Bonds rarely make for high drama. Yet, an op-ed in the Wall Street Journal recently drew a lot of attention. Not for the argument itself, but because he used AI to help write it.
What’s different here is the investor reaction: a big ¯\_(ツ)_/¯. “I was a B student in English, but an A+ in economics,” he said. “At 73 I’m kind of proud of using it,” reacted Stanley Druckenmiller.
The Journal’s opinion editor doesn't seem bothered either. "AI is a fact of modern life.". "The question for us is whether what we publish from contributors reflects an author’s original argument, and if the author has the standing and credibility to make it."
A few weeks ago, the Financial Times took a different path and added an editor’s note to a column by Harvard professor Ricardo Hausmann after discovering that he had used AI to shorten a longer draft. The newspaper said this violated its editorial rules.
Same tech, two very different ideas about what a byline guarantees.
What we are watching is professional norms taking shape in public and everyone seems to be arguing about a slightly different thing.
AI doesn’t come with an instruction manual
We shouldn’t be too much surprised by this debate. In 1994, Wanda Orlikowski and Debra Gash described how people understand the same technology through different assumptions and professional values - technological frames. That’s basically what’s happening here:
For the WSJ’s opinion section, an AI-assisted op-ed can still be authentic if the underlying argument belongs to the person whose name appears on it.
For the FT, the process of producing the words is also part of the editorial contract.
And even inside the same publication, those standards differ. The Journal’s opinion section and its newsroom operate separately and have different approaches to AI use.
In other words, we are still negotiating what counts as assistance, authorship and what readers believe they are being promised.
Trevor Pinch and Wiebe Bijker described a similar process decades ago in their work on the social construction of technology. New technologies do not show up with one stable meaning attached. Nor is adoption just about whether something works. It also depends on whether people consider it appropriate and eventually unremarkable.
Druckenmiller’s unapolegetical stance is interesting precisely because it pushes against the current stigma. He considers the question already settled.
For plenty of editors, readers and writers, it clearly hasn’t. Writing is the most contentious flashpoint in the media industry.
We already had this argument about ghostwriters
All comparisons are imperfect, but ghostwriting provides an interesting anchor in this debate. Politicians and executives have relied on speechwriters and comms teams for decades, and we find it normal. But not for everyone.
In 2004, legal ethicist Stephen Gillers wrote about ghostwriting and that sounds remarkably familiar today. He argued that ghostwriting could be acceptable for politicians or CEOs because readers primarily care about their ideas. But it might be more problematic for academics whose work includes developing and articulating those ideas themselves. Sounds familiar?
It might seem strange to assume that paid human help is automatically authentic while technological help is automatically suspicious.
What’s in a name?
The scientific litterature offers a few interesting studies on authorship:
In a study involving 602 participants, researchers asked people to assess a fictional author who received different levels of help from either a human assistant or ChatGPT. The main factor shaping perceptions was how much help they received, not whether the assistant was a person or an AI system. When the assistant contributed more, participants were more likely to expect disclosure.
Another study of 253 people found that the stage of the writing process matters too. AI help with planning reduced writers’ sense of ownership only slightly, whereas AI-generated drafting produced a much stronger decline.
This recent conversation between Nick Thompson and Steven Johnson, who created NotebookLM (Notebook if you prefer), is worth listening to. Johnson argues this:
“Is there a world where, as these models get better at understanding prose style, you would start to steer the model toward the paragraph you want, but let it write a little bit more of it? I don’t know. I probably wouldn’t do that because I like to write. But for folks who have a hard time writing or aren’t professional writers, is that a valid way to generate paragraphs? (…) That seems acceptable to me as a potential future.”
In other words, is AI a sparring partner or is it taking your place in the ring? Using it to organize notes or challenge an argument plays a very different role from having it supply the argument itself.
Call me Captain Obvious, but the two do not preserve the same relationship between the writer and the work. The real risk is outsourcing the thinking. Writing is often how we discover what we actually think. Hand off too much of that process and you might end up with a persuasive argument that you never really examined.
A study of 319 knowledge workers found that greater confidence in generative AI was associated with less self-reported critical thinking during AI-assisted tasks. It doesn’t prove AI inevitably makes people dumb thinkers, but it suggests that AI has an impact on our our judgment when it becomes too convincing or too convenient. Do we challenge the machine appropriately? And do we have the skills to do so?
This seems to me a more interesting question than whether an em dash appeared in the final copy.
Detection won’t settle the argument
You can pangram Druckenmiller’s piece and get a verdict that it’s an entirely AI-generated text. Druckenmiller disputed this and said he rejected many of the system’s suggestions.
Yes, a detector can offer an estimate about textual patterns, but it can’t give you breadcrumbs of the writing process.
Research also shown that people are surprisingly bad at identifying AI-generated writing. So does a recent preprint analyzing Reddit and Hacker News comments. Accusations can become a way of policing what feels authentic rather than establishing what actually happened.
That creates a strange incentive. People who openly acknowledge using AI can be punished for their transparency. I mean, who hasn’t run their own text through Pangram to see how it rates them, even though they know they wrote it themselves?
A proposal of a jurisprudential test
Instead of just pasting a text in an obscure blackbox, we as editors or readers could ask better questions. A starting point could be:
Did the ideas come from the person whose name appears on the piece?
What did the system actually contribute (editing, translation, reporting…)?
Who checked the facts and the sources?
What does the byline lead readers to expect in this particular context?
Would knowing how AI was used extensively change how readers interpret the piece?
There’s no universal answer, especially because an op-ed or an investigative report carry different promises. But treating all AI assistance as equivalent makes the whole debate less useful.
And there’s one more irony here: we’re spending a lot of energy policing the purity of the finished article at the exact moment AI is changing how people access information in the first place. The article is already becoming less central but we’re focusing on the proverbial tree hiding the forest.
P.S.: I ran this text through Pangram and got a 68% AI-generated score, even though I only used AI to proofread it and suggest a few headings. QED :P
Six Journalists, Six Different Lines. Where Do You Draw Yours?
WSJ and WIRED both just published pieces about journalists using AI to write their stories. Same day. Same taboo cracking open.