When he returned from his Christmas break, Marten immediately called a meeting with some of his colleagues at DPG Media. They agreed that it was time to get serious about AI. “With the launch of GPT-3.5, the potential of AI began to grow exponentially. Initially, you’d get a lot of garbage results, but now you can get so much more out of this technology. It has definitely made life easier for our editorial teams. You can give the system all kinds of pre-programmed tasks, and the results are excellent.”

These tasks are based on extensive prompts – questions and instructions – which took some time to fine-tune. “A year ago, I used the first version I made to check a story that featured a pastor from Brabant who said something about a ‘stukske vuur’ (‘bit of fire’) in his regional dialect. Which ChatGPT obviously changed to ‘stukje vuur’, the way you would say it in standard Dutch. So I instructed it not to edit quotations, because these kinds of variations occur in other dialects as well.” With the help of early adopters in the various newsrooms, Marten wrote an ever-expanding prompt containing information on what ChatGPT should and shouldn’t use, and which even takes into account the specific style and language level of each individual paper.

The editorial assistant was just one of six ideas Marten came up with during his holiday, which he would later flesh out. Another one was a tool for writing headlines. “Headline quality has a huge impact on how many people read an article, which is why the editorial team is always given several options to choose from. Coming up with all those headlines tends to be a time-consuming process, but with this CMS tool you can instantly generate some good suggestions that will also fit a news article, background story or feature.”

Marten can’t stop raving about his new AI colleague. He sees many more opportunities and isn’t afraid to think outside the box, for example when it comes to football coverage. “Can we use AI to generate the first draft of a match report based on the liveblog we write anyway? That could save reporters a lot of deadline stress. And AI could also extract all kinds of interesting tidbits from the available data during the match itself, for instance that a player is covering much less distance compared to previous matches. Observations like that might give the reporter new ideas.”

Marten also thinks that AI could be used to quickly produce follow-up coverage for a news story. “It would be a tool that could come up with fresh angles and also suggest potential sources for the journalist to call. It’s fine if those lists include some of the ‘usual suspects’, but there should definitely also be a few surprising names – people from more diverse backgrounds.” 

For now, Marten is mainly focused on the journalistic side of the business, but DPG Media is also using AI in other areas, such as sales, marketing and, of course, IT. There are plenty of opportunities to simplify work processes and increase productivity across the entire organisation.

But there’s also scepticism. “Of course, that’s how it always goes with major changes. When trains were invented, people were afraid they wouldn’t be able to breathe going 38 kilometres an hour. A legitimate concern when it comes to AI is that these incredibly smart systems can also be used by people with bad intentions. But that shouldn’t stop us from using this technology to make our work more fun, and to improve things. This change does differ from previous waves of automation in that it’s not affecting low-skilled work. It’s now becoming clear that computers can also do things that used to require highly skilled creatives. That’s going to take some getting used to.”

Sceptical questions about AI usually come from colleagues who have little experience with it. “I recently had a conversation with editors who were worried that AI was going to turn every article into a boilerplate news report. When I showed them how it really works and how they can tell AI in plain language what to do and what not to do, they immediately got excited.”

The opposite also happens, says Marten: some colleagues expect so much from AI that they’re disappointed when it makes a mistake. “AI definitely isn’t perfect, which is why we don’t let it make decisions. It’s just a helpful tool. When it’s used to come up with headlines, the journalist decides what the headline is going to be – AI only makes suggestions. That specific tool is even a hit at home: my wife makes a podcast for the NPO, and every week she asks me if she can use the DPG Media headline generator to come up with a title for the latest episode. So in the end, she also got something out of that Christmas holiday I spent fiddling with AI.”