According to Bourdieu (2005), the heteronomous pole reflects a type of journalism exposed to external influences, mostly political and economic, as illustrated for instance by the influence advertisers exert on commercial television news. On the opposite end of the spectrum, the autonomous pole can be seen as a manifestation of an independent form of journalism that Bourdieu perceives as being the “purest,” on the grounds that it would be exempted from external pressures. Print journalists that set up the news agenda for the day as well as journalists being awarded the Pulitzer Prize could in a sense be representative of this autonomous pole (Bourdieu 2005; Benson and Neveu 2005).

Lastly, in addition to the confrontation between these two poles, Field theory also introduces concepts that relate to an individual’s behaviour within the field: hence, the “doxa” reflects the “universe of tacit presuppositions that we accept as the natives of a certain society” (Bourdieu 2005, 37), the “habitus” assumes that “individuals’ predispositions, assumptions, judgments, and behaviors are the result of a long-term process of socialization,” while the “illusio” represents “an agent’s emotional and cognitive ‘investment’ in the stakes involved” (Benson and Neveu 2005, 3). In the journalistic field, the doxa could then be conceptualised as the “rules of the game” (Tandoc and Jenkins 2017), the habitus as a “feel for the daily news game” (Schultz 2007) and the illusio as a belief that this game is “worth playing” (Benson and Neveu 2005). Additionally, Bourdieu also evokes a situation of “hysteresis” when individuals “judge and act today according to dispositions previously acquired under quite different social conditions” (Benson and Neveu 2005, 10), resulting in their habitus being out of touch with a new order (Bourdieu 2000; in Wu, Tandoc, and Salmon 2019a).

In media studies, these concepts at the core of Field theory have mostly been operationalised in two different ways. First, a research stream looking at the forms of capital at stake to document the characteristics of a journalistic subfield or to locate it within the larger field of media production. For instance, Siapera and Spyridou (2012) concentrate on economic, cultural, social and symbolic capital to study the specifics of the online journalism subfield while English (2016) draws on the opposition between economic and journalistic capital to situate the subfield of sports journalism within the journalistic field.¹ Second, a research stream analysing the influence of new entrants (e.g., bloggers, BuzzFeed, entrepreneurial journalists, news startups, technological companies) on either changing or preserving the doxa that prevails in the journalistic field.

This second stream of research is based on Bourdieu’s belief (2005, 39) that “to exist in a field … is to differentiate oneself.” According to this view, new entrants in the field or a subfield of journalism could disrupt or abide by the prevailing norms or doxa, thus reinforcing or changing the nature of both economic and cultural capitals. Vos, Craft, and Ashley (2012) consequently concluded that bloggers seemed to have accepted the prevailing journalistic doxa while Tandoc and Jenkins (2017) illustrated that BuzzFeed was perceived as willing to go along with it. In the same line of thought, Vos and Singer (2016) showed that entrepreneurial journalists tended to be well accepted among their peers despite the potential they hold to change the prevailing doxa while Usher (2017) demonstrated that news startups do not fundamentally change the doxa of the field, but rather challenge the internal hierarchies within it.

Our research follows this second perspective as we believe that the impacts of automated journalism on media practitioners could be better understood by looking at how they relate to the prevailing journalistic doxa, either through changes or preservation. Our research questions therefore are:

1. Is the prevailing journalistic doxa likely to be changed or reinforced by the introduction of automated journalism within newsrooms?

2. Based on these reflections, what are the essential dimensions to consider for an empirical investigation of the impacts of automated journalism on media practitioners?

