During COVID-19, one of my more vivid memories was designing the photo gallery of this website. At 3:00 AM, I was planted in my chair and hunched over an ancient MacBook. My entire house had been asleep for at least five hours by that point, and they believed I was asleep as well. I sat in that chair for so long that my lower back ached, and a weird adhesion developed between my legs and the chair.
Yet, for whatever reason, the interface design enamored me. I used a third-party program called Nicepage to create a photo gallery showcasing three years of photography. This collage featured images from across Southern California: deployed national guardsmen during the Black Lives Matter protests, masked politicians at the onset of the pandemic, and still photographs of fishermen in downtown Santa Monica.
I tried to emulate the photojournalism seen in the New York Times's annual "Year in Photos" and split the pictures into large, bold headlines to give it a more newspaper feel.
In my experience working with other people, this entire process is uniquely human. To sit there for hours on end, moving text boxes a centimeter to the left, or to labor over minute syntax, requires an almost fantastical level of devotion unknown to AI. Anybody who has ever written anything knows that their writing tendencies came from years of playing with sentence structures or leveling their tone. Changed by circumstance and practice, writing is distinctively evolutionary.
That is why AI writing feels so hollow.
The large language model didn't go through anything that materializes real writing. Mark Twain's time as a steamboat pilot in Missouri, in part, inspired him to write The Adventures of Huckleberry Finn. Ernest Hemingway was a reporter during the Spanish Civil War and later based For Whom the Bell Tolls off of his journalism.
We all know this. It's why we all have a visceral reaction when a company releases an AI-slop-filled commercial or why we dismiss an essay riddled with AI patterns. It's uncreative and boring. It feels like SGA at the free-throw line.
And yet, despite our aversion, AI is slowly encroaching on everything. Disgruntled, burned-out, or just lazy students are relying on it to write their ten-page essay on how property taxes fund education. Professors use it for lesson planning, while researchers rely on its flimsy authority as a baseline for axioms. Law firms and tech companies use it for their menial tasks.
Generally, AI's allure lies in its expediency.
This is especially true at underfunded local newspapers, which increasingly rely on AI to synthesize large bodies of information into digestible, attractive language. The Wall Street Journal recently reported that small news outlets are relying on AI for reporting. Instead of attending city council meetings or budget hearings, some reporters submit the published transcript to AI for a summary.
More shockingly, the Journal revealed that 60,000 articles published by USA Today’s local U.K. papers utilized AI for first drafts. Here in the US, Axios — which does genuine investigative reporting — partnered with the ever-expanding OpenAI. The multi-year partnership involves funding AI-assisted journalism in 34 markets across the country.
And honestly, with proper guardrails, this doesn’t seem that troubling. When I was an unpaid intern at my hometown’s newspaper, I saw the stark financial reality of local journalism. If AI saves time and costs, I don’t see a tremendous potential danger in using AI to summarize a school board meeting. That is,again, assuming there are proper guardrails on how AI processes this information and whether it consolidates it for sellable data.
For some time, AI has also had a role in creative writing and opinion journalism. Some editorial boards I know use ChatGPT to generate feedback on how to sharpen their voices to accommodate their audiences better. Especially for local newspapers, editorial boards calibrate their tone and political positions to avoid losing their core readership. And instead of asking those core readers for their input, they rely on auto-generated prompts as reference points.
In my experience, I relied on AI to analyze some of my draft opinion columns in the Berkeley newspaper. I never used it to replace my writing, but for suggestions on where to trim unnecessary paragraphs or for a general analysis of the writing quality. For a while, I believed that was permissible: that while AI should never replace opinion writing, we can use it as a secondary tool for evaluation.
Yet, this creates an echo chamber for feedback. These LLMs are supposed to accommodate and flatter the user. A group of Stanford researchers recently found that AI models “are highly sycophantic: they affirm users' actions 50% more than humans do, and they do so even in cases where user queries mention manipulation, deception, or other relational harms.” The study also found that users are more likely to trust and rely on sycophantic responses.
Conclusions from the Stanford study.
In the realm of opinion journalism, constructive feedback imparts one’s perspective on somebody else's writing. Critical, sometimes ruthless, editing improves not just the current article but also the writer’s approach. Everybody who begins as a reporter or editorial assistant in some writing capacity has to get their ego bruised.
When I drafted my first tweet for a congresswoman, the press secretary simply said, “It’s verbose, clunky, and not her voice. See previous posts for ideas.” I found that old draft and inserted it into ChatGPT, and it spat out, “Your instinct is good—commemorative + resolve—but trimming the lofty phrase will make the tweet much more effective.” This advice is still somewhat critical, but that initial compliment cushions it, and one certainly can misconstrue the severity of the error.
That’s not to say that people have to be antagonistic when editing; just that they impart their experience and perspective on that person’s writing. In an interview with Tucker Carlson, Sam Altman spoke of the “moral foundation of AI” and referenced AI’s model spec; that is, AI follows the “collective experience, knowledge, and learning of humanity.” I don’t think anyone can really define humanity's “collective experience,” let alone the execrable CEO of OpenAI.
Yet, the fact remains that AI is not a member of a political party, religion, or even a country. Its raw coding cannot supplant human judgment or predict how human reaction to writing. The problem is our trust in a program designed to subvert the painstaking process of editing.
Human ingenuity requires human feedback, and AI is nothing more than a regressive reaffirmation of the writer we hope to be.