David Ricardo and Intellectual Courage

The International Monetary Fund estimates that AI will affect 40 percent of all jobs. Not just low-skill positions. Skilled work—the kind that required years of training and provided solid middle-class paychecks.

You’ve probably heard the optimists’ response. Every disruptive technology looks threatening at first, they’ll tell you. Automobiles put carriage makers and stable owners out of business, but they created the oil industry, motels, drive-in theaters, and hundreds of other new sectors. Technology destroys old jobs and creates new ones. The workforce adapts. Why should AI be any different?

The answer comes from 1821, when an English economist named David Ricardo did something remarkable: He publicly admitted he’d been completely wrong about the most important economic question of his time.

The Confident Prediction

David Ricardo, born in 1772, watched England’s textile industry transform before his eyes. New spinning machines could convert raw cotton into yarn faster and cheaper than any home-based spinner working at a spinning wheel. Some spinners lost their livelihoods, true—but Ricardo observed that most found new work weaving the suddenly abundant and affordable yarn into cloth.

The system balanced itself. Productivity increased. Employment remained strong. Workers and entrepreneurs both prospered. Technology had disrupted one job while creating another.

Based on this evidence, Ricardo developed an optimistic theory about mechanization. On the floor of the House of Commons in 1819, he declared with confidence that “machinery did not lessen the demand for labour.”

That same year, in August, cavalry charged into a crowd of 60,000 protesters in Manchester. Eighteen people died. Hundreds were injured. They called it the Peterloo Massacre.

The protesters were demanding political reform. But their desperation had an economic source that Ricardo was only beginning to understand.

When the Pattern Broke

Power looms arrived roughly a generation after the spinning machines. Each one could produce more cloth than 10 to 20 handweavers working from their cottages. But these machines were enormous—they required factory buildings to house them. Home-based weaving became impossible.

When spinning machines had eliminated home spinning, displaced workers transitioned into the growing weaving trade. The math worked: fewer spinners needed, more weavers required. But when power looms eliminated home weaving, the displaced workers had nowhere to go. The factories didn’t need 10 to 20 workers for every loom they installed. The math didn’t work anymore.

In two Lancashire towns, family earnings for handloom weavers fell by half in just five years starting in 1814. The numbers tell the story more clearly: Handloom workers in the English cotton industry averaged 240 pence per week in 1806. By 1820—the year Ricardo stood before Parliament and assured them that machinery didn’t reduce labor demand—these same workers were making less than 100 pence weekly. More than half their income, gone.

Even the factory workers who secured jobs tending the powerful new looms weren’t prospering. Between 1806 and 1835, they experienced almost no real wage growth. The wealth created by these remarkable machines flowed somewhere else—just not to the people operating them.

The protesters in Manchester weren’t demanding abstract reforms. They were demanding survival.

The Revision

Ricardo was nearly 50 years old, wealthy, well-connected, at the absolute peak of his professional influence. He had every incentive to maintain his optimistic theory. His own wealth was tied to the system these power looms were enriching. Admitting error would undermine years of economic writing and parliamentary speeches.

He admitted error anyway.

The third edition of his Principles of Political Economy and Taxation, published in 1821, included an entirely new chapter on machinery. “The same cause which may increase the net revenue of the country,” he wrote, “may at the same time render the population redundant, and deteriorate the condition of the labourer.”

It was a complete reversal. The new machines could make the nation wealthier while making workers poorer. Both things could be true simultaneously. He’d been wrong to think otherwise.

This matters today because we’re repeating Ricardo’s first mistake.

The Hollowing Out

Since the 1980s, digital technology has been doing to middle-skill workers what power looms did to handweavers. Computers didn’t just speed up existing work—they eliminated entire categories of employment. Administrative support positions disappeared as offices adopted word processing and spreadsheets. Clerical jobs vanished as databases replaced filing systems. Blue-collar production work moved offshore or became automated.

These were jobs that employed people without four-year college degrees. They were the economic foundation of the middle class.

