EatingThe Seed Corn (Updated)

The Village Idiom explains:

In the idiom "eat one's seed corn", the phrase refers to consuming or disposing of valuable resources without considering the long-term consequences. It implies a short-sightedness and lack of foresight, often resulting in detrimental effects on future prospects or success.

Salomé Balthus
Uwe Hauth, CC BY-SA 4.0

More than a year ago in Going Out With A Bang I quoted Salomé Balthus:

'The elephant in the room is climate change. Everyone knows it can't be prevented any more,' she said, adding that the 'super rich' could generally be split into two groups on the topic.

'The one group thinks it only affects the poor, the "not-white race", while the others fear that it could get worse but there's no sense in trying to do anything about it so they just enjoy themselves,'

In fact both groups have decided to steal everything they can get their hands on to fund their retreat to bunkers on private islands protected by armed autonomous humanoid robots. Below the fold I look at yet another example of this trend.

Stanford has an on-going research program tracking the impact of generative AI on US employment. Their first report was published in August 2025 entitled Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence, and a year later they published a an update whose abstract reads:

Using a sample of high-frequency administrative payroll data from ADP covering millions of U.S. workers through June 2026, we document six facts about the labor market following the widespread adoption of generative AI.

1. We find no evidence of widespread, economy-wide job displacement.
2. However, employment of young workers (ages 22–25) in AI-exposed occupations now stands 19% below where it would be had it kept pace with that of their less-exposed peers; experienced workers show no comparable gap.
3. This divergence has widened steadily since we first documented it in August 2025.
4. It operates primarily through reduced hiring of young workers rather than increased separations.
5. Declines are concentrated in occupations where AI usage primarily substitutes for human tasks; where usage primarily complements workers, employment is flat or rising, especially for experienced workers
6. Adjustment is occurring through employment rather than base compensation.

Dodini & Smith Chart 3

Samuel Dodini and Tucker Smith of the Dallas Fed agree with the Stanford team in Job postings show early signs of AI automation impact:

After the release of ChatGPT in late 2022, job openings fell for occupations whose tasks are automatable by GenAI. The decline was not confined to new firms or driven by a reduction in the number of surviving firms. Surviving incumbent firms posted fewer openings and shifted the composition of their job posts away from more AI-exposed occupations.

These results characterize the early effects of the AI transition on labor demand. Follow-up analyses suggest this shift in demand has negatively affected the labor market outcomes of recent college graduates from Texas universities and prompted current students to change their educational decisions.

The authors argue that the major impact is on recent college graduates:

There is strong evidence that GenAI has decreased labor demand for occupations consisting of tasks that can be performed by these new tools. Though the overall effect on aggregate online job posting behavior thus far has been modest, demand reductions have been significant and meaningful for specific types of workers. Even in the absence of layoffs, a hiring pullback in these jobs would negatively impact individuals entering the labor market for the first time or those attempting a job transition.

In the Lightcast data, fewer than half of firms’ typical job ads explicitly require more than two years of experience and very few firms require more than five (Chart 4). Because the types of jobs posted online typically require little prior work experience, a decline in job postings is likely to disproportionately affect new labor market entrants.

These dynamics suggest that recent college graduates, whose unemployment rate rose to unusually high levels during this period of rapid GenAI adoption, are where effects of GenAI on employment and earnings are likely to first appear.

A year ago the employment gap for AI-exposed young workers was 13%. Now it is 19%. There are two big problems this loss of entry-level jobs causes:

• Loss of expertise
• Loss of social stability

Loss of Expertise

@HowToPrompt had a long tweet on the topic. An extract:

Every industry relies on what researchers call the "Cognitive Commons”, a shared pool of deep professional expertise that regenerates itself generation after generation.

How is that expertise built?

Through friction. Through doing the boring, difficult, junior-level work. Through manual trial and error.

That is how you build "Internalized Mastery."

When companies adopt AI to automate entry-level work, the logic looks airtight on a quarterly balance sheet. Cut costs. Speed up output. Skip the grunt work.

