Autonomous AI tools can be used to asses journalists’ hunches before they sink too much time into an idea.
One expert told Press Gazette this could “enable some rejuvenation of small-scale investigative reporting”.
Dr Chris Brauer, chief innovation officer at AI academic research and advisory firm Symmetry, said new “agentic” AI technology can be used to statistically check the validity of a reporter’s hunch, saving time and encouraging experimental research.
Agentic AI encompasses systems that are designed to pursue complex goals with minimal human intervention.
Brauer said: “In the same way as WordPress transformed the ability of newsrooms to produce at scale through CMS, there’s a really powerful application of this type of technology at any scale of newsgathering.”
There are multiple ways a newsroom can build a hunch-testing AI, he explained. The approach taken depends largely on the data you want to test the hunch against.
If it is a data “lake” like the Panama Papers archive then the system works by converting natural language or unstructured data into mathematical information which can be read and analysed by the AI.
If the data is a content stream, such as ongoing corporate announcements, the hunch can be built to test against every new news item.
Brauer said agentic AI is the most powerful way to assemble a hunch-testing system.
He said: “In an agentic approach, you need to assemble your AI agents which involves creating roles and backstories.
“For instance, you could create a hunch investigator, a hunch rationale writer, and a critical thinking editor, and ascribe all these AI agents distinct goals and access permissions to specific software and datasets in performing their roles autonomously and as a multi-agent team.
“In the same way that you would onboard a junior reporter in the newsroom and provide them with guidance around best practice, you would do the same for the AI agent, in the same natural language.”
The journalist would then input their hunch and the agents would provide a suggestion of how strong the theory is.
The information input into the AI system can be controlled by the journalist to avoid misinformation or inaccuracy.
Brauer said: “You could have the COP Summit coming up and you could scale up an AI agent network whose sole purpose was to evaluate every possible implication of every story and statement.
“This would take about five days but would then do the heavy lifting for you.”
He added: “It may offer promise to enable some rejuvenation of small-scale investigative reporting which is under-resourced.
“You’re encouraged to be much more adaptive and experimental in your reporting – you don’t have to commit to any single angle early on.”
Brauer acknowledged two main obstacles to introducing this agentic AI hunch-checking software: it is computing intensive and it is not cheap.
However, he said: “There’s lots of concerns about the impact of AI on the newsroom and the ways it’s being used to automate and displace people…
“The journalist who is vulnerable in this context is the one who chooses not to consider the role that AI might play in their day-to-day work.”
Email pged@pressgazette.co.uk to point out mistakes, provide story tips or send in a letter for publication on our "Letters Page" blog
