Permanently at the Bottom, No One Can Answer That 17-Year-Old Child

Note: This article is speculative; please bear with it to the end

By: Xiao Bing

On July 26, Sam Altman stood on the stage of the Chase Center in San Francisco and said: "The notion that 'if you don't get into a frontier lab, you'll fall into the permanent underclass' is utterly stupid".

"Permanent underclass," this buzzword has come into view again.

It gradually gained popularity on X in the tech circle during the summer of 2025. Its theoretical skeleton comes from The Curse of Intelligence, as well as Trammell and Patel's paper Capital in the Twenty-Second Century.

This claim is not just predicting that "AI will cause many people to lose their jobs"; it describes a more extreme future: when AI can complete the vast majority of cognitive and physical labor, enterprises creating wealth will no longer need to employ so many people. The importance of wages in the economy will decline accordingly, while profits will increasingly flow to those who control models, compute power, energy, and data centers.

It is not just that a group of people temporarily become poor, but that the labor they rely on for upward mobility loses value; it is not just that the rich are richer than before, but that people without capital cannot become capital owners through work. This is the so-called "permanent underclass."

Without large-scale wealth redistribution, public ownership, or a global progressive capital tax, this theory ultimately leads to an unsettling conclusion:

Nearly all assets may gradually flow to the group of people who were already the wealthiest when the AI transformation occurred.

Of course, this is just a deduction, but it reminds me of a small matter.

Over the past year, writer Jasmine Sun conducted a nearly clumsy field survey. She met with researchers from frontier AI labs one by one, talking for an hour each time, without recording or keeping records, questioning them like an anthropologist about their true views on the future of AI.

There was one question she asked almost every time:

Suppose an ordinary 17-year-old American child is standing in front of you. He is not a genius programmer, his grades are only Bs, and he doesn't care much about AI usually. How would you advise him to prepare for the future?

Almost no one could answer.

These researchers had various political stances and their judgments on AI ranged from optimistic to pessimistic, yet their answers were surprisingly consistent:

Don't know. The situation now is scary; there are probably not many jobs left for him, and he happens to be trapped in the most painful transition period...

But the trouble that 17-year-old child truly faces might not be whether he will become poor.

What Is Permanent Is Not Poverty

Every era has poor people and an underclass, but the "underclass" of the past was not necessarily a permanent identity.

A person could sell labor for wages, then exchange wages for education, housing, and assets; workers could also organize, strike, and collectively bargain, demanding capital convert a portion of productivity growth into higher wages and better benefits.

This mechanism was not fair, but it at least preserved a channel for upward mobility.

What truly stings about the "permanent underclass" is that it suggests this channel itself might disappear.

Looking back at labor history, unions, minimum wages, and the weekend system were all built on the same fact: Capital and labor cannot completely substitute for each other.

Factory owners needed workers, and tech companies needed engineers, so both sides had to sit down and negotiate.

Marx predicted the high concentration of capital and also provided a set of solutions. Because in his world, workers still held something capital needed but could not create out of thin air: labor.

Strikes were effective because workers could temporarily withhold it. If workers didn't enter the factory, machines couldn't run; if engineers didn't write code, products couldn't launch.

But if compute power can buy all the labor a person can provide, the things workers can withhold will approach zero.

At that time, the problem is not just that some people's wages decline, but that capital no longer needs to negotiate with the majority. Workers will find it difficult to accumulate assets through wages, and also difficult to demand redistribution of returns by stopping work.

What makes the underclass "permanent" is that the machine that once could send the poor back to the middle class—labor, bargaining, and asset accumulation—has been dismantled.

Evidence that this machine is loosening has moved from theory into payrolls.

The Stanford Digital Economy Lab used ADP microdata covering 4.6 million workers and found: After the popularization of generative AI, employment among workers aged 22 to 25 in occupations with the highest AI exposure saw a relative decline of about 16%; the unemployment rate for recent college graduates in the US reached 5.6%, up 1.6 percentage points from three years ago; the scale of recent graduate recruitment by large tech companies dropped by 25% within two years.

