Until last October, Peter Vandermeersch was one of the most powerful figures in Irish media.
First as publisher and then chief executive of Mediahuis Ireland, he was responsible for operations and strategy at the country’s largest newspaper group. His stint in Ireland followed a distinguished journalistic career that included roles as editor at De Standaard in Brussels and NRC in Amsterdam.
During his time running Mediahuis Ireland, Vandermeersch cultivated a public profile as a straight-talking, clear-eyed Belgian unafraid to confront the harsh truths facing the business but committed to protecting quality journalism.
The picture one hears from current and former employees of the company can be rather different. They describe a reluctance to invest in original reporting and a preference for diverting resources elsewhere, particularly into podcasts.
READ MORE
Last Thursday Mediahuis announced that Vandermeersch had been temporarily suspended from his position as the group’s “Journalism and Society” fellow, following revelations in NRC, the paper he used to edit, that he had attributed false quotes to multiple media commentators in his Substack newsletter.
Some of these posts also appeared as articles in Mediahuis titles. The Sunday Independent and Belfast Telegraph have since taken these down.
One of those misquotes was attributed to Columbia University professor Emily Bell. When asked by NRC if she had said that “the next frontier of journalism is not immersive technology, but immersive understanding”, she responded that “I certainly hope not, because it makes me sound like an idiot”.
As Bell points out, fabricated online quotes have real-world consequences and often end up being impossible to extinguish once they’re let loose on the internet.
The difficult choices that could dramatically increase housing supply in Dublin
For anyone familiar with artificial intelligence’s (AI) large language models, the Bell quote reeks ofAI slop: superficially plausible but clearly inane. Unfortunately, the same could be said of much of the corporate-speak from media executives that long predates the arrival of AI.
Journalists are, and should be, a suspicious bunch. Many are sceptical that high-flown rhetoric about values and democracy counts for much when it comes to the bottom line, and that senior executives will, when necessary, be happy to reduce human headcount and roll out AI automation of articles and editing. Those sceptics will feel justified by the events of the past week.
The glaring irony here is lost on no one, including Vandermeersch himself. In the explanation he posted on Substack last Thursday, he acknowledges that the central theme of the articles he has been posting since becoming a Mediahuis fellow was the need to maintain journalistic standards in the age of artificial intelligence and digital disruption.
The false quotes, he explains, were taken from summaries and queries generated by AI tools such as ChatGPT, Perplexity and Google Notebook. They were “hallucinations” of the sort that have become familiar to users of these tools, and his error lay in not adequately verifying them with a reputable source.
Vandermeersch declined to comment further when I contacted him for this article.
But his Substack post leaves a number of questions unanswered. Did the AI tools hallucinate entire quotes, presenting them falsely as direct speech, or did he take AI summaries of people’s views (which may or may not have been accurate), and transpose them into direct speech?
[ Avoiding the temptation to replace journalists with AI is crucial for news mediaOpens in new window ]
That is not a minor distinction. If the quotes were pure AI hallucinations, the fault lies in Vandermeersch’s failure to verify them. If they were transposed from indirect to direct speech, that represents a further failure of journalistic standards.
More questions are raised by what Vandermeersch describes as his second mistake. It reflects well on the Mediahaus-owned NRC that it appears to have reported without fear or favour on a controversy surrounding a former senior colleague. But Vandermeersch acknowledges that he became aware of problems with some of the quotes months ago, yet did nothing to rectify them until the Dutch paper’s story broke.
Whatever about the Substack, which he presumably manages himself, Vandermeersch had a clear obligation, once he became aware of the problems, to inform the editors of the Irish newspapers that had published some of this material. These, after all, were editors who only months previously had reported directly to him as chief executive. He does not explain why he failed to do this.
Over the past year media organisations have been drawing up guidelines on how their journalists should and should not use AI. Mediahuis’s own guidelines are a model of clarity. They recognise the ways in which these tools can streamline routine tasks, augment certain forms of research and possibly automate some back-end processes.
But they also call for renewed attentiveness to basic journalistic principles. None of these principles is new: always verify that information is accurate and comes from a reputable source; ensure all direct quotations represent the actual words spoken; correct mistakes promptly and transparently.
Vandermeersch acknowledges he failed to meet those requirements. The reason he offers is that he was, in his own word, “enthusiastic” about the possibilities of AI, and that he underestimated the power of hallucination.
[ Mediahuis Ireland prepares for the digital stormOpens in new window ]
He says he now uses AI more judiciously, primarily for translation, document analysis and as a sounding board for ideas. But while he says he previously used AI “while writing”, he doesn’t address the other obvious question: were any of the articles themselves AI-generated?
Some believe a cordon sanitaire around journalism is required to protect it from the corrupting effects of AI. That seems unrealistic and probably undesirable.
Do journalists really want to cut themselves off from a technology that, for good or ill, is transforming communication? The unfortunate truth for Vandermeersch, though, is that this episode illuminates AI’s challenges to day-to-day journalism more vividly than anything he has written, with or without assistance from a large language model.