Methodology

To address RQ1, we needed to build an analytical framework that would enable us to reflect on the changes and reinforcements automated journalism could bring to the prevailing journalistic doxa. First, to materialise this journalistic doxa, we relied on Deuze’s interpretation of journalism ideology, which can be understood as “how journalists give meaning to their newswork” (2005, 444). Often considered as a normative model in media studies (Lindén 2017; Usher 2017), Deuze’s concepts introduce five ideal-typical values that he believes are representative of journalism’s ideology: public interest, characterised by a “watchdog” style of reporting that makes elites accountable; objectivity, which speaks to a sense of impartiality, fairness and professional detachment; autonomy, a notion that has to do with journalists’ freedom to tell the stories they want without external forms of pressures, constraints or influences; immediacy, which refers to the speed of breaking news but also to a non-stop news cycle, and ethics, that can be understood as individual behaviours regulated through professional standards.

Second, to take into consideration the modification or reinforcement of the prevailing journalists doxa as a result of the implementation of automated journalism, we followed Wu’s, Tandoc’s and Salmon’s interpretation of structure and agency, which stipulates (2019a, 428) that “social structures shape the logics of the journalistic field and the behavior of agents to adopt automation” and that “the agency of actors, in turn, will reshape the structures over time through the skillsets they accumulate and their attitudes towards field transformation or preservation.” Wu, Tandoc and Salmon suggest to look for these tensions in three areas: first, outside the journalistic field, through the influence external structures such as political, economic, social and technological forces exert on the field; second, inside the journalistic field, through the kind of cultural capital that journalists need to acquire in the automation age; third, through the reactions of journalists to increased automation that contribute to creating tensions within the field.

Deuze’s materialisation of the journalistic doxa as well as Wu’s, Tandoc’s and Salmon’s perspective on structure and agency in the era of newsroom automation will constitute the two pillars guiding our analysis. To answer RQ1, we will come up with our own reflections on the possible combinations between the ideals outlined by Deuze and the areas of struggle highlighted by Wu, Tandoc and Salmon, following a structure that is inspired by the latter (i.e., Structures External to the Journalistic Field, Accumulating Cultural Capital, Adversarial Reactions within the Journalistic Field). In doing so, we will rely on Bourdieu’s concepts of autonomous and heteronomous poles, illusio and hysteresis, but also on other scholars’ lecture of journalistic capital, which constitutes the specific cultural capital of the journalistic field (Schultz 2007) and encompasses many specialisations that include “knowledge and expertise, sources, technical skills, research, and content itself” (Møller Hartley 2013, 573–574), and news habitus, “a bodily knowledge and feel for the daily news game which can be seen in the journalistic practices surrounding qualification and legitimisation of newsworthiness which almost takes place without words” (Schultz 2007).²

Finally, to answer RQ2, our reflections will then be summed up into a list of key considerations that we developed in our Table 1, which can be used as part of an empirical investigation that looks into the impacts of automated journalism on media practitioners.

Table 1. Empirical investigation of the impacts of automated journalism on media practitioners: essential dimensions to consider.

	Public service	Objectivity	Immediacy	Autonomy	Ethics
Structures External to the Journalistic Field	Are media practitioners able to employ their traditional journalistic skills to collect public interest data? Are they able to explore other ways to get public information data?	Is a diverse set of data sources used in the conception of automated news? Is an assemblage of templates being used or just a single one? Are alternative data sources taken into consideration?	What is the extent of human verification in the production of automated news? Are computational tasks employed to verify content when automated news are published on a large scale?	Are organisations with a large open data catalogue overrepresented in automated news output? In advanced machine learning models, is it also the case for organisations that generated a lot of training data? Are less affluent groups or movements not covered on these grounds?	Are media practitioners well aware of algorithmic biases? Are they trained in ethics of artificial intelligence and data management?
Accumulating Cultural Capital	Are journalists with the computing background to programme automated news also able to engage with algorithmic accountability reporting?	When authoring templates, do media practitioners refrain from picking a “winning” side and a “losing” side beforehand?	How do media practitioners handle predicting elements of the story in advance when authoring templates? How familiar are they with the concept of abstraction at the core of computational thinking?	Are journalists with a computing background able to programme their own automated news? Are media practitioners with little or none programming background able to use third-party NLG tools or “No-code” development platforms?	Are media practitioners able to adapt their expert knowledge of core ethical values in journalism to new ethical challenges that arise from the utilisation of automated news? Do media practitioners supervise the incorporation of journalistic standards and practices into automated news scripts? Do they also make sure that those are well maintained and up-to-date?
Adversarial Reactions within the Journalistic Field	Does the deployment of automated news threaten the sustainability of less profitable areas of journalism? Does it impact the livelihood of journalists specialised in these areas?	When complementing an automated draft, do media practitioners remain in control of the story while exerting their critical thinking? In advanced machine learning models, do media practitioners prevent algorithmic biases from creeping into the final copy?	Are media practitioners able to engage with formats that demand more time as automated news take on routine tasks? Are they redirected to stories that reflect the priorities of media owners or of the marketing and advertising departments?	How do media practitioners views on automated journalism collide with those of technologists and business people? Is the displacement of human journalists likely to create tensions?	Does the discussion around revamped ethical values happen within professional journalistic forums, or is it too closely tied to external organisations?