The elevator is still operating normally; it's just that the button for the first floor has been removed.

Those Who Cannot Answer Are Buying Insurance for Themselves

Back to Sun's interviews.

Those researchers who could not answer "what should the 17-year-old child do" did not stop their work because of this.

Sun continued to ask: Since you believe the future is so dangerous, why continue building it?

The answers roughly fell into three categories.

The first category sincerely believes that as long as they endure the transition period, AI can ultimately cure diseases and eliminate scarcity.

The second category believes in technological determinism: even if they don't do it, others will.

The third category is the most honest, and also the most awkward:

If the great turmoil really comes, at least secure a position for yourself in the future first.

This impulse to "secure a position" is changing the talent flow across the entire AI industry.

A friend of Sun's pursuing an AI PhD at Berkeley said that about half of their cohort decided to graduate early, and some even chose their PhD thesis topics based on "which research direction makes it easiest to get an offer from an AI lab."

Many independent writers and policy researchers around Sun also gave up their original positions and joined AI labs.

These stories sound like jokes, yet they are truly deciding the life choices of a generation, and continuously draining talent from the independent research and public policy ecosystem.

And those being drained are precisely the group most likely to build the negotiation table.

Factory automation in the 20th century did not generally evolve into intense conflict. An important reason was that before machines entered the factory, the factory side usually had to negotiate with the union first: automation had to be explained as a safety upgrade, and wage increases needed to be tied to productivity improvements.

Between capital and labor, there was a negotiation table.

Today's white-collar workers do not have this table.

More subtly, those most capable of building the negotiation table—scholars, independent researchers, and policy talent—are being persuaded into labs by the narrative that "labor is about to lose value." Because only by entering the lab can they obtain equity and stand on the side of capital in advance.

Thus, a closed loop is formed:

The more people believe that labor's bargaining power is about to disappear, the more people will give up building bargaining power and turn to competing for equity; and the fewer builders there are, the faster labor's bargaining power disappears.

No One Is Absolutely Safe

That 17-year-old child is not entirely without advice.

Facing the uncertainty brought by AI, people have already given many answers, most directly spawning two reactions.

One is to learn AI as soon as possible and strive to be the last person replaced; the other is to own AI assets as soon as possible and strive to stand on the side of capital before labor loses value.

The latter impulse is especially obvious in Silicon Valley, and even among the wealthy class in China.

Some people will pay any price to hope to get equity in Anthropic, OpenAI, or other frontier AI companies. The theory is simple: If the singularity really arrives, labor may depreciate rapidly; at that time, what determines a person's situation will no longer be what they can do, but what they own in advance.

According to this logic, as long as labor income is exchanged for equity in AI companies before the window closes, there is a chance to transform from someone replaced by technology to someone who owns technology.

This is also the most tempting part of the "permanent underclass" narrative; it not only creates fear but also hints at an escape route:

Since capital may replace labor, then become a capital owner from a laborer as soon as possible.

The problem is that this road does not belong to the majority in the first place.

OpenAI and Anthropic are not companies that ordinary people can easily buy on the public market. Those who truly have a chance to obtain this equity are usually lab employees, early investors, and wealthy individuals who can access the private market.

Therefore, this is almost a circular argument:

To avoid falling into the underclass because of lacking capital, he needs to first possess capital that only the upper class can obtain.

More importantly, even if a person successfully obtains equity, this insurance is not necessarily permanently valid.

Fernando Borretti, author of the programming language Austral, pointed out that there is actually a contradiction within the "permanent underclass" theory.

If AI can truly complete almost all cognitive and physical labor at a lower cost, then while ordinary workers may indeed lose economic value, those who hold shares in AI companies in advance may not necessarily become the "permanent upper class."

The reason is simple: wealth is not written into the laws of nature.

The reason a person owns a company, land, or compute power is not just because their name is on the contract, but because courts, police, and governments are willing to recognize and protect this property right.

But continuing the deduction according to the setting of "super AI replaces everything," AI may ultimately be able to complete even production, management, governance, and even war. At that point, today's rich provide neither labor nor hold actual power beyond machines. So why must future states or super intelligences forever recognize the ownership they acquired in the old era?

