Taking stock: the state of play of journalism and artificial intelligence
Ongoing experiments in Italy and around the world
Editorial staff
Artificial intelligence, i.e. the set of techniques for collecting and processing large amounts of data, has been present in newsrooms for years: as early as 2014, the US international news agency Associated Press was already using Wordsmith1, a specific Large Language Model software, i.e. capable of recognising and reproducing words and thus producing content. However, it was only with the popularisation of these models, which took place with the free demo of Chat-GPT in November 2022, that the question of whether, how and why these algorithms should be used and with what consequences became part of the public debate.
Like every editorial choice, this one is not neutral and has an impact on the quality of information, on economic sustainability, on the relationship of trust with you who read and listen to our content and, above all, share our idea of the world: and, as we wrote last week, we believe that, in 2024, a digital editorial project cannot fail to include a thought on the use of artificial intelligence in its processes.
Through this lens, we have observed the dynamics surrounding us on both the “user” side (editorial teams, bloggers, national and international editorial organisations) and the “provider” side (companies offering AI solutions for publishing and information). From this standpoint, we’ve chosen to not only explain how and when we’re utilizing AI in the coming weeks but also to specify which AI we’re using, why we’ve selected it, and the standards it meets. We kick off this week by providing an overview of the AI and journalism landscape, both domestically and internationally.
What is happening in the world
An Associated Press2 study published in April 2024 and conducted on a sample of 300 people employed in European and US media shows how generative artificial intelligence is reshaping the roles and workflow of newsrooms: almost 70 per cent of the sample said they use artificial intelligence technologies to create social media posts, newsletters and headlines, for translating and transcribing interviews, and multimedia content, including social graphics and videos.
AI can be used in the newsroom for both the collection and production of stories. On the collection side, AI automates the search for information in internal and external newsroom databases. On the production side, artificial intelligence makes it possible to increase reporting capacity through the automatic transcription and translation of interviews or the publication of automatic data-driven content (stock market trends, sports results, election results).
Among the more innovative activities, some newspapers are experimenting with the use of artificial intelligence as a “self-criticism” test, to identify and reduce representation bias before stories are published.3
On the distribution side, there are also various systems capable of identifying and analysing the characteristics of audiences and producing, based on these, different versions of the same story, targeting audiences not only their preferred consumption formats but also their age profile and geographical location.
One of the aspects that makes it difficult to implement these AI functionalities in medium-sized editorial offices is the cost of developing or renting algorithmic platforms: on one hand, there is a cost related to the number of activities that one wants to automate or speed up; on the other hand, there is a cost linked to training and maintaining algorithms based on extensive datasets, which small editorial entities often do not have or do not have the resources to produce.
For these reasons, the solutions often chosen represent a compromise between the cost of these models and the requirements of each editorial entity. As in our case: for Mangrovia, we decided to invest in artificial intelligence not to generate more content, but to try to offer added value to our main activity: searching for stories to tell that intersect art, culture, technology, and society starting from places and subjectivities considered marginal.
What is happening in Italy
In Italy, only a few newspapers claim to use AI. The case of Slow News is well known, as it is the first Italian slow journalism project, which has worked on and publicly released a draft AI policy on its website explaining how the editorial team uses generative AI. By declaring the rules governing AI usage in the newsroom, there’s a commitment to establish a fresh pact of trust and transparency with readers, in pursuit of accountability. Such a principle is frequently invoked by journalism from entities and institutions, yet internal implementation can sometimes fall short.
The 2023 report by the Digital Journalism Observatory of the Order of Journalists documents various other experiments, both within Italy and internationally. As stated in the conclusions, the state of the art varies significantly between Italian and foreign newsrooms, with numerous experiments reported: from newspapers cloning beloved TV presenters (Deep Brain AI, in South Korea) to those using algorithms to decide when it’s time to charge readers for articles (The Wall Street Journal) and to predict the emotions an article will evoke (The New York Times), to those creating new teams in newsroom to prototype potential uses of machine learning for the benefit of both writers and readers.
More widespread are applications of AI to analyse user behaviour and realise fully data-driven content strategies (The Times and The Sunday Times have developed James, a technology based on machine learning algorithms that offer the public personalised experiences of the same edition of the newspaper).
In Italy, the report highlights the production of automated news about the pandemic by Ansa, the experimentation with automatic press reviews and subtitling at RAI, and automated editing assistance at “Il Secolo XIX”.
While in the rest of the world, even small publishers are embarking on experiments, albeit always financed by foundations or multinationals in the sector, as emerged from this panel hosted in the last edition of the Perugia Journalism Festival, in Italy innovation is mainly linked to large publishing groups and concerns semi-automated micro-texts, intelligent CMSs and customised experiences.
Beyond individual experiments, the entire field of journalism is at the centre of a regulatory process that aims to regulate the implementation of artificial intelligence technologies while protecting the indispensable social and democratic role of information: the main specific reference in this area is the Council of Europe guidelines published in December 2023 for responsible implementation of artificial intelligence systems in journalism. All these elements are considered in Mangrovia’s experimentation with artificial intelligence.
The cover image is “Hidden Labour of Internet Browsing” by Anne Fehres and Luke Conroy & AI4Media Browsing, distributed under a CC-BY 4.0 licence. This image explores the hidden labour of artificial intelligence in everyday Internet use. The image shows a chaotic intersection filled with reCAPTCHA elements such as zebra crossings, fire hydrants and traffic lights, representing the invisible work in data labelling.
- The Associated Press. (2024, March 12). Artificial Intelligence. Leveraging AI to advance the power of facts. In AP. https://www.ap.org/solutions/artificial-intelligence/. ↩︎
- Diakopoulos, N., Cools, H., Li, C., Helberger, N. (2024). Generative AI in Journalism: The Evolution of Newswork and Ethics in a Generative Information Ecosystem. https://www.researchgate.net/publication/379668724_Generative_AI_in_Journalism_The_Evolution_of_Newswork_and_Ethics_in_a_Generative_Information_Ecosystem. ↩︎
- An example is Janet Bot, the alert used by the Financial Times to rebalance gender images on the homepage. https://labs.ft.com/product/2018/11/07/janetbot.html. ↩︎
