Compute and equity

Energy can grow concurrently only to a point. Below a threshold of per capita wattage, motors improve the conditions for social progress. Above this threshold, energy grows at the expense of equity. — Ivan Illich, 1973

I came across this while reading Ruan Yifeng’s blog. He illustrated it this way: “Higher speeds widen the distance between people, but only a minority can command high-speed transport. The faster it gets, the more it helps those people pull ahead.

Time also becomes more urgent, which benefits the wealthy more. For those at the bottom, life may become more valuable too—but also more exhausting.”

The broader implication is this: once a resource is no longer quite so scarce, it tends to end up in the hands of a few. They then use it to widen the gap between people and deepen social inequality. Today, that resource is compute. A small group possesses more powerful compute and better models, giving them a far greater edge than ordinary people in accumulating wealth.

I agree, but this predicament does not seem to have a systemic solution. So I spent some time discussing it with GPT:

---

question:

AI has clearly increased productivity in fields such as design, programming, and analysis, but it has also sharply reduced the number of related roles. At the same time, it has not created many AI-specific jobs. It feels as though more companies are laying people off and unemployment is rising. Is there data to support this? And where is this headed?

---

answer:

I think the feeling is partly supported by the data, but it also reflects the combined effects of AI, the business cycle, and corporate cost-cutting. It cannot all be attributed to AI.

An increasingly clear consensus in research from 2025–2026 is that AI changes hiring before it changes the unemployment rate.

In other words:

• Companies may not be laying off huge numbers of people;
• but they are noticeably hiring fewer.
• An employee who once did the work of two can now, with AI, do the work of three.

This shift is especially tangible for new graduates.

First: what data supports this?

The International Labour Organization (ILO) has released several studies over the past two years. Their conclusions are broadly consistent:

1. AI really does increase productivity

In customer service, software development, document work, translation, data analysis, and design, many experiments have found productivity gains of 20%–60%. There is extensive experimental evidence for this by now.

2. But employment has not grown in step

This is the part you are feeling most clearly.

The ILO’s latest summary this year says that the productivity gains from AI are real, but have not yet translated clearly into more jobs or higher wages.

Put differently: much of the cost saving goes to profit, or is reinvested in AI, rather than used to hire more people.

3. Entry-level roles are seeing the clearest effect

Nearly every institution is now observing this, including among:

• Junior developers
• Junior designers
• Junior copywriters
• Junior customer-service staff
• Administrative staff

Hiring demand is falling fastest there. The ILO believes young people are being hit hardest. Many companies are effectively saying: “Hire one senior instead of three juniors.”

Second: why does it feel as though there are more layoffs?

There is a major misconception here. Many companies are not saying, “AI is here, lay off 500 people.” What actually happens is that a 100-person team loses people through attrition, the company does not replace them, and the remaining 80 use AI to absorb the workload.

So layoff numbers do not necessarily explode. But hiring keeps shrinking. Many studies suggest that AI’s current impact on the labor market shows up more in reduced hiring than in mass unemployment.

Third: why do stock markets rise while ordinary people feel none of it?

This has become a major topic in Western economics over the past two years. Take a software company that once had 1,000 employees, 10 billion in revenue, and 2 billion in profit. After adopting AI, it has 800 employees, 11 billion in revenue, and 4 billion in profit.

Shareholders make a fortune and the stock reaches new highs, but ordinary employees do not. That creates a pattern: capital markets keep improving, while labor markets become increasingly ordinary.

Some analysts have even begun discussing the possibility of a jobless productivity boom.

Fourth: I worry that the next five years may bring a new phase with no real historical precedent

During the Industrial Revolution, machines replaced physical labor. Later, computers replaced repetitive work. But AI is the first technology to begin replacing cognitive labor—and white-collar work in particular: programmers, lawyers, accountants, analysts, designers, translators, and educators.

People used to think higher education made them safer. AI is the first technology to seriously disrupt that assumption.

Fifth: will this produce mass unemployment?

Economists are divided into two camps.

Optimists argue that history has always worked this way. Cars displaced horse-drawn carriages, then the auto industry created more jobs. Computers displaced typists, then created the internet economy. In the end, AI too will create new industries.

Pessimists argue that this time is different. AI is not like a car: a car cannot write code. AI can continue improving itself. The jobs it creates may not keep pace with the jobs it eliminates. More recently, over two hundred economists and people in technology have jointly called for countries to study AI’s employment shock in advance, rather than wait until the problem fully arrives.

My view—and what I consider one of the most important questions of the next decade

The real concern is not whether “AI will replace every job.”

It is that AI may split the labor market into two tiers.

The first tier is the very small group that can manage, design, and lead AI. Their incomes keep rising.

