How algorithms are augmenting the journalistic institution: In search of evidence from newsrooms and its innovation labs

How algorithms are augmenting the journalistic institution: In search of evidence from newsroom innovation labs 

Digitalization has dramatically changed newsrooms in recent years; for example, journalists increasingly use tools to gather, write, verify, and disseminate news. These tools, in the form of algorithms, are latent in the news ecosystem and take the form of recommender systems (labeling what is newsworthy), speech-to-text generators (helping with the writing of articles), or metrics systems (measuring reading behavior through audience analytics). However, little research has been done on how journalists interact with these relatively novel algorithms in the (1) news gathering, (2) news production, (3) news verification, and (4) news production and moderation. This dissertation focuses on this journalist-algorithm interaction and enhances our understanding of how these tools are reshaping rather than reinventing the journalistic institution. 

Based on the results of one theoretical and five empirical studies, this dissertation engages with algorithms and journalism as an institution. The first study proposes a typology that maps the various levels of automation and autonomy in the newsroom. The five-level typology of computational journalism starts from a 'manual level' (level 0, where the news worker has full autonomy) and ends with a 'full automation support level' (level 4, where the algorithm has full autonomy or decision-making power). In doing so, the interaction between a journalist and an algorithm becomes more granular, and the level of automation and autonomy might affect the journalists' perception of whether or not they will interact with that algorithm. In addition, a research agenda is distilled from the existing literature, which recommends where future research in the field of computational journalism could go.

In a second study, a topic modeling and framing analysis was conducted on a corpus of US news coverage on AI and automation from 1985 to 2020. The results of the topic modeling show that articles on AI and automation are most prominent within "Work," "Art," and "Education. In terms of the framing analysis, news coverage has been optimistic rather than pessimistic over time, with an increase in optimism towards AI and automation observed, especially in recent years. However, when looking at the dystopian frames, the results show that in the corpus, there has been an increased focus over the last five years on the impact of AI and automation on ethical dilemmas associated with these technologies. 

In the third and the fourth study of this dissertation, there was a focus on the individual level of the utilization, the interaction, and the perception of algorithms in the news reporting process. For the third study, recommender systems that could label what is newsworthy were investigated in the (1) gathering and the (2) production of news. In-depth interviews with news workers from Europe and North America show that when they rely on suggestions and summaries to evaluate what is newsworthy, they mostly interact with them when there is a 'spike in information' such as elections or a pandemic. In a fourth study, Belgian journalists were interviewed on how they interact with metric systems in the (4) news distribution and moderation phase of news reporting. Results have revealed that journalists in the sample who are younger are inclined to align their editorial choices with the metrics systems, demonstrating that a 'shared decision-making' emerges in the gatekeeping process.

In a fifth and a sixth study, the organizational level of algorithms and the journalistic institution were considered. In the fifth study, innovation lab members from newsrooms in Europe and North America were interviewed as they are often in charge of developing and implementing algorithms in the broader news ecosystem. The organizational structure was considered, as were the perceived roles of these lab members in relation to the broader newsroom. Results have shown that there are three types of innovation labs: static, dynamic, and hybrid forms. Regarding the perceived roles, I conclude that members of innovation labs see themselves as 'service providers'. In the last study, the journalism-algorithm interaction is researched within one news organization, The Washington Post. Based on digital ethnography, it is concluded that the algorithms are trusted by a small group of the newsroom. In addition, the journalists in the newsroom who trust these tools tend to be data-savvy and are thus very homogeneous. Transparency is essential to increase trust in these innovation labs and their tools across the newsroom.

Overall, this dissertation concludes that the development and the implementation of algorithms in the newsroom are in a preliminary state of institutionalization. Therefore, I argue that algorithms are augmenting the journalistic institution rather than reinventing it. Therefore, the utilization of these tools results in a more accurate way of gathering, producing, verifying, and distributing the news. Vigilance is crucial when developing and implementing these algorithms as agency and autonomy need to be guaranteed to sustain algorithmic accountability and transparency in news outlets.