Sign up for the daily CJR newsletter.
Three weeks ago, a Dutch freelancer named Menno van den Bos contacted CJR and the Tow Center for Digital Journalism to tell us he had found a writeup of our Journalism 2050 issue in his country’s press that he suspected of containing AI-generated inaccuracies. In itself this is not surprising. As we have written here before, the incidence of fake quotes and citations from journalists and academics is a growing menace. The Tow Center is frequently cited in error, I and other journalists and researchers are misquoted, and a core citation problem in news that Tow researchers originally identified a year ago is not only not improving, it is generally infecting journalism and academic citation on an increasingly alarming scale.
What was surprising about the incident was that the AI distortions came from a very senior figure in European journalism, Peter Vandermeersch, a former editor in chief at leading Dutch newspaper NRC and chief executive at Mediahuis’s Irish division. Now a kind of intellectual emeritus at Mediahuis, Vandermeersch is an in-house sage, in the official position of a “Journalism and Society fellow.” Or at least he was, until the unchecked and inaccurate use of AI in his work caused his suspension.
Vandermeersch made up quotes from eight different commentators, academics, and journalists from our Journalism 2050 issue. None of the experts appeared anywhere in the project. Their quotes did not exist anywhere else that van den Bos could find. I was one of the quoted commentators, and I do cohost the Journalism 2050 Podcast, but this quote was not something I had said. Not on the podcast, nor, as far as I can tell, anywhere else. Van den Bos’s reporting into Vandermeersch’s Substack uncovered incidents of using AI beyond this one: in total, fifteen out of fifty-three posts contained AI-generated, or otherwise fabricated, quotes.
In a mea culpa, which truth be told could have been a bit more culpa, and a lot more mea, Vandermeersch describes what happened: “Even I—with all my years of experience and knowledge—fell into the trap of hallucinations. I summarised reports using AI tools and worked from those summaries, trusting they were accurate. In doing so, I wrongly put words into people’s mouths, when I should have presented them as paraphrases. In some cases, it reflected my interpretation of their words. That was not just careless—it was wrong.”
There is another problem here—these words could not have been paraphrased because they did not exist. I wasted many minutes of my averagely priced time looking for where I might have said something asinine about “immersive knowledge.” Google AI Overviews pointed me to a speech made by someone else, in which I was quoted, but not about that.
Writing in the Irish Times about the unfortunate case, Fintan O’Toole neatly summarized the embarrassment of his former colleague as boiling down to there being no effective “responsible use of AI” in this type of journalism.
“So what can have come over him?” asks O’Toole. “I suspect the answer lies in his life as a newspaper business executive who can’t help having an eye on the bottom line. That part of his self-described role—‘exploring the responsible use of AI in newsrooms’—is not an abstract intellectual quest. It’s a search for ways to cut more jobs and make higher profits.”
At the heart of this problem lies the original sin of the relationship between platforms and publishers. The haste of publishers to think that, when a technology company presents itself as providing “help” to journalism, this is in fact what it is doing. In nearly every example I can think of, the opposite has been the case. While executives paid far more than editors lectured newsrooms about trust being the “number one problem” for journalism, and shoveled money into underwriting research that “proved” this, their own aim was to borrow the authority of reporting without paying for it, and to create products that washed away the foundations of fact in a tide of cash.
In Paris last week, at the excellent Saving Journalism conference put on by Columbia and McGill University, Julia Angwin talked about the class action case she was filing against Grammarly for appropriating her name and reputation—along with those of dozens of other journalists—to suggest to their users that we were in some way lending our editing skills to their product. In the United Kingdom, another AI authorship scandal broke this week when the new book by Matt Goodwin, a right-wing commentator and would-be politician, appeared to be riddled with made-up “facts” and quotes that didn’t exist; the occasional citation appeared to have ChatGPT left in the URL, pointing to a possible source of the absence of truth.
Sometimes, as a journalist, it is embarrassing or uncomfortable to be held accountable for the things that you said or wrote. Those of us who have been trundling around the hamster wheel of word production, for longer than the internet has existed, now live in a gloam of low-level anxiety that an offhand remark or an aside at a panel will become ingrained in countless AI biographies as the very essence of our intellectual belief system. Attributing values, opinions, actions, and ideas to people who do not hold them has always been a flaw in the worst excesses of journalism. Accountability, although thin and unevenly applied, was at one time a possibility. This too is now largely illusory.
Has America ever needed a media defender more than now? Help us by joining CJR today.