Every individual company is making a rational choice. But collectively, they are draining the pool dry.

When you eliminate junior roles, you destroy the pipeline that creates senior experts.

Who needs senior experts?:

Effective human oversight of AI depends entirely on deep domain expertise.

As professionals increasingly delegate thinking to machines, they lose the mastery required to spot when the AI is wrong.

The experts validating AI outputs today look competent.

The terrifying question is: Who comes after them?

When the current generation of veterans retires, who will be left with the skills to audit the systems running our economy?

Nobody.

We are trading our long-term intellectual independence for short-term productivity boosts.

The tweet was triggered by Nolan Lovett's The Tragedy of the Cognitive Commons: How AI Could Disrupt the Regeneration of Professional Expertise:

Artificial intelligence is reshaping cognitive work, but Human Resource Development scholarship has treated this transformation as an organizational training challenge, leaving the collective regeneration of professional expertise unexamined. This conceptual paper introduces the Cognitive Commons framework, integrating commons theory, HRD scholarship, and distributed cognition to explain how rational AI adoption decisions can deplete the shared expertise pool professions require for renewal. The framework distinguishes Internalized Mastery (deep domain knowledge from sustained practice) from Distributed Mastery (orchestrating human-AI systems), and develops the Validation Tether: effective AI oversight depends on the expertise AI adoption may undermine. Early labor market and clinical evidence suggests possible disruption to expertise-regeneration pathways in highly AI-exposed sectors, though adoption is recent and the strongest signals come from leading sectors rather than all professions. Five factors determine occupational vulnerability, and governance arrangements may form across organizational, professional-association, and policy levels. The paper reframes expertise development as collective stewardship rather than organizational optimization, with implications for HRD theory and workforce policy.

The key to Lovett's argument is this framing, clarifying the way everyone acting in their short-term interests is eating the seed corn:

The tragedy framing identifies a collective action problem that current arrangements leave unaddressed; the governance question, taken up later, is whether institutions can be designed to sustain the regeneration pathways on which professional expertise depends.

The problem is the side-effects of the AI drug in reducing understanding of the underlying problem:

Wiles et al.’s (2024) experimental evidence is instructive: when AI assistance was provided during a skill-building task, participants performed significantly better during the access period, but this performance advantage did not transfer to subsequent unassisted performance. AI-assisted productivity and AI-independent expertise are not the same capability, and developing one does not automatically build the other. Human factors researchers have recognized this paradox for decades: Bainbridge (1983) demonstrated that automation simultaneously increases the need for human skill and reduces the opportunity to develop it, an insight extended to system-level analysis by Hollnagel and Woods (2005). What this literature has not addressed is the collective action structure that emerges when every organization in a profession faces this paradox simultaneously.

This effect is greatly enhanced by the sycophantic nature of chatbots, which are deliberately designed this way to enhance engagement and over-confidence, i.e. addiction.

Lovett asks the key question:

If cognitive capability can distribute across human-AI networks (Hutchins, 1995), why does concentrated individual expertise matter? Hutchins’s naval navi­ gation analysis showed that effective coordination requires humans who understand what distributed systems are doing and why, not merely how to operate them. Extended through recent research on transactive memory in human-AI partnerships (Bienefeld et al., 2023; Woolley & Gupta, 2024), this reveals what is termed the Validation Tether: effective orchestration of opaque AI systems requires independent domain understanding sufficient to recognize when systems operate outside competence boundaries

This necessary understanding will have eroded away just as the LLMs suffer model collapse because the vast majority of the available training material is AI slop.

