The Situation of AI Causing Mass Unemployment Might Not Be That Severe Yet

@Lanxi: Times writer Ezra Klein wrote a column saying that based on his interview records, the situation of AI causing mass unemployment might not be that severe yet. The bad news is that the severe problems lie elsewhere:

A March poll showed that 70% of Americans believe AI will reduce their job opportunities, up from 56% a year ago, and 30% of respondents worry about imminent job loss.

What's so surprising about this? The bosses of AI companies have been frequently issuing warnings about the imminent collapse of the labor market:

– The founder of Anthropic stated that within 5 years at the earliest, half of entry-level white-collar jobs would permanently disappear;

– Microsoft's CEO claimed that most clerical jobs will be completely taken over by AI within 18 months;

– OpenAI released a document calling for a 32-hour workweek to mitigate mass unemployment by reducing human productivity;

– Right outside the NYT building, an AI company I've never even heard of booked a massive billboard that read, 'Stop Hiring Humans';

Well, thank you very much.

If you are deeply convinced by this round of AI narrative, then undoubtedly surrender is the only option.

AI is designed to cheaply mimic all the behaviors humans can do on a computer, but it never needs to sleep, has no intention of forming a union, and indeed performs better than real humans on many tasks.

So of course companies are eager to replace humans with machines, and they are already doing so. Meta, Oracle, and Block are all trying to buy out employees' tenure, openly citing AI as the reason.

However, it's always right to be cautious. These tech companies might just be at the end of a hiring cycle, while also wanting to tell a story to the capital market that stimulates investors' G-spots. The bigwigs in the AI industry are indeed experts in neural networks, but they may not be experts in the labor market.

First of all, macroeconomic data is not cooperating with the AI industry. The unemployment rate in March 2026 was 4.3%, compared to 4.2% in the same period last year, and average hourly wages also remained stable.

Claude Code is awesome, but the demand for software engineer positions is still rising, and the reason shouldn't be hard to understand.

Mainstream economists are also skeptical of the doomsday prophecies of mass unemployment. Alex Imas, an economist at the University of Chicago, believes that the vast majority of discussions about AI miss the point: 'The answer always depends on what becomes scarce.'

– From human history, calories were scarce for a long time, and our skills were all geared toward finding food;

– The development of agriculture gradually solved the food shortage problem, and then it was goods' turn to become scarce;

– Industrial manufacturing brought a massive supply of affordable goods, and then technology became scarce. Doctors, lawyers, and engineers earned high salaries because they possessed knowledge;

– Now we worry that AI will make knowledge worthless too, just like clothes and strawberries are consumable by everyone today, the fruits of learning have also become consumer goods...

But scarcity itself is relative. The tasks AI can accomplish have been discussed too much; what is more noteworthy is which tasks people do not want AI to do.

Here is a finding from econometrics: the wealthier humans are, the more they demand from other humans, not less.

Therefore, goods and services that are more human, experiential, or meaningful will definitely be scarce. Clothes with a backstory, food with a distinct style, a doctor you can meet face-to-face, a physical therapist who makes you feel understood, a tutor who knows your child... these social relationships will see explosive growth.

Yes, AI has stripped away the scenarios where humans work with computers, and humans have been driven to work with other humans.

It turns out that the more automated things are, the more we value human warmth. Take coffee, for example. In the past, making espresso at home was a hassle; now, everyone has a coffee machine.

So did this cause coffee shops to go out of business? Of course not. There are more baristas than ever before, and more coffee shops than ever before. Coffee as a commodity has actually spawned more demand for the coffee experience.

This is the value bestowed by scarcity.

Alright, this is Alex Imas's conclusion. Under AI's ruthless disruption, human jobs will enter an era where being human-centric is the standard. But there is another possibility: human labor might not undergo a massive change either.

In 1979, the first spreadsheet software, VisiCalc, was released on the Apple II. It could do in minutes what used to take an entire team of accountants days.

At that time, some predicted the accounting profession was doomed. However, over the next 40 years, the number of accountants actually grew more than fourfold.

Eldar Maksymov, an accounting professor at Arizona State University, believes: 'Spreadsheet software unearthed the latent financial needs in the market. These needs weren't discovered before simply because the costs hadn't dropped low enough.'

This is a school of thought based on the 'Jevons Paradox.' In 1865, British economist Jevons discovered that coal consumption did not decrease due to the steam engine's improvement in productivity. On the contrary, because it became cheaper, the application of coal became more widespread.

