AI discourse is out of touch
While knowledge workers get AI to help them do their jobs, the rest of the world is operating on a whole other level.
Welcome to the The Closing Argument, our verdict on the news, plus everything The Argument published and appeared in this week.
The Verdict
I recently wrote about how the people who shape public debate, who I call the messenger class, share a narrow band of experiences that we mistake for the norm, producing a systematically skewed picture of reality. This class includes everyone from tech workers and CEOs to journalists and politicians, and my critique is less ideological than experiential.
The big problem isn’t that the messenger class is lying or purposefully hiding the ball, which in many ways would be a much simpler problem. The problem is that important stories go undercovered and trends overlooked because very few messenger class participants live in high-crime neighborhoods, rural America, or outside of high-income countries, to name a few distinguishing features.
AI discourse is a clean case study of this problem. There are, loosely speaking, two mainstream concerns with AI:
It’s going to take your job.
It’s going to kill education.
Unlike previous generations of workers impacted by technological revolutions, highly educated workers like me are the ones who find ourselves worried that a new machine could take our jobs. People who work with their hands can, for now, feel confident that robot roofers or day care workers are far away. It’s the computer scientists and, uh, journalists who are — and should be — worried.
Many people I speak to seem completely unaware that these are not the major use cases for AI right now. In fact, the most common use cases for AI seem practical and largely unobjectionable.
A new study, which examined 202,590 ChatGPT conversations from 1,252 users in India, Nigeria, Brazil, and Pakistan from Dec. 2022 to Feb. 2026, found that the majority of queries were about personal, not work, matters. “The most prevalent topics are health, education, translation, and everyday practical guidance,” the researchers reported.
In Pakistan, the most common use case was Urdu-English translation, and in Brazil, Portuguese-English translation was also very common. The health category — which was common across all four countries — is largely about helping users interpret symptoms, and obtain diet, mental health, and fitness advice.1
And there’s evidence that the messenger class uses AI differently:
According to a study from last fall2 that looked at roughly 1.1 million ChatGPT messages from May 2024 to June 2025, work usage was more common for educated users in highly paid professional occupations, but 70% of all usage was for non-work purposes.
The researchers broke that into three buckets:
Practical guidance: Health/fitness/beauty/self-care, tutoring and teaching, how-to advice, and “creative ideation.”
Seeking information: basically a substitute for googling. People are asking for recipes, information on specific people, products, current events, etc.
Writing: Help producing emails as well as “editing, critiquing, summarizing, and translating” existing text.
If this suggestive evidence is right and AI is largely a general-purpose digital advisor for everything from “What is this rash on my leg?” to “What’s a healthy, cheap meal I can make for my family in under 30 minutes?” then a lot about how we think about AI should change.
The most important question becomes: Is the advice people are getting any good? What is the economic impact of all this free advice?
That being said, if someone can go to a chatbot for legal advice, for medical advice, for tutoring, and for many more use cases that many of us went to college to become experts in, that does directly threaten the messenger class’s economic position.
Top stories this week
As we grow, I want to make sure you see everything we’re doing without flooding your inbox with dozens of emails. But for the real libs, you can get every post as it drops by opting into The Mag here.
Education research is weak and sloppy. Why?
Most social sciences have improved their research methods since the 20th century. But not the field of education, and it shows.
Helen DeWitt is the psycho we need
Acclaimed novelist Helen DeWitt recently lost $175,000 because she couldn’t find Wi-Fi in Amsterdam. Is she the unplugged hero we've all been waiting for?
🌟Abundance Wins of the Week🌟
Worth watching...
The second episode of Matt and Jerusalem’s podcast released this Thursday. They bring back a debate as old as the internet: online anonymity. Subscribe wherever you get your podcasts and watch below!
Subscribe: Apple Podcasts | Spotify | YouTube | Overcast | Pocket Casts
Our director of political data, Lakshya Jain, was joined by VoteHub’s Head of Data Science Zachary Donnini to talk about the state of the California governor’s race after credible allegations of sexual assault led Democratic frontrunner Rep. Eric Swalwell to drop out of the race:
The Sidebar by Maibritt Henkel
This week, two pieces prompted discussion of a recurring political question: When feelings don’t fit the fact pattern, do we reconsider the facts or pathologize the complainers?
On Tuesday, researcher Paddy Maher challenged popular narratives about the loneliness crisis. It is not old men who are the loneliest, but young people — especially poor young people.