Update: 10^th September 2026

Source

Two pieces of evidence show that the deleterious effect of AI is showing up even earlier in the expertise pipeline. First, Brad DeLong cites David Stromberg et al's The Generative AI Learning Penalty: Evidence from Chinese Secondary Education:

Using 30 months of panel data on 26,811 Chinese students in grades 7-12, we study how generative AI affects homework productivity and learning. The data combine monthly closed-book exams, high-school and college entrance exams, and homework scores and completion time across nine subjects. We exploit staggered AI adoption in a difference-in-differences design. AI adoption raises homework scores by 18% and reduces completion time by 30%, but lowers monthly exam scores by 20% within six months. High-stakes entrance-exam scores fall by 18 and 24%, with the full penalty emerging only after about two years. The losses are largest in social science subjects, followed by STEM and languages, and are especially large for junior students, high-achieving students, and boys. The learning losses are concentrated among roughly 80% of AI users whose behavior is consistent with homework outsourcing, as indicated by exceptionally short homework completion time coupled with high homework scores. AI users who maintain similar homework completion time as non-AI users experience small learning losses.

Source

Second, the OECD's Programme for International Student Assessment (PISA) has released their 2025 results. Among their results are two very worrying charts:

• The average scores in science, reading and math peaked in 2009 then started a slow decline. The 2022 and 2025 scores show a marked acceleration of the decline. The 2022 decline was initially explained by the effects of the pandemic, and some recovery was expected. But the further decline, especially in reading and math, in 2025 makes that explanation suspect.

• 

  Source

  The use of generative AI in schools produces a significant decrease in math score, increasing with the amount of generative AI use.
• The gap between the low and high scores in the US was the largest of any country. The tech oligarchs are notorious for placing very strict limits on their children's screen time, and thus use of AI. They are happy for the expertise of the masses to decay, because this will give their children a huge advantage as they shelter in their bunkers.

Loss of Social Stability

Source

Lets look at the New York Fed's data on graduate employment. First, the rate of unemploment for recent college graduates (ages 22-27) in blue, compared with the rate for all college graduates in red, all young workers in grey and all workers in black. Note that the three comparisons include the recent graduates. Observe that the gap between recent graduates and all graduates has been widening gradually over time, suggesting that AI is not the only factor making it hard for recent graduates to find work. And note that around 6% of recent graduates are unemployed. The Fed also tracks youth unemployment for ages 15-24, currently 9.34%. The rate for 22-27 is 7.2%, so age does push people toward the workforce in their twenties.

Source

Second, the rate of underemployment for recent college graduates compared with all college graduates. Underemployment is defined as the "share of graduates working in jobs that typically do not require a college degree". Note that about a third of all college graduates are underemployed, and about 42% of recent college graduates.

Even before the major impact of AI, almost half of college graduates under 28 years old are either un- or under-employed. This seems like a waste.

As in many technology-related areas, China is well ahead of the US. Barclay Bram's The 19 Percent Revisited: How Youth Unemployment Has Changed Chinese Society describes their stage of the problem:

China’s urban youth unemployment rate for ages 16 to 24 has hovered around 20% since the pandemic. When it reached a record 21.3% in mid-2023, the National Bureau of Statistics stopped publishing it. When reporting resumed a few months later, figures were calculated using a new methodology that corrected downward. But even with the new system, youth unemployment stood at 16.9% in February 2025, ... Many believe the true figure to be much higher.

With the history of the "iron rice bowl", early unemployment's impact on youth is bigger in China:

while the 16–24 cohort does not represent a significant segment of the labor force, the cliché holds that “the young are the future.” That many of them are struggling to find work, settling for unstable or unfulfilling jobs, or staying in school to forestall entering the workforce altogether means that their first experiences are ones of significant setback and rejection. Moreover, while many generations have struggled and overworked, what is unique about today’s cohort is their education level and a sense that their hard work will not pay off as the economy sputters.

Jobs in traditional manufacturing are vanishing:

In the Pearl River Delta and other areas of China, a “reverse China shock” has caused once-booming factory towns to grow quiet. The most labor-intensive, low-skilled factory work has shifted to Southeast Asia and other places with lower wages. Further shifts have been canceled or scaled back because of Washington’s tariffs. Many factories that remain open are employing automated systems that require fewer workers. The twelve most labor-intensive industries shed 3.4 million jobs from 2019 to 2023,15 following a loss of 4 million jobs from 2011 to 2019. The textile industry also created 40% fewer jobs during this period. Factory labor has become increasingly flexible, reflecting a broader trend toward gig work—over 40 million flexible laborers in the manufacturing sector now account for 31% of total employment.

