In a recent podcast from Semafor Media’s Mixed Signals, the acclaimed magazine journalist Patrick Radden Keefe discussed London Falling, his latest work of narrative nonfiction. The book, described by The Guardian as “a compulsive tale of money, lies and avoidable tragedy”, started life as a long-form piece in The New Yorker, and is an investigation into the death of a 19-year-old man called Zac Brettler, who in 2019 fell from a London luxury apartment balcony. The investigation uncovers Brettler’s secret life, posing as the heir of a wealthy Russian oligarch, and how he got caught up in London’s dangerous criminal underworld.
This column isn’t about London Falling, though, but about the craft of journalism necessary to produce a work like this. And by extension, about how we define journalism in an era of AI-slop and over-engineered clickbait, one powered by the adoption of AI solutions into newsrooms. I’ve written about this before, emphasising the point that if you, the reader, can’t find a news source that you can trust, and that you’re comfortable recommending to other people, one of the glues that holds our democracy together risks coming unstuck.
This is not to suggest there is one source of truth. It’s to suggest that, no matter our political or cultural views, we can all agree that there is a way for us to understand that there is such a thing as a trusted news source, rather than the wasteland of mistrust that bad actors want us to inhabit.
Keefe is the author of several books that started life as journalistic pieces. 2021’s Empire of Pain: The Secret History of the Sackler Dynasty, for example, followed Keefe’s 2017 New Yorker article on the Sackler family, titled “The Family That Built an Empire of Pain”. The book recounts the history of the Sackler family and their company Purdue Pharma, and reveals how their aggressive marketing of OxyContin helped fuel the US opioid epidemic, allowing them to reap vast wealth while shielding themselves from accountability and blame.
I tend to think the best ideas aren’t on the Internet. You’re probably not going to find them by just sitting there on ChatGPT or Google
— Patrick Keefe
In an April 17 interview with Mixed Signals, Keefe said: “I think it’s good to stay curious and stay in conversation with people. I tend to think the best ideas aren’t on the Internet. You’re probably not going to find them by just sitting there on ChatGPT or Google.”
This struck me as an indictment of those who believe (and I count myself among them, to a degree) that AI will replace journalists. Not all journalists, to be sure. And a better way of putting that belief might be to say, AI will replace much of what constitutes the production of journalistic content, but it will do a very bad job of replacing investigative journalism.
Stellenbosch University’s Centre for Information Integrity in Africa (CINIA) has just released an interesting report entitled “Navigating risks and rewards: How South African journalists use AI in the newsroom”. The report doesn’t give an explicit, fixed definition of journalism, but it does give an implicit one, by way of journalism’s functions and values.
As you might imagine, this implicit definition emphasises newsgathering, verification, editorial ethics and oversight, and the report suggests that implementing AI in journalism “necessitates a reaffirmation of journalistic basic principles, forcing journalists to define their professional logic as a separate ‘creative process’ that gives crucial context and meaning.”
There’s a data visualisation that troubled me, though. It asks the survey respondents “What tasks do you use AI for?” According to the report: “The chart shows that transcribing speech to text and language translation make up the majority of use cases of AI in the sample of journalists surveyed.”
In fact, there are four data points. 20% of journalists use AI for “language translation”, 29% for “speech to text transcription”, and 9% for “other transcription purposes”. The biggest single category by far, at 43%, is “other”.
What is this “other”? I don’t mean to imply that it’s a mystery. The report does answer this question, and details other uses including AI‑assisted research, summarisation, drafting and reversioning of copy, editing support, and multimedia editing and image generation. But still, it struck me as an ominous catch-all. If there are so many disparate uses of AI, how are we ever going to be able to control how its deployment affects and changes journalism?
Deployment is the key term here. Publishers and business owners are AI lemmings, leading the charge to jump over the cliff of replacing people with machines. Once you’ve jumped over that cliff, though, you can’t change your mind halfway down. To take us back to Keefe’s quote, “the best ideas aren’t on the Internet. You’re probably not going to find them by just sitting there on ChatGPT or Google.”
The CINIA report digs into some of the pitfalls of using AI to produce journalism. One journalist interviewed said (the report goes so far as to use the word “worryingly”), “‘It mistranslates political or cultural terms,’ highlighting the inability of such AI tools to pick up nuance and context. This is a potentially hazardous flaw when trying to understand cultural and racial identity issues, and where accuracy is instrumental to news storytelling, and establishing trust and credibility with an audience.”
The report further points out that, “by creating what one journalist described as ‘bland copy’ when delegating translation tasks to AI chatbots, such nuances or contextual understandings are lost. In a country like South Africa, where political and cultural terms have been weaponised by disinformation practitioners, the potential for journalistic credibility to be lost is high if newsrooms do not carefully understand and manage how their teams use AI.”
At this point, you might be wondering why you should care if journalists are replaced by AI. One reason I would advance is that, while AI companies audit their training data to reduce bias and errors, add various safety guardrails, and follow governance frameworks, there is no way to judge whether AI has your best interests at heart. And not just because AI doesn’t have a heart.
The CINIA report discusses how “some sceptics argue that AI fundamentally shifts the relationship between reporter and audience as reporters are effectively ‘turned into writers and editors’, suggesting a mechanical relationship that strips content of its humanity.”
In a New York Times interview, Keefe alludes to why this humanity is important. His interviewer writes: “We started having a conversation that journalists have all the time, in these days of artificial intelligence and political turbulence and nonstop distraction: Does what we do matter? I said something self-deprecating. But then Keefe surprised me. He said something positive. ‘I don’t want to sound self-righteous at all,’ he said. ‘But I feel like I’m living a righteous life.’
“He said that everywhere he looked, there were people, including a number of other Yale Law graduates, who were doing morally grotesque things. ‘Listen, I may not be part of the solution here in any fundamental way,” he said. “But I’m not part of the problem. And that’s something.’”
Leaving aside the question of what constitutes righteousness, and perhaps risking a naive definition of how AI works, it seems to me that AI is trained on the morally grotesque as much as on the righteous, and we just have to look at Grok to understand that. We can only be sure of what we consider to be journalism we can trust if we are reading a journalist who has built trust through action.
Mixed Signals describes Keefe as “one of the last [celebrity journalists] who’s not standing in front of a camera or their laptop and a microphone producing content.” This is a little dismissive of the new types of journalists, and I wouldn’t want to be part of that.
There’s no point in lamenting the fact that old-school journalism platforms are dying, if not dead, and that audiences now want to consume journalism that’s produced in a new way. What is important is working out how we can redefine journalism, and how to take advantage of the new economy of meaning that AI has brought us.
I use the word economy here to refer both to a system, as well as the act of curtailing meaning. Journalism is supposed to present a limitless understanding of the world, not one hamstrung by the limitations of technology.
Would you like to comment on this article?
Sign up (it's quick and free) or sign in now.
Please read our Comment Policy before commenting.