Last week I asked Google, “Why did bishop Gijsen step down?” (read the article about chaplaincy troubles). I knew it had something to do with sexual misconduct, but I couldn’t remember exactly what had happened. In the AI overview at the top of my screen, terms like “ongoing conflict” and “affairs” popped up. Rather vague, so I clicked on the references to the AD, Trouw, NRC, Wikipedia. Half an hour later, it turned out to be a lot more complex than I had thought. The wonderful (and time-consuming) thing about journalistic detective work is that you can follow a trail of articles and so learn more about the facts of the matter, facts based on reliable sources.
There was rather a to-do about the use of AI in journalism last week. Peter Vandermeersch, former editor-in-chief at NRC, had made a mistake. As a ‘fellow’ of Belgian media company Mediahuis, Vandermeersch had blogged about topics in journalism, such as AI. He knew all about the risks, that AI sometimes just makes up stuff. And now he had copied some of that made-up stuff (quotes) onto his blog. He admitted he had “underestimated the power of AI hallucinations”.
We don’t have any strict rules at the Observant office yet, because we only occasionally use ChatGPT or Copilot. Nevertheless, we are always on our guard. It is a language model that predicts words, not a database of facts. Not to mention that ChatGPT has come under fire after OpenAI announced it had made a deal with the American government. People are worried that AI models might be used in military operations, among other things. That gave us pause for thought at the office, too.
What do we use AI for at Observant? We occasionally use it to translate a few sentences into English (all our articles are sent to real, human translators), or for a spark of creativity for an article’s headline. Before concluding that they’re all horrible clichés. I recently had ChatGPT run a summary analysis on an impenetrable memorandum, just to see if it could make anything of it (answer: no).
Although here we have to hold our hands up and admit, we’re AI-amateurs, not trained in inputting the right prompts (or questions), and that plays a large part in determining what AI produces. Would we call ChatGPT a ‘writing assistant’? Absolutely not. For now, we’re happy to rely on each other’s brains and creativity.
The editor-in-chief gives a look behind the scenes at the editorial office