China Domestic Savings

China's savings rate has always been very high, but it has recently decreased a little. But it isn't the young who are saving:

Young people have few options to escape this intense competition, and many of them are falling into debt. China’s high savings rate has only increased in the years since the COVID-19 pandemic; from 2021 to 2024, total household savings increased by 50%. But the ability to save is predicated on having income and assets. By some estimates, as many as 87% of people under 30 carry some form of debt, and the risk of default rises with unemployment and a sluggish housing market.

China is an example of trends worldwide:

One of every three Indian graduates is unemployed. In the United States, Gen-Z intellectual Kyla Scanlon has written about the “end of predictable progress” as young graduates enter the job market saddled with huge student loans and skillsets poorly matched to a world of increasing automation and AI. Falling birthrates are a salient issue in democracies and authoritarian states alike. It is therefore important to examine the circumstances particular to China and how its political system responds.

Against the backdrop of a difficult job market and the broken dreams of tens of millions of young Chinese, a marked shift in perceptions of inequality is occurring. In representative national surveys before 2014, the majority of respondents believed that inequality in Chinese society was largely the result of individual failings in an ascendant China. By 2023, the majority saw inequality as a structural failing, related to unequal opportunities, corruption, and a failing economy
...
An entire generation is growing up with deep reservations about the status quo, questioning what it means to live the “good life.” As anthropologist Biao Xiang has noted, the anxieties expressed by many Chinese youth today are fundamentally existential.

Li Yuan reports on the way these "deep reservations" are expresed on-line in Social Media in China Is Getting Really Dark:

Across RedNote, Douyin, Weibo and other popular platforms, you can scroll endless posts about meager wages, scarce jobs, falling property values and fear about the future. Some turn their hardships into dark humor. Others hijack official posts and hashtags and turn propaganda into spectacles of mockery.
...
“In this environment, you’ve won if you didn’t buy apartments, didn’t buy cars, didn’t invest or didn’t get married early,” read one post on RedNote. People born in the 1980s who strove the hardest, it said, are likely the biggest losers. The post was liked nearly 2,000 times.

This is becoming a problem for the government:

China’s internet censorship has grown increasingly ruthless over the past decade. That makes the sheer volume of the pessimistic posts and sarcastic comments all the more striking. At the same time, it’s difficult to know why some posts survive long enough to go viral while others quickly disappear.

One possible explanation is that policing too aggressively could hurt traffic and advertising revenue. Another potential explanation is that recommendation algorithms allow scattered grievances to find one another.

For the bigger picture we can turn to Wikipedia's Youth Unemployment: Social Stability:

Youth unemployment is often seen as a key driver of revolution, political instability, societal upheaval, and conflict against the government or state. Historically, high levels of youth unemployment have been linked to significant political and social change, including the overthrow of established political systems. Major events such as the Arab Spring, Russian Civil War and the French Revolution all largely being caused by large scale youth unemployment.

The rise of political unrest and anti-social behaviour in the world has been recently attributed to youth unemployment. During the course of 2011 it became a key factor in fuelling protests around the globe. Within twelve months, four regimes (Tunisia, Egypt, Libya, Yemen) in the Arab World fell in the wake of the protests led by young people. Riots and protests similarly engulfed a number of European and North American cities (Spain, France, United Kingdom between 2008 and 2011 for example). The lack of productive engagement of young people in wider society, underlined by high levels of unemployment and under-employment, only serves to add to this feeling of disenfranchisement.

Conclusion

The combination of a world filled with LLMs unchecked rampant hallucinations, and a revolution of uneducated, un- and under-employed youth, is a future that will make us all wish for a refuge on a private island.