Collaboration key to leverage AI in newsrooms: Findings from JournalismAI

JournalismAI is a project of Polis, the journalism thinktank of the London School of Economics. This global initiative is funded by Google News Initiative and empowers news organisations to use AI responsibly.

In 2019, when JournalismAI came into being, the team released their first report on the use of AI in newsrooms globally. Their second report, in 2023, talked about the use cases of GenAI in newsrooms around the world. Of the 105 newsrooms interviewed for this report, 75 percent noted that they use AI for functionalities across the news cycle - from news gathering to production to distribution.

An issue common between JournalismAI's 2019 and 2023 reports were the gaps between small and legacy news organisations, in terms of resources, staff strength and funding required to use AI at scale, said Lakshmi Sivadas, Senior Programme Manager at JournalismAI, Polis LSE, at our recent AI Summit in Bengaluru.

And to bridge that gap, she said, comes the need for collaboration.

AI's current role in newsrooms

The primary reason for newsrooms to use AI is to improve journalistic and organisational efficiency and productivity, Sivadas said. She cited current use cases from the JournalismAI fellowship across news gathering, productionand distribution.

News gathering

• MP Interests Tracker:This UK-centric project, being developed through a collaboration between the BBC and The Times, will help journalists uncover who pays MPs (Members of Parliament) and who gives them gifts using publicly available data. By simply installing the tracker's Python library, data journalists can leverage the tool to automate the process. The tracker handles downloading the Register of Members' Financial Interests, cleaning and segmenting the raw data, and using GenAI to extract and refine entities. The final cleaned data is then ready for journalists to use in their research and stories.
• Real Estate Alterter: This is a collaboration between the Detroit Free Press (USA), Gannett (USA), The Globe and Mail (Canada), and E24 (Norway). This tool will use anomaly detection methods and Large Language Models (LLMs) to uncover hidden news stories within real estate data, to make use of in breaking news and investigative stories.

News production

• Oriel: Developed by BBC, this toolaims to enhance efficiency and discoverability in image search, thereby saving journalists' time. The team uses machine learning methodologies to sift through images based on multiple filters such as the number of people in an image, gender, and emotions.
• StyleCheck: AFP (France), Novaya Gazeta (Latvia), and Irish Examiner (Ireland) collaborated to build an AI-driven tool to check the adherence of articles with the newsrooms' style guide, thus improving the quality and credibility of the produced news. Integrated with a cleanup tool for article copy, it will provide suggestions that can be overwritten.

News distribution

• AI-powered article summaries: South Africa's Daily Maverick went about experimenting with ChatGPT to help create summary articles and summary cards. And with that came the realisation of the ups and downs of working with the technology but ultimately improving reader engagement.

WAN-IFRA's Media Watch | Take the pulse of AI in news media

The ideal role of AI in newsrooms

Sivadas listed a few guiding principles learnt from working on these reports with journalists and technologists.

First is that AI should perform tasks that humans realistically cannot. "The Idea here is to not throw AI at every little thing, but to problem solve through that lens and see if AI is indeed the way to go. Perhaps, a simpler tech or a human can do that task," she said.

Second, AI should be a catalyst for bridge building. The report findings emphasise that collaboration between news organisations, intra and inter-departmental collaboration, and collaboration between academics and news organisations will be vital, going forward. "The idea is to build interdisciplinary teams and have a more diverse information ecosystem," she said.

Third, AI should improve user experiences. "The focus here is on the fact that ultimately, it's your users who will be interacting with the system, which very few people give thought to," she noted. "The recent Reuters report talks about how GenAI is preferred by younger generations. This doesn't mean you cater to only that one segment, but audience-based segmentation is key."

Finally, AI should do no harm and mitigate harm. This could be possible through improving transparency with consumers while building AI systems, Sivadas said. "AI use cases go beyond just text/headline generation and translation. It's important to educate the consumer, and disclose the use of AI in all cases," she said, adding, "It's also essential for every newsroom to have an ethics panel in every newsroom."

How might AI in journalism evolve?

Sivadas said their research foresees AI to have continued use in finding needles in editorial data haystacks, enhancing the practicality of how journalists work. She listed a few examples.

• AURA: Advanced Understanding and Research Assistant is a conversational AI platform designed to help journalists navigate and extract stories from complex, unstructured data. By leveraging GenAI, AURA offers instant context and investigative leads. This is a collaboration between Indian Express, The Economist, Danish Broadcasting Corporation and Sweden's Aftonbladet.
• IntelliNewsComparer: This GenAI-based document comparison tool will employ machine learning to semantically compare text documents in English, Finnish and Tagalog. This tool is being developed by GMA Network (Philippines) and Helsingin Sanomat (Finland).
• Sivadas also sees a new focus on providing hyperlocal and personalised "services" that meet the information needs of individuals and societies. "We will see a switch from merely writing stories to actually meeting the information needs of consumers," she said.

"We will see improved, and in most cases, completely new forms of user experiences to meet those information needs," Sivadas said, citing two examples of chatbots:

• Know Your Leader:This AI-powered chatbot aims to watch over political leaders' positions taken in the past, and make it possible for the public and journalists to track political speech, comments and opinions. This is being developed by Al Jazeera and Nigeria's Centre For Journalism Innovation & Development.
• Data Robot Aide:This open framework will help newsrooms create LLM-powered AI chatbots from different structured datasets such as elections, census, crime, healthcare etc., to speed up storytelling and initialise story ideas. It is being developed by The Washington Post, Bloomberg, Il Sole 24 Ore (Italy), and India Today.

Future areas of AI integration

"AI will drastically change the way we interact with information. To keep up we must educate, introspect, and be open to adapt to newer forms of journalism which may or may not involve AI. We need everyone who is interested to have a seat at the table of making these AI decisions," Sivadas said.

JournalismAI's 2023 report chalked out four potential areas for the future integration of AI.:

• Fact-checking and disinformation analysis
• Content personalisation and automation
• Tech summarisation and generation, specifically in terms of push messages, summarising and headlines.
• Conducting preliminary interviews and gauging public sentiment: "The newsrooms we surveyed said this was a good aid before conducting the actual interviews," she said.

Neha Gupta

Multimedia Journalist

[email protected]