I am admitting my mistake

About the use of AI and hallucinations

Since October, I have been writing here about journalism. What threatens journalism? Why is good and transparent journalism so important for our society? How can journalistic organisations deal with mistakes? What is the value of journalism in times of AI? I try to address these questions with dedication and care.

But I was not careful enough. An NRC staff member—the newspaper I led for nine years—noted that he could not verify certain quotes I had used in editions of this newsletter. He brought these to my attention before publishing an article about them. I include a link here to his piece in NRC.

If someone else would have done this, I probably would also write about this.

Let me be clear: he is right. I acknowledge that, when I started my newsletter in the last quarter of last year, I used AI language models such as ChatGPT, Perplexity, and Google Notebook while writing. I was enthusiastic about the possibilities these tools offered and wanted to experiment with them extensively.

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.

It is particularly painful that I made precisely the mistake I have repeatedly warned colleagues about: these language models are so good that they produce irresistible quotes you are tempted to use as an author. Of course, I should have verified them. The necessary “human oversight,” which I consistently advocate, fell short.

When I realised this a few months ago, my enthusiasm diminished, as did my use of AI. Yes, I still use it—for translations, for working through documents, as a sparring partner for ideas, possible headlines, and angles—but with far less naïve trust than before.

One more point about transparency: I made a second mistake. When I knew that some quotes were incorrect, I should have corrected them immediately. Instead, NRC did so in my place.

Together with Mediahuis I decided to temporarily suspend my activities as Fellow “Journalism and Society”.

Journalism is human work. I remain convinced that AI can be a powerful tool—one that can help journalism become better, dig deeper, and be more precise. But not by using AI in the way I did in the early months of this blog.