On The Wall Street Journal's A.I.-Written Op-ed And the Headache It Is Giving Me
Why You Shouldn't Use A.I. to Write A Professional Op-Ed, by Me, a Professional Op-Ed Writer Who Has Less Real Competition If You Do
[Stanley Druckenmiller's WSJ op-ed, which he apparently wrote with A.I.--and Paul Gigot's defense of it--were making me a little crazy, so I banged out a Medium post about it and hit publish. Here it is: ]
Billionaire investor Stanley Druckenmiller made news this week after the Wall Street Journal published his op-ed critiquing Treasury Secretary Scott Bessent for buying back long-dated bonds in an attempt to affect price. The op-ed got more negative attention than it warranted, not because Druckenmiller’s argument was bad, but because he’d written it with A.I.
I write opinion columns as a contributing writer for The New York Times Opinion section, and my contributor contract expressly forbids the use of A.I. for any aspect of writing columns that are published by the Times. But not every publication I’ve written for has those restrictions and Wall Street Journal editor Paul Gigot defended Druckenmiller — or “Stan”, as Gigot refers to him — and his use of A.I., which I suppose means it doesn't matter to the WSJ:
Gigot: "Shock and outrage! How dare this 'billionaire,' as many put it, not write every word himself."
I think readers do have a reasonable expectation that people writing op-eds in the Wall Street Journal are responsible for their own words and viewpoints, and if Gigot — Paul, I’ll call him, because why not? — wants to make the point that they’re often assisted by professional writers, he disingenuously pretends that the objection is about the writer having any kind of editorial help. Obviously, that’s not the problem or the editing process itself, which entails humans making the writing of other humans more readable, would be objectionable.**
Paul, who seems somewhat enamored with the billionaire in question, and perhaps enamored with the idea that the very rich should not have to do something so crass as put their own words into a Google doc like some common dirty writer, misses the point.
Generative A.I. is not a human editor helping you sharpen your argument or think of an original way to say something. It is a hallucinating, cliché-ridden, original thinking destruction machine. It hallucinates because it cannot reason; it peppers writing with clichés and hackneyed conventions because it is trained heavily on mediocre and bad writing; and its “logic” hews toward consensus, which is not a recipe for innovative thinking. Inasmuch as using A.I. is like having an editor, it’s like having the world’s worst editor.
Druckenmiller’s piece gets his point across, but largely because what he’s saying is something that’s fairly obvious to anyone watching the Bessent buybacks: the U.S. bond market is extraordinarily liquid, as it always is, and buybacks do not fix the real reason why yields are so high, which is that Donald J. Trump has managed to increase our national debt to $40 trillion dollars. Druckenmiller is just expressing the long held consensus viewpoint of Wall Street that you can’t boss around the bond market, but in a LinkedIn-post-esque A.I. cadence.
It’s not an original point of view; it's only interesting to hear Druckenmiller saying it because he’s a former mentor of Bessent’s and a high profile investor. For [almost never] better or worse, people with power and this sort of standing are often allowed to publish in our country’s most powerful newspapers solely because they’re powerful, even if what they’re saying is banal, dumb, self-serving, or as in Druckenmiller’s case, obvious. It’s a holdover from the mid-19th century when opinion pages were almost exclusively alloted to CEOs, statesmen, and other “people of importance.” If you were not rich and powerful in some way, you’d never be published. (And certainly, the managing editor of the Wall Street Journal would not applaud your use of what is essentially a plagiarism machine.)
But putting aside the sometimes questionable dynamics of our nation’s most illustrious opinion pages, people do not want to read the arguments and “thinking” of a generative A.I. model. They want to know what Druckenmiller thinks (in theory) and not what an A.I. would conclude or argue.
The problem with using an A.I. to write is also that A.I.s are shallow “thinkers.” (I’m going to keep putting “think” in scare quotes because that’s not what they’re literally doing, but it’s shorthand that is fairly pervasive now.) The relationship between thinking and language is not incidental. Language is the workhorse of abstract thought, especially if you’re working with complex ideas. Outsourcing it to an A.I. is not just outsourcing the assembly of words on the page, but the architecture of an argument and the actual human thinking behind it.