A few days later, I published a piece about who is experiencing the most financial distress right now. Based on a bunch of data, including our original polling here at The Argument, I argued that women disproportionately feel the weight of rising living costs.
Both of our pieces came down firmly on the side of: reconsider the facts. But some commenters felt differently:
In response, one subscriber remarked that despite women appearing to do better than men on many metrics (number of friends, economic prospects etc.), The Argument had made the point twice this week that women continue to feel worse than men. “Strange things, feelings,” she mused in note.
“By far the most likely explanation for the gender gap is the one you’re dismissing: women tend to be more neurotic and anxious than men,” commented another.
Now, both my article and Paddy’s point to material reasons for people’s poor perceptions of the economy. Paddy’s argument is that the much-hyped “male loneliness epidemic” is largely an artifact of an unreliable dataset and that, once you use better data, the loneliness signal tracks financial insecurity rather than gender.
The bulk of my own argument was also about women’s real-world financial conditions; the fact that they are more likely to work low-wage jobs, raise children on a single income, and shop for groceries (putting them on the front line of inflation).
Feelings aren’t noise. They’re often a signal that our map just doesn’t match the territory, that our model is missing something important — like jobs numbers not capturing how frustrating the hiring process has gotten.
The mistake isn’t taking feelings seriously. It’s taking them seriously only when they flatter a story we already wanted to tell.
What’s News with The Argument
The Argument recommends
The team really pulled through with their cultural recs this week.
I personally have a longstanding fear of board games, but sharing an office with game enthusiast, Eli Richman, has made me realize I should probably work through that. He recommends Spirit Island, “a great cooperative board game that sees players taking the role of primordial spirits defending an island of native inhabitants.”
Apparently it’s a bit like the game Pandemic, “except the disease you’re fighting is European colonialists.” Say less, Eli.
Justin Zuckerman, our video producer, turns out to be a more diligent cultural critic than me. Instead of just reading buzzy internet takes, he actually went and saw The Drama in person. Verdict is in, folks: “dark, funny, shocking. Really good!”
Angela Tracy, who has had a “nostalgic 2010s-ish Bollywood-themed” week, recommends the 2014 religious satire PK, about an alien stranded on earth. Intriguing.
Now for some music:
Tech Fellow Kobe Yank-Jacobs described a you-had-to-be-there kind of live performance he went to by the artist Tōth, who apparently “plays trumpet and guitar while his one bandmate plays bass and drums.” I was not at this concert but have since listened to a few Tōth tracks and approve.
My other fellow fellow, Milan Singh, takes us to a very different corner of the musicverse: U.K. rap. In particular, he recommends Dave’s 2025 album, The Boy Who Played the Harp. I strongly second this one.
From me, you get a recommendation in three parts to it. First, watch the epic “Sandstorm” music video by Finnish DJ Darude from 1999. Second, marvel at this footage of the annual Sandstorm Run, a seven-kilometer race through Helsinki that passes all the landmarks featured in the original music video. Third, sign up for the 2026 race in August (I’ll see you there).
Finally, although Jerusalem Demsas did not come to me with any explicit recommendations, anyone who has been within one meter of her this week knows that she has Helen Dewitt on the brain. So, I guess go read The Last Samurai.
-Maibritt Henkel
More to read:
Can you tinker your way out of the permanent underclass?
AI has people terrified about the future of employment. And that terror has naturally begotten advice columns about how to personally come out ahead.
Who are America's biggest antisemites?
POLLING: Antisemitism, however you measure or define it, is on the rise.
This paper is not a random sample; instead the researchers paid gig economy workers to export their chat histories for money.
Yes, in AI terms, this is like 100 years ago. But nearly 10% of the global adult population had used ChatGPT by this point.
I think it's also relevant that, for professional writers who are trying to minimize their use of AI, their interactions with it are mostly going to unavoidable enounters integrated in legacy applications (Google search, MS Office suite, etc.), which tend to be annoyingly obtrusive and relatively low-quality as models go. That, and AI slop in advertisements. They make for a lousy impression.
"There are, loosely speaking, two mainstream concerns with AI:
1. It’s going to take your job.
2. It’s going to kill education."
I'm not so sure! There's a Pew poll from last September where only 9% of US adults are concerned about job loss and education isn't even on the list. The top two concerns were "erodes human abilities and connections" at 27% and "negative impact on accuracy of information" at 18%. Things could have changed since then, but probably not too much. I find all these vastly different conceptions from person to person to be quite interesting.