Eldar Maksymov believes in historical precedents: 'Among occupational groups that extensively adopted computer technology, employment growth far outpaced those that didn't. The drop in costs corresponds to an increase in demand, ultimately driving the expansion of employment scale.'

Simply put, the enhancement of capabilities makes humans realize there is much more to do.

My own experience seems to validate this point. When I started doing podcasts 10 years ago, I was the only researcher on the team. Now I manage an entire team running the show. And has this made my job easier?

Not at all. The effort I put into preparation has become heavier because the incremental information brought by the team has multiplied the time I need to digest and think, and my podcast has also grown bigger.

All the people I know who are keen on embracing AI are now busier than ever because they can do more things. Wasn't AI supposed to free everyone from work to enjoy life?

Of course, as to whether AI has actually boosted productivity or created an illusion of boosting productivity, the conclusion varies from person to person:

– Slowly chewing through a difficult book is far better than quickly absorbing the summaries of ten books;

– Earnestly and thoroughly writing a first draft also sparks more ideas than editing five AI-generated outlines;

What I mean is, the sense of efficiency itself is something to be wary of. I've seen plenty of people who hand all the work over to the 'lobster' (AI), but honestly, their work quality has declined.

But allow me to digress. Wharton School professor Ethan Mollick once proposed his benchmark test for AI: Is it better than the most suitable human you can find at the moment?

In his view, the question is not whether AI outperforms top-tier editors, programmers, therapists, or travel consultants, but whether it is better than the best person you can contact when you urgently need help.

I thought about it, and over the past year, I have indeed found AI to be better than the people around me:

– My editor is very knowledgeable, but he needs rest and has to divide his time among other writers;

– My massage therapist is also fantastic, but generally she can only see me once a week;

– I can also find a specialized doctor, but making an appointment is a hassle...

So perhaps I have reached the repeatedly warned critical point—AI is starting to have the ability to replace the human roles in my life.

But that's not the case. The more powerful AI becomes, the more I need to communicate with the people around me:

– AI suggested I should take a certain physical symptom seriously, so I went to the doctor, only to find it was just a common allergy;

– AI made sharp comments about my personal dilemma, which then prompted me to start a new conversation with my therapist;

– AI helped me verify a research idea, which became a new topic for discussion with my editor;

– AI made video production easier, so I can have more demands when communicating with my video editor...

How should I put it, although I don't believe that a fully automated economy and a wave of mass unemployment are inevitable, I can't completely rule out that probability. AI is a technology vastly different from the past; its flexibility and growth potential have helped it transcend the category of a mere tool.

A more likely scenario is that AI won't replace all or most jobs, but only a portion of them. Strangely, it is precisely this possibility that leaves us underprepared.

A world where AI replaces 8 million workers is harder to cope with than one where it replaces 80 million. If a true 'the big one' event does happen, a comprehensive restructuring of the economy might actually offer an opportunity for 'destruction before construction.'

The pandemic is an example: that shock was so thorough that society could no longer blame workers' misfortunes on themselves as before, so it had to establish an unprecedented subsidy system to secure millions of people.

But when the scope of unemployment is smaller, we are actually more cruel. The US lost about 2 million jobs due to globalization, which is not a huge number in the overall job market, but for those 2 million specific households, it was a devastating blow.

If all the truck drivers or marketing managers in the country lost their jobs, we would take action quickly. However, if only the unemployment rate for truck drivers or marketing managers triples, we will just do what we've always done: imply it's their own fault, give them a few months of unemployment insurance and training opportunities, and then continue to ignore this structural predicament.

Another reality is that even if AI makes skills rooted in deep relationships more valuable, it will simultaneously weaken humans' ability to learn these skills.

The time young people spend with friends has already dropped from 12 hours/week in 2003 to 5 hours/week in 2024. The proportion of high schoolers who have ever dated also plummeted from 80% in 2000 to 46% in 2024, and about 1/4 of Gen Zers had no sexual activity in the past year.

AI might be an accomplice to this social disintegration; it provides a digital simulation of relationships, removing the motivation to experience the joys and pains of real relationships.

If Alex Imas is right—and I believe he is—then our ability to form deep relationships with others will become a core, high-value personal asset. What I worry about is that this is exactly what technology is destroying in young people.

When I optimistically look forward to the future created by the AI era, the picture is one of abundance, where humans are encouraged to live a life following their hearts. But when that optimism is shattered, the world remains as it is, except that wealth is monopolized, and the deep relationships we value have long since forgotten how to be maintained.