Reflections

Structures External to the Journalistic Field

In automated journalism, one of the areas in which the influence of external structures could impede on journalistic objectivity is related to the over-reliance on a unique source of data to fill a single template. For instance, although automated stories on sports or election results could feed directly on sports leagues’ and governments’ open data portals or APIs, it is nevertheless important to balance those with complementary data sources, so the overall story remains impartial. This could be achieved through an assemblage of pre-written templates (Caswell and Dörr 2018) that could feed on alternative databases, such as those of supporters’ clubs or government watchdog groups. To do so, journalists’ news habitus in selecting balanced viewpoints and credible sources would be deemed an essential qualification.

The over-reliance on a single type of data could also hamper the autonomy of the journalistic field while making it more porous, or heteronomous, to external influences such as political or economic forces. Indeed, corporations and institutions with the means to maintain a large data catalogue as well as those previously covered that generated a lot of training data for advanced machine models are rather likely to be selected for the creation of automated news, as opposed to less affluent grassroot movements and citizen groups. These influential organisations could end up being overrepresented in the news agenda as they would exert a form of control over journalists’ capacity to “tell the stories they want.”

To avoid being too much dependent on external datasets, media practitioners could garner sensitive data materials while using their own reporting skills. This could be envisioned for instance through “structured journalism” (Caswell and Dörr 2018), which stands for the idea of reporting the news in a data format, or in other words turning “narratives into databases” (Anderson 2018, 13). To uphold the value of public interest, media practitioners could adopt a structured journalism approach to gather sensitive data: on the one hand, they could make use of their usual journalistic capital such as filling FOIP requests to access valuable datasets, and on the other hand, they could also engage with new types of journalistic capital such as organising crowdsourcing campaigns to collect a large amount of public interest data (e.g., The Guardian’s crowdsourced investigation on MPs expenses in 2009).

Another issue that relates to over-relying on external data feeds has to do with the possibility that algorithmic errors could make their way into the final copy. This is especially relevant in the case of immediacy, as demonstrated by the automated news story in which The Los Angeles Times, in 2017, warned against an earthquake that occurred in 1925.³ If human verification is necessary to avoid such mistakes, new forms of journalistic capital could also be needed when automated news are published within minutes on a massive scale (e.g., Le Monde’s 36,000 stories to cover the results of a regional election in France in 2015, TaMedia’s 40,000 news articles to report on the outcome of a Swiss referendum in 2018). This could translate for instance into computing skills to programme a wide range of computational tasks such as advanced statistical calculations and text recognition mechanisms (e.g., OCR), so as to be able to pinpoint discrepancies in the final news output.

