At last year’s World News Media Congress in Krakow Tom Trewinnard and Fergus Bell, co-founders of Fathm that helped WAN-IFRA design and lead its Newsroom AI Catalyst accelerator programme, shared some of the initial learnings from the nearly 130 editorial teams. On Monday in Marseille, the pair gave an update on a year’s worth of learnings from OpenAI-supported programme.
Read about how Prisa Media in Spain prioritises AI governance
Trewinnard kicked off by explaining that publishers are beyond the prototype stage, and now building production systems, but with that it’s become undeniable that projects have to be grounded in real business needs, not a love of the technology’s possibilities.
“The process that we take newsrooms through… is to try and really identify how to apply AI into real work and real pain points.”
Trewinnard pointed to Otto, an AI Chief Operating Officer (COO) developed by the Australian digital publisher Man of Many. The key learnings there were that; “the editorial human-in-the-loop accountability judgments here are really important”, not least for editorial control and the message that sends to staff. He flagged the safeguards built into Otto, for example; “no agent can modify live ad campaigns. No agents can send emails or publish articles.”
Tom Trewinnard presenting the case of Man of Many.
Show me the money
Adopting new technology doesn’t mean forgetting the basics, like ROI. With Otto, Trewinnard notes that “they’re saving about $6,000 a year in subscriptions to enterprise products.” Secondly, there’s the productivity of the people involved; “Senior leadership meetings are going to be reduced from two-plus hours to 15 minutes.”
The power of people
Fergus Bell was even more emphatic about the value of human judgement. “AI lacks intuition … data might be technically correct, but without human judgment, there is a chance that it might be user hostile,” concluding that “human in the loop is non-negotiable.”
He underlined the importance of fitting into the way people already work. “Integration is king.” Why? Simply because stand-alone AI tools, however ‘shiny’, “force journalists to leave the CMS, open new tabs, copy and paste text,” which results in “high friction and zero adoption.”
Similarly, Bell emphasised that one-size-fits-all AI solutions simply don’t acknowledge the fact that every publisher’s actual workflow is slightly different. Imposing a new AI solution without taking into account the ways of working will only heighten resistance.
Don’t let the shiny thing distract from the dirty work
One of the earliest rules of computing has never been more true; “if you put garbage data in, you’re going to get garbage data out.” Bell points out that many AI system failures can actually be traced to the ‘fuel’ that is data. “You need to think about cleaning the fuel.”
In fact, he went further, pointing out that not only is the success of an AI system dependent on the quality and legibility of data, content, and meta content but also that success comes down to being prepared to do the dirty work, cleaning it up first: “you cannot build a scalable AI system until you do the glamorous work of being that data janitor.”
Steve Shipside is a freelance journalist and media consultant who has covered the news industry for more than 25 years, and has contributed to WAN-IFRA’s content for several years.