I teach a handful of writing workshops, including one on opinion writing,++ and I routinely have to explain to participants when you’re building a case to a reader, there are subtle things that matter that an A.I. can’t organically do. It can’t register the likely emotional effect of the information you’re delivering, or where the reader might need to take a breath, or where they might be less tolerant of exposition. It can’t deploy reporting with sensitivity to sources and legal considerations. It cannot reason the way the large fatty organ full of nerves and blood vessels trapped inside your skull can.
Druckenmiller argues in his own defense that we use calculators all the time because it doesn’t make sense for us to think through every calculation, and that is true. But it’s a bad analogy, because A.I.s are not actually better at writing, and what they produce is not what a newspaper reader is looking for in an op-ed, which the opinion of a specific human.
I would also make the argument that A.I. writing is bad and cringey, but given the sheer amount of it on LinkedIn and elsewhere, I think there are a lot of people who really can’t tell bad writing from good, and for a certain kind of writing, that might not matter. If you’re banging out six posts a day about what your divorce taught you about B2B software sales or the one weird trick you can use to lose belly fat, nobody’s expecting Shakespeare from you. But if you want to publish an op-ed in a national outlet and you’re not rich and powerful, the quality of your writing matters quite a bit. This is partly because you’re competing with a lot of other submissions for limited space, including submissions from rich and powerful people who already have the advantage of being rich and powerful and do not need to be as articulate or careful in their argument. You’re also competing with in-house full time columnists and people who are on contract like me. They are all starting at mile 20 of the marathon, and you are just lacing up your shoes. So you have to be as good or better than they are.
If you’re going to write for a place that might put your argument in front of millions of people, it also behooves you to do your own writing yourself because readers will assume that you’re doing that anyway. If you leave them with the impression that you’re a shallow thinker whose expression of ideas is mediocre and unremarkable, it probably undermines whatever you expected to accomplish by having an op-ed in a big publication. The payoff for the shortcut isn’t worth it.
If not for the fact that Druckenmiller used A.I., in his column, I wouldn’t remember a word of it. But I do remember a JP Morgan executive’s analysis of the Bessent buybacks: It’s like “paying your mortgage with your credit card.” It’s a good analogy, original — and formulated by a human.
** Speaking of objectionable, another line from Gigot’s editorial: “But it wouldn’t be honest for me to grandstand that our contributors never use a tool like AI. Mr. Druckenmiller’s sin against the media pharisees was admitting its use.” Media pharisees? Sir. Your newspaper constantly publishes editorials blaming the left for rising anti-semitism and you write that? Maybe the anti-semitism is coming from inside the house?
++ I didn't write this as a pitch for my op-ed workshop, though I think the workshop is good for people who are looking to publish formally in a national outlet. I make about enough money from teaching workshops to pay the utility bills--and if I were being that utilitarian about everything, well, it wouldn't exactly behoove me to teach people how to compete with me in the NYT Opinion section. But I enjoy teaching and now that NYU's Studio 20 is on hiatus because my erstwhile boss Jay Rosen is retiring, I have to find other ways to do it. (If anyone out there works at an educational institution that's looking for adjuncts, send 'em my way.)
But even among people who take my workshop, I get a lot of first drafts that are A.I.-written and I have to explain why you shouldn't write with A.I. to people who actively want to write and publish--so much so that they're willing to pay to sit in front of Zoom for a couple of hours to learn how to do it. It wasn't intuitive to Druckenmiller why you wouldn't do it and he's not a dummy, so I think it's worth explaining.
Elizabeth Spiers is a digital media strategist and contributor to the New York Times Opinion section. She also co-hosts the Slate Money podcast with Felix Salmon and Emily Peck, and thinks Treasury should not be intervening in the bond market beyond its already scheduled regular-course-of-Treasury-business buybacks. There are probably some typos in this column because she generally needs some copyediting, but she is not going to use A.I. to do it, sorry. There are legitimate uses of A.I.s in newsrooms, but that's another column.
Thanks for reading! If you'd like to support my work, there are several ways to do it, and it's always appreciated. - Elizabeth