In a similar way to catching algorithmic errors, media professionals also need to be aware of algorithmic biases that could creep into the final automated story. This is especially true in the case of advanced machine models, for instance when computer software suggests pre-made sentences for the journalist to include in a copy (Lindén 2017). However, this could constitute an excellent opportunity to revamp the ideal of ethics, especially by equipping media practitioners with a news habitus that would be composed of a good understanding of ethics of artificial intelligence and data management. This could be provided either through the organisation’s own standards and practices (e.g., the BBC's principles on responsible machine learning) or through an authoritative source in journalism (e.g., a potential addendum to the Society of Professional Journalists’ Code of Ethics).

Accumulating Cultural Capital

The adoption of automated news within newsrooms prompts media practitioners to re-examine their journalistic capital, whether through re-emphasising their human potential such as adding more context to the story and focusing on in-depth forms of reporting (van Dalen 2012) or through engaging with the concept of “computational thinking” (Diakopoulos 2011; Stavelin 2013; Gynnild 2014), a means of solving problems through ways of abstract reasoning that are at the core of computing skills (Wing 2008). The application of computational thinking to the area of automated news challenges the very notion of immediacy as it demands a whole reconsideration of the news habitus: authoring templates for automated journalism indeed requires predicting elements of the story in advance (Thurman, Dörr, and Kunert 2017), a craft that is difficult to acquire as it necessitates to be familiar with abstraction, what could further entrench an effect of hysteresis among media practitioners.

At the same time, a news habitus that includes predicting elements of the story in advance could impede on the ideal-typical value of objectivity. For instance, media practitioners could feel cornered into choosing a “winning” and a “losing” side beforehand, which could reinforce the perception among news readers that the journalistic field is influenced by heteronomous forces such as political or economic pressures or caught into its own set of autonomous values that may sometimes collide with the “best practices of democratic government” (Schudson 2005, 222).

The adoption of computational thinking could be pushed even further through the integration of programming skills to the journalistic capital, which would be likely to reinforce the autonomy of tech-savvy journalists as they could programme their own automated news systems, as opposed to outsourcing their conception to an external NLG provider, which further limits journalists’ capabilities to influence the algorithmic models adopted by their newsrooms. Moreover, the hysteresis effect that one could expect between “hacker journalists” (Usher 2019) and media practitioners with little or none programming background could meanwhile be mitigated through the use of third-party NLG tools (e.g., Arria Studio, AX Semantics) or platforms using a “No-code” programming language.

Additionally, media practitioners equipped with an adequate understanding of computational thinking could adapt an existing news habitus, their knowledge of journalism ethics, to address new ethical aspects that arise from the deployment of automated news, for instance when “defamatory content slips through the cracks” (Lewis, Sanders, and Carmody 2019, 15). At the same time, they could make sure that current ethical standards are well embedded into the computer scripts that trigger automated news, and verify that they are adequately maintained and up-to-date. As previously stated, a news habitus that would be made of a good understanding of ethics of artificial intelligence and data management will help them in this task.

Finally, the integration of programming skills to the journalistic capital could open new avenues in the domain of public interest journalism, notably with regard to “algorithmic accountability reporting,” a type of journalistic investigation that looks into black box algorithms to reveal “the power structures, biases, and influences that computational artifacts exercise in society” (Diakopoulos 2015, 398). As algorithmic accountability reporting relies on reverse engineering, a set of techniques used to investigate the input–output structure of algorithms, journalists equipped with the programming skills to build their own automated news systems could also engage with this new type of investigative format, which could lead them to reveal, among others, potential biases in hiring and credit scores algorithms or predictive policing software, thus consolidating the autonomous pole of journalism. These programming skills could be also used to equip data journalists with better data scraping techniques to collect information on the stories they want to investigate.