MIT labor economist David Autor calls it “the hollowing out of the middle class.” The statistics bear this out with brutal clarity. Between 1981 and 2021, average household income in the United States rose 95 percent after adjusting for inflation—a seemingly impressive number. But that average conceals the real story.

Income in the highest earning group climbed 165 percent. Income in the lowest earning group grew just 38 percent. The wealthy did more than four times better than the poor. The middle, meanwhile, stagnated or fell. The gains went to the top and, to a much lesser degree, trickled to the bottom. The middle got squeezed.

Where did those middle-skill workers go? Many were forced into lower-paying service work that didn’t require their expertise. Their training became worthless. Their experience didn’t transfer. Like the handweavers of 1820, they discovered that the economy no longer had a place for what they knew how to do.

This is the world AI is entering. Not a level playing field, but one already tilted by four decades of technological change that concentrated wealth and eliminated middle-class work.

The Optimistic Case: AI as Equalizer

MIT economists Daron Acemoglu and Simon Johnson—two of the three recipients of the 2024 Nobel Prize in economic sciences—see a different possibility. They argue that AI, handled correctly, could rebuild what earlier technology destroyed: meaningful middle-skill, middle-class employment.

Their argument turns on a simple observation. Our economy contains thousands of tasks that currently require doctors, lawyers, engineers, or professors—but don’t actually demand that level of expertise. These are decisions that credentialed professionals make because nobody else is allowed to make them, not because the decisions themselves are impossibly complex.

AI could change that calculation. It could give middle-skill workers the tools to perform higher-value work.

Consider healthcare, where this pattern is most visible. An experienced medical worker—someone who’s not a doctor but who has years of clinical experience—could perform lung ultrasounds with AI assistance. Lung ultrasound is powerful for diagnosing patients with shortness of breath, but it typically requires a doctor to perform and interpret. In a recent study, clinicians without ultrasound experience successfully obtained high-quality lung images using an AI tool. They didn’t replace doctors. They expanded the number of people who could do diagnostic work.

Or look at HVAC systems—heating, ventilation, and air conditioning. These systems have become increasingly complex, incorporating sophisticated sensors and control systems. When something breaks, technicians often need to call experts for guidance. Now they’re using ChatGPT-style AI tools that provide real-time troubleshooting guidance. A technician standing in your basement can solve problems that previously required backup from a mechanical engineer. Same worker, more capability, higher value.

Simon Johnson compares this to the rise of nurse practitioners over the past 30 years. “There weren’t very many nurse practitioners 30 years ago,” he notes. “And now every pediatrician’s office has them. And parents are very grateful for the advice and access that they can get.” It’s not about replacing doctors. It’s about enabling skilled workers to do more than they could before. Give them better tools, better training, better support systems—and they can take on responsibilities that previously required additional degrees.

“We’re just saying: Redress that balance,” Johnson explains. “Boost the people who don’t have four years of college. Enable them to be more productive.”

It’s an appealing vision. Technology that empowers workers rather than replacing them. AI that restores the middle class rather than further eroding it.

But it requires deliberate choices about how we deploy AI. Left to market forces alone, nothing guarantees this outcome.

The Pessimistic Case: AGI Changes Everything

Anton Korinek, an expert on the economics of AI at the University of Virginia, sees a fundamental problem with the optimistic scenario. It assumes AI will be like previous technologies—powerful tools that augment human capabilities in specific domains. That assumption might be wrong.

Artificial general intelligence (AGI, with human-level cognitive abilities) wouldn’t just automate specific tasks. It would automate thinking itself.

“The key distinction from past technological changes is that AGI would be capable of performing any cognitive task,” Korinek argues, “potentially leaving few unique economic roles for human workers.”

Past technologies automated physical labor or routine cognitive work. The printing press automated copying. The power loom automated weaving. The computer automated calculation. But humans always retained advantages in reasoning, judgment, creativity, and adaptation. Those advantages justified our wages.