Buying equity in AI companies may allow a person to survive the transition period, but it cannot guarantee they remain in the upper class permanently.

If owning AI assets is not a path everyone can take, the more common answer is to learn AI.

Another paradox has emerged on this path:

The better a person is at using AI to improve productivity, the more likely they are to help enterprises reduce demand for other workers.

He may be able to stay in the elevator temporarily, but he is also participating in removing the buttons for other floors.

This does not mean people should not learn AI. For individuals, skillfully using AI may still be the most reasonable choice at present. The problem is that when everyone tries to avoid being replaced by improving their own substitutable efficiency, the final result may be that enterprises need fewer workers.

Individual rational choices, when aggregated, may instead accelerate the decline of labor's overall bargaining power.

There is another answer: leave the fields most easily replicated by AI and turn to work that requires physical presence.

If the supply of digital products approaches infinity, then physical operations, on-site responsibility, and interpersonal trust that cannot be replicated remotely may indeed command a higher premium.

Data centers need electricians, an aging society needs caregivers, and people are still willing to pay for the real presence of doctors, teachers, and service personnel, but this is also not a retreat path everyone can take.

The above look like three different paths, but actually all answer the same question:

How to make oneself fall down a little later?

But the answer that 17-year-old child truly needs might be: Why must a person's survival depend on whether they are still needed by capital?

The reason researchers cannot answer is that this question ultimately requires not a personal plan, but a new distribution system.

When labor was still irreplaceable, wages, unions, and strikes together constituted the distribution mechanism; if labor really begins to lose scarcity, then society must find another way to distribute the wealth created by machines to those who are no longer needed by machines.

This matter cannot be solved by everyone learning AI harder, nor can it be accomplished by everyone buying a few stocks in advance.

What individuals can buy is only buffer time; what is truly missing is still that negotiation table.

A Negotiation Table

Building a negotiation table is not something any individual effort can substitute.

China unexpectedly provides a control sample.

In December 2025, a court in Beijing ruled that a position being replaced by AI cannot directly constitute a legal reason for dismissal. There were also state-owned enterprise employees who said that the AI tools they used could complete the work of about two employees, but the company promised not to lay off staff on the grounds of AI.

After staying in China for two weeks, Sun summarized that the social attitude there is not simple technological optimism, but a pragmatism of "resistance is unrealistic, so get on the bus first."

But at least, someone is trying to catch the people falling during the technological transition with a large number of detailed rules and legislation.

What the US faces is closer to an institutional vacuum.

In April 2026, someone threw a Molotov cocktail at Altman's residence in San Francisco; in the same month, a city councilor who supported a data center project was shot at home. At graduation ceremonies, booing targeting AI company executives also began to appear.

If emotions cannot find an institutionalized outlet, they will find an outlet themselves.

At the same time, data from the Federal Reserve Bank of St. Louis shows that 39% of US GDP growth in 2025 came from data centers and AI-related investments.

The entire country is betting growth on a technology from which most citizens temporarily feel no benefits.

The US will hold midterm elections in 2026, and the 2028 party primaries are also destined to be crowded. Mark Kelly, Ro Khanna, and Josh Hawley have already released AI action platforms respectively.

Therefore, there may be only one indicator worth observing most in the next two years:

Can that negotiation table be set out before anger arrives?

It could be a law stipulating that enterprises cannot simply convert all productivity improvements into layoffs; it could be a new form of union, allowing employees of the same company to participate in negotiations on AI deployment together without having the same profession; it could also be a check printed with the name of an AI company, redistributing the returns created by technology to those bearing the transition costs through taxes, dividends, or public funds.

The specific form has not yet been determined, but the core question can no longer be avoided:

When enterprises no longer need to employ the majority to create wealth, on what basis should the majority share the wealth?

As for that 17-year-old child with only B grades, he will probably never know:

When the smartest group of people in an entire industry were asked "what should be done about him," their answer was "no answer."

Then, they each went back and continued to buy insurance for themselves.