Adversarial Reactions within the Journalistic Field

A first type of struggle arising from the reactions of media practitioners to the adoption of automated journalism has to do with their autonomy. Indeed, as newsmaking involves technologists and business people as well (Lewis and Westlund 2015), the confrontation between each of their individual habitus could result in increased tensions within news organisations. For instance, although media practitioners and technologists do share similarities in their respective doxa (Wu, Tandoc, and Salmon 2019b), their views could be conflictual due to misunderstandings around what constitutes the boundaries of journalism (Lewis and Usher 2016). Meanwhile, the economic appeal of automated news could translate into less human employment in recruiting strategies (Kim and Kim 2017), which will possibly fuel dissensions between editorial staff and the business side of media organisations.

Such a diminution in human employment could in fact hamper less profitable forms of coverage, such as local news, and ultimately impede on the development of public interest journalism. For instance, from a media management point of view, the savings occasioned through the deployment of automated news combined with the possibility to tailor them to niche and geo-specific audiences (Lokot and Diakopoulos 2016) could constitute an strong economic incentive that would eventually threaten the livelihood of specialised journalists and local correspondents, thus creating an effect of hysteresis for those unable to adapt.

Automated journalism is therefore likely to introduce a considerable reshuffle within newsrooms. A positive outlook would be that automated journalism relieves media practitioners from the challenges associated with immediacy while executing routine tasks such as news recaps, so that they can concentrate instead on more demanding forms of journalism (van Dalen 2012; Clerwall 2014), what would ultimately strengthen the autonomous pole. These could include traditional formats such as investigative or international reporting, in-depth types of news reporting in which the journalist is a participant to the story (e.g., narrative, immersion and “Gonzo” types of journalism) and new journalistic formats tackling the growing datafication of society, such as data journalism or algorithmic accountability reporting. That being said, if reporters are instead assigned to stories reflecting the personal views of media owners or the priorities of the marketing and advertising department (e.g., “clickbait” stories or native advertising), this would reinforce the heteronomous pole and trigger an effect of illusio among media practitioners.

Another area of struggle that relates to automated news has to do with the ideal of objectivity as new forms of co-authorship gradually emerge, either through media practitioners reworking an automated first draft (Wölker and Powell 2018) or through pre-made sentences generated thanks to machine learning techniques that can be directly dragged and dropped into a copy (Lindén 2017). Media practitioners need to remain in control of the story while exerting their expert knowledge, or news habitus, of critical thinking to make sure that the story is overall objective, especially when it draws on a single dataset, and that no algorithmic biases creep into the final copy, which could be the case in advanced machine learning models that suggest pre-made sentences, thus making the field more heteronomous.

To conclude, as previously mentioned, the introduction of automated journalism within newsrooms brings about new perspectives on journalism ethics, thus fostering a discussion on a potential renewal of journalistic standards and practices. Should this conversation take place within news organisations, journalism research centres and professional associations, this would reinforce the autonomous pole, but if this is too closely tied to external organisations such as Big Tech companies, there is an increased risk that heteronomous forces penetrate the field of journalism.

These reflections illustrated the many ways through which automated journalism could change and reinforce the prevailing journalistic doxa. In summary, the influence external forces exert on the journalistic field could be visible through an over-reliance on external datasets, the accumulation of cultural capital would mostly have to do with media practitioners acquiring a computational thinking mindset and the reactions of media practitioners that could trigger conflict within newsrooms relate to potential tensions not only with the business and technology sides of news organisations, but also with external players such as Big Tech companies.

That being said, these changes and reinforcements should be considered in relation to existing journalistic practices that can present similar challenges. For instance, relying on the same data sources to automate news stories is essentially similar to the more traditional challenge of relying on a limited number of authoritative and official sources for routine news, thus hampering the diversity of opinions and voices in the overall media coverage (Gans 1979; Molotch and Lester 1974). Likewise, trying to predict elements of the story in advance while authoring templates for automated news, and risking to pick up a “losing” and a “winning” side before an event occurs, could correspond to “prep copies” that journalists write beforehand, which could be a problem with regard to journalism bias should reporters not refrain from introducing perspectives into their copies the way journalists writing “advance obits” do (Adams 2020).

