A billionaire investor criticizes the White House in The Wall Street Journal. A Dartmouth provost writes in The Washington Post about what students should learn to do without AI. Someone runs their columns through an AI detector, and suddenly the question is not Are these arguments right? but Did they write every word themselves?
Stanley Druckenmiller gave a direct answer when asked about his Journal column: “Of course I used AI.” He compared it to using a calculator. The published piece, he said, was his message. The Journal’s opinion editor defended it: Druckenmiller had the standing to make the argument, and the argument was his. (NOTUS)
That seems like a reasonable starting point. I want to read Druckenmiller’s judgment about bond markets. I don’t need proof that he personally arranged every comma.
We have been here before, in another medium. Digital photo editing created genuine problems for journalism. During the Iraq War in 2003, a Los Angeles Times photographer combined two images into a single, more compelling scene. The paper fired him. The problem wasn’t that a computer had touched the photograph; it was that the finished image depicted a moment the camera had never captured. (Poynter)
Newsrooms did not respond by abandoning digital photography or demanding that every published image look exactly like an untouched file. They developed—and enforced—distinctions. The Associated Press, for example, allows limited adjustments such as cropping and toning but prohibits changing a photograph’s content. The standard is whether the image still tells the truth. (AP policy)
Text deserves a similarly useful distinction. Did a tool help an author express an argument more clearly, or did it supply claims the author never checked and would struggle to defend? Did the author follow the publication’s rules? Did the finished piece mislead readers? Those questions tell us more than Was AI involved?
Long before ChatGPT, published writing could pass through researchers, editors, speechwriters and ghostwriters. We have also spent decades using spell-check and grammar tools. AI can do more than those older tools, which means it needs more scrutiny—not an entirely different theory of authorship.
The recent cases aren’t all alike. The Financial Times added a clarification after AI was used to condense a longer draft of Harvard professor Ricardo Hausmann’s column. That mattered because the use conflicted with the paper’s policy. UC Berkeley mathematics professor Zvezdelina Stankova said she used AI to help edit an op-ed, while describing substantial human work on its drafts. Neither process can be established merely by assigning the finished prose an “AI percentage.” (FT clarification; The Daily Californian)
The Dartmouth case raises a harder question. Provost Santiago Schnell says the ideas and drafts were his, and that he used AI for language and clarity. He has since apologized for not routinely disclosing that use, and Dartmouth has announced a review. Because his Washington Post column concerned what students can accomplish independently of AI, students have good reason to ask how his own practices fit the standards he advocates. But that question deserves an examination of what he did and what rules applied—not a detector score treated as a verdict. (The Dartmouth; Schnell’s response)
I use AI to organize paragraphs and find better words for my ideas. I would be comfortable if a reader used it, too—to turn this post into a one-line summary, translate it, or explain it in a style they prefer. I care about whether the idea reaches them. I would care if a summary changed my meaning, just as I would care if an editor changed my argument.
That is where the photo analogy lands for me. We do not need every pixel—or every sentence—to be untouched. We need the person and the publisher behind the finished work to ensure that it represents what they mean, follows the rules they agreed to, and does not distort the truth.
Leave a Comment