AGI could eliminate those advantages. And if machines can think as well as humans—and think faster, more accurately, more tirelessly—then why pay human wages?

“This could lead to widespread labor displacement and significant wage declines,” Korinek warns, “unless governments intervene to avoid further widening of the wealth gap.”

His short-term recommendation focuses on jobs requiring authentically human qualities that machines can’t easily replicate or that humans prefer to receive from other humans: psychotherapists, childcare providers, religious counselors, hospice workers. These roles demand emotional intelligence, ethical judgment, and genuine human connection. We could train displaced workers for these positions while machines handle cognitive tasks.

But Korinek’s long-term vision is more radical. If AGI truly can perform any cognitive task, societies might need to implement universal basic income—no-strings-attached cash payments from governments to all citizens whether or not they work. The idea sounds extreme until you imagine an economy where machines outperform humans at most valuable activities. How do people earn wages in that world? How do they survive?

The answer can’t be “they don’t.” It has to involve sharing the prosperity that AGI creates.

What Ricardo Understood

We don’t have definitive studies yet on AI’s total economic impact. Researchers can’t fully account for jobs that don’t exist in industries not yet invented. The self-driving taxi didn’t appear in anyone’s 1990s job projections. The social media manager didn’t exist as a category until the 2000s. We’re bad at predicting what new technology will create because genuinely new things are, by definition, unanticipated.

But Ricardo’s story offers something more valuable than predictions: a model for how to think about technological disruption.

Ricardo was embedded in the establishment. He was wealthy, connected, respected. He had built his reputation on his economic theories. The industrial system enriching England was, in many ways, the system his ideas helped justify. Admitting that mechanization could harm workers would undermine his life’s work.

When he saw the evidence that his theory was incomplete—that the power looms were creating wealth for factory owners while impoverishing workers—he changed his mind. He revised his theory. He published his revision. He didn’t qualify it or hide it. He added an entire new chapter explaining what he’d gotten wrong.

“We have much to learn from Ricardo’s openness to new ideas and new ways of thinking about economics,” Acemoglu and Johnson write in their analysis of this period.

England eventually learned from the handloom weavers’ collapse, though the learning came slowly and at enormous human cost. Industrial cities like Manchester received representation in Parliament for the first time in 1832. The Factory Act of 1833 created the world’s first enforceable child-labor regulations, requiring factory inspections to ensure compliance. The government repealed the protectionist Corn Laws in the 1840s, making food more plentiful and affordable for the urban working class. Ricardo himself, an advocate for free trade, had pushed forcefully for the Corn Laws’ repeal.

These reforms improved workers’ lives. But they came a generation after the crisis began. Thousands of families suffered through those decades of political paralysis.

The Choice We’re Making

Daniel Kokotajlo, former governance researcher at OpenAI who now heads the AI Futures Project, has sketched one possible timeline. AGI arrives by 2027, driving an economic boom as AI-powered software outperforms humans in coding, research, and other cognitive tasks. Productivity soars. GDP grows. Unemployment also grows as millions discover their skills are no longer needed. He acknowledges now that “it might take a few years longer than 2027,” but the basic scenario remains plausible.

Other technologists and economists offer different timelines, different scenarios. Some see concentrated wealth at the top, a small class of capital owners and AI engineers capturing most gains. Others see reduced inequality if governments ensure broader distribution of AI’s benefits. The predictions span from utopian to catastrophic.

What they share is recognition that outcomes depend on choices we make now—not on the technology itself.

“All the smarts, all the talent around computer science is being drawn into the pursuit of AI, to a degree that I haven’t seen since the internet really boomed at the end of the 1990s,” Simon Johnson observes.

The question isn’t whether AI will be powerful. It will be. The question is who benefits from that power.

History offers a clear lesson on this point. Ricardo learned it when he saw the wage data from Lancashire. England learned it after Peterloo and the decades of unrest that followed. The lesson is simple but easy to forget: Just because you have miracle machines doesn’t mean most people will benefit from them.

The machines don’t make that decision.

We do.