In the table that follows, we used the reflections that we came up with as a starting point to summarise the essential dimensions to consider for an empirical investigation of the impacts of automated journalism on media practitioners. These dimensions are listed as questions, so as to equip researchers with a resource that would facilitate fieldwork that looks into this area and can be used as such to interview media practitioners.

Conclusion

To answer RQ1, we first reflected on the potential changes and reinforcements that automated journalism could bring to journalistic doxa through a series of self-reflections guided by Field theory concepts. The materialisation of the journalistic doxa through Deuze’s five ideal-typical values of journalism (2005) together with Wu’s, Tandoc’s and Salmon’s (2019a) interpretation of structure and agency to locate the tensions that pertains to newsroom automation proved to be an adequate analytical framework to guide these reflections. In doing so, we demonstrated that the journalistic doxa is, indeed, likely to be changed or reinforced by the introduction of automated journalism, and that the over-reliance on external datasets, the adoption of a computational thinking mindset among media practitioners and conflicts between editorial staff and, on the one hand, the business and technology sides of news organisations and, on the other hand, players external to journalism (i.e, Big Tech companies) could constitute the most pressing issues to be addressed. This stresses the need for an empirical investigation into these impacts, which could draw on the list of essential dimensions to consider that are summarised in the Table 1 that we elaborated in answer to RQ2.

Such an empirical investigation could essentially be constituted of newsroom ethnographies or semi-structured interviews, as future research endeavours need to assess media practitioners’ own evaluations of these impacts and to observe how these pressing issues play out at the level of journalists’ daily working practices.⁴ Once these are exposed, common patterns and possible differences across media organisations could then emerge. If such a research undertaking is fruitful, researchers could also consider using the same analytical framework to examine other instances of newsroom automation such as the use of algorithms to retrieving newsworthy data in investigative reporting (Broussard 2015; Stray 2019) or to automate fact-checking processes (Graves 2018). It is also important to keep in mind that, as journalistic actors were actively involved in the “internal tensions” that resulted from the adoption of digital news (e.g., new media increasing competition online, journalists in low-income countries covering outsourced domestic news, media managers paying closer attention to metrics and audience engagement), they can likewise have a role to play in the tensions that may follow newsrooms’ algorithmic automation. These tensions can then neither be singled out as heteronomous forces nor be seen as a manifestation of the autonomous pole only, but rather belong to a “grey area” that would be situated somewhere in-between those two poles.

Finally, it could also be worth looking at employing Field theory concepts to evaluate the impacts of advanced algorithmic techniques on other types of highly skilled professions, for instance in the medical industry as health professionals rely on neural networks to detect skin cancers among patients (Esteva et al. 2017) or in the legal industry as advanced classification or natural language processing methods can be deployed to predict judgements (Katz, Bommarito, and Blackman 2017; Medvedeva, Vols, and Wieling 2019).

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 765140.

Notes

1 Following Meyen and Riesmeyer (2012), English argues that journalistic capital regroups cultural, social and symbolic forms of capital.

2 Essentially, journalistic capital would be mostly constituted of “visible” specialisations that have to do with newsmaking, while news habitus is rather about “invisible” journalistic assumptions. However, news habitus could also be considered as one of the specialisations of journalistic capital, even though it is not as obvious as the other ones.

3 A revision of the exact location of a 1925 earthquake that occurred off the coast of California mistakenly triggered a United States Geological Survey alert that was sent to newsrooms across the country, prompting The Los Angeles Times’ Quakebot to publish this information as such.

4 Particular attention should be paid to the differences between developing automated news in-house, outsourcing their production to an external NLG provider or resorting to a third-party tool so that journalists can craft their own automated stories, just as the outsourcing of copy editing services to different providers rose questions that relate to journalistic credibility and control on ethical and professional aspects of news production (Martin and Martins 2018).