At the tail end of a recent episode of The Grill Room, Dylan Byers and Julia Alexander discuss how each of them uses A.I. Julia is the vibe coder, Puck News’s heaviest Claude user, creating personalized apps and processes to help her do her job while employing strict guardrails that protect her thinking, voice, and authorship. She thinks deeply about how she would and should disclose how A.I. informs any article, analysis, or research she puts out, suggesting that maybe 5-10% of it could be credited to artificial intelligence. And all of that would be in the research phase and never in the writing.
Byers isn’t a power user. He rotates through Claude, ChatGPT, and Gemini primarily as an answer engine. “Google 3.0,” as Julia puts it. While scrolling through his chat history live, he reflects on how he sneaks away after his eight-year-old son asks him a question he doesn’t know the answer to and prompts his way to one. Julia then asks, jokingly, if he takes credit for the answer ChatGPT gave him.
He doesn’t reply.
The main topic of this episode of their podcast was the fallout from billionaire investor Stan Druckenmiller’s op-ed in the Wall Street Journal, which he and Paul Gigot, section editor of WSJ Opinion, defended after it was identified as 100% A.I.-written when run through the A.I. content detector, Pangram*. These detectors are prone to over-detecting, so we don’t know how much of the piece was written by machine, but Byers doesn’t think readers care as much about this as professional writers do. While he agrees that there should be disclosure policies and that audiences will want to know when an agent wrote something, he thinks much of what gets published these days is recycled consensus or “human slop,” and he doesn’t think that’s materially different than slop made by machines.
How cynical.
The Journal can’t even agree on this in their own newsroom. James Taranto, their editorial features editor, wrote that professional writers have an ethical, and sometimes contractual, duty to keep their work original (while also noting the errors produced by Pangram). Perhaps Druckenmiller doesn’t fall under that standard, but it shouldn’t be the burden of the reader to figure that out when going from piece to piece under the same masthead. I’m of the opinion that if you do not either A) take a hard line on the usage of A.I. in written work—particularly that which comes from prestigious and trusted journalistic publications—or B) explain how it was used or is allowed to be used in the creation of articles or other material that comes under your banner, then you are breaking the foundation of your relationship with your readers, watchers, listeners, and subscribers. Byers’s argument is that it is already happening and the shift can’t be stopped. Journalists trying to “stand athwart history” will soon sound crazy.
Whether or not his prediction is accurate, the willingness to concede, especially by someone working in media, is crazy-making to me. A.I. is increasingly unpopular with regular people. The idea that everyone is going to just acquiesce is the fantasy of Silicon Valley, the investor class, and Fortune 500 C-Suites who have bet it all on A.I. doing everything.
The Atlantic’s Charlie Warzel rejects this “the toothpaste is already out of the tube” framing outright. On his podcast, Galaxy Brain, he brings together Elizabeth Spiers and Teddy Brown to unpack the true value of writing as a defense against this push to hit the A.I. easy button.
They acknowledge that A.I. can do a lot of writing. As CNN reporter T.M. Brown notes, there is much utilitarian writing that was previously done by humans that can be done effectively by LLMs: white papers, product descriptions, SEO copy, corporate speak, LinkedIn posts. Perhaps we don’t lose much art and craft if a model does this stuff rather than a person. Unless, of course, you value the person(s) whose career and work experience would allow them to become a much better writer later.
I’d expand that to say there is even some basic journalism that doesn’t lose much if it’s mostly software-generated, from police and court dockets to weather reports and baseline sports recaps. That said, I prefer what I consume about the things I care about to be filed by humans who cover a topic religiously. In sports, that means I want the human beat reporter who’s with a team from practice-to-practice, game-to-game, season-to-season, and understands the people involved as much as the box score. Without that, we don’t get this piece by Rey Moralde on Rae Burrell.
Writing, or whatever your expressive medium, is how we convey our interiority to the world. It is our intellectual costume, as Brown puts it. Your heart and soul on display. Even if you accept Byers’s argument that original thought is somewhat rare, I’d still rather read derivative thinking from a real person’s own mind than plagiarized thinking from the open web, generated by ChatGPT via a prompt. Even the inelegant output of a less artful writer may have meaning. And, at the very least, I would know their argument or insight was honestly earned and worth my time. I may disagree vehemently with their stance, but I’d know I was debating their ideas, not those of a collectively plagiarized hivemind.
Why would I give my time to a machine’s backward-looking, unoriginal, overly long, and ultimately empty pieces about a human condition it can’t possibly understand when I could be in awe of Nikole Hannah-Jones’s raw and deft words? Or Roxane Gay’s? Or the many, many others who put out real human work worthy of grappling with.
Not revealing that you have used a bot to help you come up with an answer breaks the trust you have with your audience, whether that’s your kid or your reader.
I’m by no means taking a shot at Byers’ parenting, but I wonder if Julia’s gentle jab about credit got him to reconsider his stance a little.
I’m a dinosaur, but when I was a precocious elementary student with endless questions Mom and Dad couldn’t answer, my parents sent me to an encyclopedia, a reference book, or the library. If I wanted to learn something that couldn’t be found in one of those places, the best teacher was someone with that knowledge and the ability to convey it to me specifically. These tactics exposed me to the craft’s sources and methods and encouraged me to find my own solutions.
I can’t imagine my parents or teachers would take credit for someone (or some thing) else’s work as their own and definitely would not do so with pride. And neither would I. Not then and not now. To do so would feel duplicitous and hollow.
As I have written before, I’m no luddite. I’m not an absolutist or doomer taking a hard line against all A.I. use, but I do believe we have a responsibility to approach it as ethically as possible. Julia Alexander proposed a kind of gauge on columns that would show how much of the work could be credited to artificial intelligence. For my own work, I’m thinking about a kind of digital colophon, or something similar, that would outline how A.I. was used to create what you’ve consumed. It could, of course, expand to include other non-A.I. methods, but I think the specific challenge today is that people assume, expect, and want something human-made, so the appropriate and honest thing to do is reveal what role software played. That is not true when something leans heavily toward being A.I.-generated. I feel duped when I find out something was “made by A.I.”
Writing is hard. A.I. gives the illusion that it is easy because it generates many words quickly. In an age when technologists try to convince us that all our parts should be commoditized and human interiority should be given over freely to machines, we should be precious with our thoughts and the difficult process of turning them into something for others to consume.
Write. Wear your intellectual costume with pride. Don’t let some artificial intelligence engage in intellectual cosplay on your behalf.
Colophon
- My spellcheck tool has incorporated A.I. features.
- I use Claude to generate blog metadata (tags, SEO copy, etc.) based on my writing style guide and blog history.
- I have a personal knowledge base that I query with Claude to recall research notes or previous thinking on a subject.
- I write all my own copy.
*If you run this piece through Pangram, it will give it a 100% Human-Written score 💅🏾.
