AI Wealth Truth (10): Why Technological Progress Makes Ordinary People Poorer

I. We are taught to believe: technological progress is good for everyone. The steam engine made textiles cheaper. Electricity made factories more efficient. The internet made information more accessible. Every technological revolution raised productivity and improved living standards. But this hides a key question: how are productivity gains distributed?

II. Economists found: technological progress does not "lift everyone equally". It has direction. Some technology raises the productivity of high-skill labor. Some technology replaces low-skill labor. This selective impact is called skill-biased technological change (SBTC).

III. The core claim of SBTC is: over the past decades, technological progress has systematically favored high-skill workers. Computers boosted the productivity of analysts, programmers, engineers. But they replaced typists, switchboard operators, bank tellers. High-skill workers became more valuable. Low-skill workers became less valuable.

IV. This is not an accident. Who builds technology? Engineers and scientists with higher education. Whose needs does technology serve? Capital owners who want lower costs and higher efficiency. If a technology can replace workers with machines, capital owners adopt it. If a technology can make high-skill workers more efficient, capital owners adopt it too. Technology is not neutral. It serves the interests of those who build and deploy it.

V. Since the 1980s, the wage gap between U.S. high school graduates and college graduates has kept widening. In 1980, college graduates earned about 40% more on average. By 2020, the gap expanded to more than 80%. Having a college degree becomes more and more important. Not having one becomes more and more fatal.

VI. But the story is not over. SBTC is not only "high-skill win, low-skill lose". There is a more complex phenomenon: labor market polarization.

VII. MIT economist David Autor found: over the past decades, the U.S. labor market became "hollowed out". High-income jobs grew, such as doctors, lawyers, engineers. Low-income jobs also grew, such as cleaners, caregivers, service workers. But middle-income jobs shrank, such as manufacturing workers, administrative staff, bank tellers.

VIII. Why? High-income jobs require complex cognition and creativity that machines, for now, struggle to replace. Low-income jobs require interpersonal interaction, physical labor, and flexible adaptation, which machines also struggle to replace. But middle-income jobs are often "well-defined cognitive tasks": filling forms, processing data, following procedures. This is exactly what computers and automation do best.

IX. The result is that middle-class jobs disappear. Before, you could finish high school, enter a factory, earn a stable wage, buy a home, raise a family. Now that path almost no longer exists. You either squeeze into the high-skill track, requiring education and resources. Or you fall into the low-skill track, with low pay, instability, and little upward mobility. The middle path disappears.

X. AI pushes SBTC to a new extreme. Traditional automation can replace only tasks with clear rules. You tell the machine every step, and it follows. But AI can handle tasks with fuzzy rules. It can learn, judge, and decide.

XI. This means jobs once believed to require human intelligence now face replacement risk. Writing reports. Translating documents. Analyzing data. Designing images. Writing code. These were "high-skill" jobs and used to feel safe. But AI is entering these fields. The moat of high-skill workers is being filled in.

XII. More frightening: AI learns far faster than humans. A person needs ten years to become a competent lawyer. AI can digest all legal documents in days. Fields where humans still have an advantage today may be surpassed by AI in two years. Your skills' shelf life is shrinking rapidly.

XIII. Some people say: technological revolutions always created new jobs. Cars replaced carriages but created auto workers, gas station attendants, mechanics. AI will also create new jobs. This analogy has a fatal flaw: previous new jobs were designed for ordinary people. You did not need a PhD to become an auto worker.

XIV. New jobs created by AI may be highly concentrated at the top. Machine learning engineers who train models? The world may need only tens of thousands. Low-paid data labeling workers? They may also be automated by better AI. The quantity and quality of new jobs may be insufficient to absorb displaced labor.

XV. There is a deeper question: who captures AI's returns? When AI raises productivity, value increases. But that added value does not automatically flow to workers. AI is capital, not labor. Its output belongs to capital owners.

XVI. Suppose a company uses AI to replace 100 employees. Productivity stays the same, costs fall. Where does the saved money go? Shareholder dividends, executive bonuses. You are replaced. The company earns more. You lose your job. The stock price rises.

XVII. Some say displaced workers can retrain and learn new skills. In theory, yes. But retraining requires time, money, and cognitive resources. If you are already 40, have a mortgage, and have kids to support, can you quit your job and study for two years? "Retraining" is an impossible luxury for many.

XVIII. And the new skills you learn may be learned by AI before you graduate. Today, "prompt engineers" are in short supply. Two years later, they may not be needed at all. Because AI will become smart enough that it does not need carefully designed prompts. You are chasing a target that is accelerating away.

XIX. Technological progress itself is not the enemy. The problem is: how are the gains distributed? If all gains flow to capital owners, while costs are borne by workers. Then for workers, technological progress becomes the reason they get poorer.

XX. This is not technological determinism. Distribution consequences of technological progress can be shaped by policy. Higher minimum wages. Stronger unions. A "robot tax" on AI substitution. Universal basic income. These policies can let more people share the gains of technology. But these policies require political will, and politics is influenced by capital.

XXI. Without policy intervention, technological progress will continue to favor capital. High-skill workers may be okay. Low-skill workers will deteriorate. Middle-skill workers will be hit hardest, because their jobs are easiest to automate. Technological progress is not universal. It creates winners and losers.

XXII. Next time someone says "technological progress is good for everyone", ask: good for whom? Productivity rose. Who shared the increase? Costs fell. Did consumers benefit, or did shareholders? The employment structure changed. Who got the new jobs, and who lost the old ones? Technological progress is not manna from heaven. It is a redistribution with a direction. In the AI era, the winners of this redistribution are fewer, and the losers are more.