Rising to the Challenge: Applying Generative AI in Newsrooms
The journalism industry is all too familiar with technology-driven disruption and has often been accused of being slow to adapt to new technological realities like social media, smartphones and the internet itself. The advent of large language models (LLMs) and generative AI seems like another of these disruptors, but journalism isn’t lagging this time. In newsrooms around the world, senior leaders and working journalists alike are recognising the potentially dramatic power of these new tools and are actively seeking to apply them in their workflows, products and strategies. A decade of innovation projects, lessons from the growth of social media and hard-won operational experience in applying big data and machine learning to recommendations, subscription management and investigative journalism have left the news industry better equipped to handle AI than many assume. This article describes a global programme in which 12 digital-first newsrooms with track records of entrepreneurial innovation are developing generative AI projects that are both pragmatic and potentially transformative for their journalism. Developed and funded by the Open Society Foundation, the AI in Journalism Challenge — or AIJC — is now in full swing. Participating newsrooms are fully immersed in their projects, and some useful observations are emerging.
The AIJC programme is structured as a competition in which participating teams are provided with educational, mentoring and financial support for developing their AI projects. The 12 participating newsrooms were selected from a pool of 113 applicants who responded to a public solicitation with a brief project proposal. The selection process focused on the degree to which the proposals were pragmatic, measurable and potentially transformative, as well as on the commitment that the applying teams were prepared to make to their project. Following selection the participating teams were then provided with a series of workshops about aspects of generative AI, as well as with a dedicated 6-module version of the London School of Economics’ JournalismAI Academy for Small Newsrooms. Each team was also provided with one-to-one mentoring as they developed their projects into detailed specifications and designs, and each was provided with a £5000 development grant to fund their work. The 12 participating teams are now executing those projects and will demonstrate them to a judging panel of news industry experts in mid-October. Five finalist projects will be selected, and a final round of demonstrations and judging will be done at the Splice Beta journalism festival in Chiang Mai, Thailand in November. The winning project will receive a £25,000 grant, and the lessons learned from the full programme will be shared publicly in a comprehensive report. The AIJC programme is managed by Open Society Foundation staff, and I have participated as the lead consultant on applied generative AI.
Digitally native and purpose-driven newsrooms in Asia, the Middle East, Africa, Eastern Europe and South America are some of the most innovative news organisations in the world. Imbued with scrappy pragmatism and with a history of delivering high-impact journalism with relatively few resources, these deeply entrepreneurial news organisations often outpace their larger, better-resourced peers in the UK and US. The AIJC challenge is designed to take advantage of that creativity and energy to find useful ways of applying generative AI in news, and its participants reflect that ambition. The cohort includes: Agência Pública, an investigative journalism agency in Brazil; Cuestión Pública, an investigative journalism newsroom in Columbia; Daraj Media, a pan-Arab news platform; Initium Media, a Singapore-based digital media outlet serving Chinese-speaking readers worldwide; Meduza, a Russian-language news website covering Russia from Latvia; PumaPodcast, an award-winning podcast production company based in Manila; Rappler, the online news website in the Philippines founded by Nobel Peace Prize laureate Maria Ressa; Raseef22, an Arabic media network based in Beiruit; Rubryka, a Ukrainian online media outlet specializing in solutions journalism; Scrolla, a South African news start-up also covering Nigeria; The Conversation, a media outlet based in Indonesia publishing expert-written news stories, opinion and analysis, and part of The Conversation network globally; and Zamaneh Media, an exiled Iranian news publisher and broadcaster based in Amsterdam. Each of these organisations has fielded teams of 3 to 4 editorial and product staff for their AIJC project, often including senior leaders, and each has developed a project plan that reflects their pragmatism as well as the strategy of their organisation.
These projects cover a broad range of applications. Some are focused on using generative AI to adapt content to better suit the needs of new audiences, thereby broadening the reach of the organisation’s journalism. These content adaption projects are usually aimed at younger audiences, and typically seek to use generative AI models to create simplified, restyled or multimedia experiences from text articles. Other projects are focused on newsgathering, using large language models to analyse large streams or repositories of content to identify potential news stories. There are projects that seek to use structured data from investigative projects to develop a series of short summaries suitable for communicating those stories in a series of social media posts, projects that monitor the news environment in a media market and use LLMs to assess the impact of specific journalism in those societies, projects that seek to optimise news content for distribution and projects that seek to establish systems for designing, evaluating, managing and deploying news-related prompts for language models. The 12 AIJC projects also cover a broad range of technical approaches, with many being completely ‘no-code’ solutions, some using code-free ‘drag-and-drop’ online automation tools, some using a thin user interface over a prompt-based API-driven workflow, and some that include more complex software engineering by developers.
The AIJC programme is still several months from completion, and the participant teams are still deep in the design and development phase of their projects. Nonetheless there are already a few interesting observations emerging from the process. There seems to be a clear ‘before and after’ change that occurs when teams begin hands-on engagement with generative AI tools. Once teams sign up for accounts (especially for ChatGPT Plus), spend some time with the interfaces, make some initial attempts at prompting for familiar tasks, and start to evaluate the resulting output for themselves, then their awareness, excitement, motivation and abilities begin to accelerate. There seems to be no obvious relationship between the potential of a generative AI project and the availability of deep technical expertise. Instead it seems that familiarity with the tools and the ability to think clearly about the journalistic objective are more important, and that there are multiple ways of executing these projects in workflows without coding. Granted these newsrooms operate at relatively small scale (~15 to ~120 people) and are unusually flexible, but there may also be lessons here for larger newsrooms. It also seems that most of these newsrooms have no difficulty identifying a list of potential applications for generative AI, often well beyond their initially proposed project. In fact, the ability for teams to retain their focus to just their primary project objective may well become a significant advantage in developing a winning prototype! This is, of course, a self-selected cohort of newsrooms with a predisposition for entrepreneurial risk-taking, however it is hard not to notice the competence and confidence that develops as the teams define and begin their projects. Working with these newsrooms has convinced me that substantial innovation in newsrooms no longer requires dedicated labs, data science teams or special expertise, but merely a willingness to engage with generative AI tools and an ability to think carefully about the newsroom’s objectives, core functions and tasks.
As the newsrooms participating in the AIJC programme develop, complete and deploy their projects over the coming months we will all learn from their efforts. The judging rounds to select first the five finalists, and then the winning project, will provide expert assessments, and the teams themselves will publish their own descriptions of their experiences. Additional articles like this one will appear as the programme progresses, and the lessons learned from the collective effort across all 12 projects will be collected, synthesised and published in the final report. Follow-on projects and programmes will be explored. The global news industry is facing what may be its most significant technological disruption so far, and yet it is already clear that our new access to artificial intelligence can also be used in the service of purpose-driven journalism. Learning how to do that effectively, responsibly and sustainably is a critical, and exciting, challenge.