The second tier is the large number of ordinary white-collar workers whose wages are increasingly easy for AI to compress.

In other words, the greatest future risk may not be universal unemployment, but an even wider income divide.

---

question:

The cost of acquiring knowledge is indeed lower in the AI era, but the gap between people may also widen sharply. Not everyone is willing to keep learning new things through AI; some just use it to assist with their existing work. Wouldn’t that widen the wealth gap and, in turn, social tensions? Put another way: in this era, do the people willing to compete hardest reap the greatest rewards?

---

answer:

Your observation reaches one of the central social questions of the AI era: AI lowers the cost of acquiring knowledge, but will it make society more equal—or instead widen the gap?

Academia currently has two opposing views. Your concern is closer to the ideas of skill-biased technological change and winner-take-all effects.

My view is:

AI will probably widen the gap between some groups of people, but that does not mean everyone must compete without limit just to survive. The real divide may lie in whether someone can turn AI into productive capacity.

1. Why might AI widen the gap between people?

Because AI has a distinctive property: it does not simply replace low-skilled labor; it amplifies highly capable individuals.

Take a simple example. In the past, an average programmer might write 100 lines of useful code per day and an expert 300—a threefold gap. With AI, the average programmer may produce 1,000 lines a day, while the expert—who understands architecture, the business, and AI—may produce 5,000. The gap becomes fivefold.

Experts also have another advantage: they know what to ask AI to write and which problems they need to think through themselves. AI may therefore become an ability multiplier.

2. This resembles the internet era

The internet did not distribute wealth evenly. Instead, it created:

• World-class programmers
• Founders
• Content creators
• Investors

They could serve a global market. An outstanding software engineer who once influenced only one company can now influence the world through an open-source project. That is the rise of the “super-individual,” and AI will strengthen this trend further.

3. But why will not everyone make an effort to use AI?

You raise a very real point: not everyone wants to keep learning. That is indeed crucial. Technological change has always worked this way.

In the internet era, some people learned to build businesses online while others merely watched videos. In the smartphone era, some used phones to do business while others used them only for entertainment.

Technology offers opportunities, but it does not transform everyone automatically.

4. A three-tier structure may emerge

This is what many researchers worry about.

First tier: AI-augmented talent

They:

• Have domain knowledge
• Can design workflows
• Can manage AI

Their income may rise substantially.

Second tier: ordinary AI users

They can use AI to get their work done. Their productivity improves, but they remain easy to replace, so their income may stay flat.

Third tier: people displaced by AI

This includes many people doing repetitive work. Without a transition, they face the greatest pressure.

This is very similar to the Industrial Revolution. In its early days, machines raised productivity but did genuinely hurt textile workers. Later, society reabsorbed labor through education, new industries, and institutional adjustment.

5. So do those who compete harder earn more?

There is an important distinction here. The logic of the past twenty years was:

Work hard and study → good degree → good job → high income.

In the AI era, it may become:

High-leverage capabilities → high returns.

What does that mean?

It is not simply about working harder; the direction of effort matters. Someone who works ten extra hours every day at repetitive tasks will see their value decline as AI arrives. Someone else who spends two hours a day learning AI tools, understanding their industry, and building a personal reputation may see much higher returns.

Competition may therefore shift from effort-intensive competition to competition in cognitive leverage.

6. But society will not allow the gap to widen without limit

History suggests that when productivity rises but its gains become too concentrated, social pressure follows.

Possible responses include:

• Stronger tax redistribution
• Education reform
• Broader access to AI tools
• Debates around basic income
• Shorter working hours

These are not unique to China. Europe and the United States are also discussing whether labor systems and social security need to be redesigned as AI concentrates wealth.

7. An interesting contradiction

AI may produce two outcomes at once:

For individuals, those who are better at using AI gain bigger advantages, widening the gap. For society as a whole, productivity rises. In theory, everyone could work less and live better.

The question is: how are the gains from productivity growth distributed? That has been the enduring problem since the Industrial Revolution.

In light of what you have been considering, I think you have identified a key point

You have been paying attention to:

• The AI bubble
• Improvements in ordinary people’s lives
• Social fragmentation in China
• Changes in education
• Career competition

Behind all of these is the same theme:

Technological progress creates more wealth, but it does not automatically turn that wealth into a better life for most people.

The greatest challenge of the AI era may not be whether AI is capable, but whether a society can give more people the chance to use AI to increase their own value.

If it cannot, AI may become an amplifier:

• The strong grow stronger
• The weak find it harder to catch up

If it can, AI may become one of the most powerful productive tools in history—one that greatly expands the bounds of ordinary people’s capabilities.