In the Sixties, We Knew Automation Could Be a Serious Threat. Now That AI Is Here, Why Aren't We More Prepared?

In March 1964, President Lyndon B. Johnson sent Congress a warning about automation.

America was becoming more productive. New technology was helping create what Johnson called the highest standard of living in the world.

But he saw another possibility.

Technological progress could also bring “dislocations, loss of jobs, and the spectre of poverty in the midst of plenty.”

Johnson did not propose stopping technological progress. He proposed studying where it might lead before the country got there.

He asked Congress to create a National Commission on Automation and Technological Progress to examine current and future technological change and recommend how the country could obtain the greatest benefits while minimizing the harm. Congress created the commission later that year.¹

That was more than 60 years ago.

Computers were still enormous and expensive. There was no internet. There were no smartphones. There were no large language models capable of writing, coding, analyzing documents, creating images and performing pieces of professional work in seconds.

Artificial intelligence (AI) existed primarily as a research field.

And yet the United States was already asking a question that has become much more urgent today:

What happens to people when technology can perform more of the work people once performed?

So why, now that the technology has become vastly more capable, does our preparation still seem so tentative?

They Saw More Than Job Loss

Something remarkable was happening in the debate of the 1960s.

People were not merely worried that a particular machine might replace a particular worker.

Some were asking what automation could eventually do to the structure of the economy itself.

In 1964, an independent group of scientists, academics and social thinkers sent President Johnson a document called The Triple Revolution. They used the word “cybernation” for the combination of computers and automated machinery.

Their prediction of how quickly mass technological unemployment would arrive proved wrong.

But one of the economic mechanisms they identified is strikingly familiar today.

They argued that the industrial economy relied heavily upon an “income-through-jobs link.” People participated in production through their labor, received income for that labor, and then used that income to purchase what the economy produced.

If machines increasingly took over production, they asked, what would replace the mechanism that distributed purchasing power to consumers?²

That is remarkably close to the problem I call the Paradox of Automation.

For one company, replacing labor cost can be rational. For many companies doing it simultaneously, the result may be very different because wages are not merely a cost to employers. They are also purchasing power.

People were thinking about that problem when a single computer could fill an entire room.

The Government Took the Concern Seriously

The National Commission on Technology, Automation, and Economic Progress issued its report in 1966.

It did not conclude that automation was about to eliminate most employment. In fact, the commission concluded that technological change displaced individual workers but was not primarily responsible for the overall level of unemployment at the time.³

That distinction is important.

But look at what else the commission considered.

Its recommendations included public-service employment with government acting as an employer of last resort for some workers, an income floor to protect families, relocation assistance for people stranded in declining areas, a computerized national job-matching system and continuing study of technological change and national economic goals.⁴

Those recommendations sound reassuring on paper. Their scalability is another matter.

Helping a modest number of displaced workers relocate, retrain or temporarily replace lost income is one thing. Trying to do the same after a large, persistent reduction in the economy's need for human labor is something entirely different.

The United States would not enter such a transition with unlimited financial capacity or a flawless social safety net. Existing programs already struggle with ordinary unemployment and worker transitions. The Government Accountability Office has placed unemployment insurance on its High Risk List because of longstanding administrative, access and technology problems.⁵

The federal government also enters the AI transition while already running large deficits and carrying historically high debt relative to the economy.⁶

A modest displacement might be manageable with the institutions we have, supplemented by additional resources. A much larger disruption could overwhelm systems that were never designed for it.

And there is another problem. If significant labor displacement reduces wages and employment, government may simultaneously face greater demand for assistance and weaker tax revenues tied to labor income. The very moment when more people need help could also be the moment when part of the traditional revenue base used to help them is under pressure.

That does not mean government would be powerless. It could borrow, change taxes, redirect spending or create entirely new mechanisms for distributing income. But those are major fiscal and political choices.

A safety net designed to catch people temporarily between jobs is not necessarily capable of supporting a society in which substantially fewer jobs exist.

The commission could be wrong about the scale of the coming problem and still be doing something valuable.

It was thinking ahead.

Then Something Important Happened: The Worst Predictions Failed

Automation continued.

Factories became vastly more productive.

Agricultural employment collapsed as a share of the workforce.

Computers entered offices.

Automated teller machines appeared in banks.

Industrial robots entered factories.

Software eliminated enormous quantities of clerical work.

The internet destroyed some industries and created others.

But the feared permanent mass technological unemployment did not arrive.

Workers were displaced. Occupations disappeared. Communities were hurt. Some people never recovered their previous economic position.

Yet at the national level, the economy repeatedly generated new forms of employment.

That history created a powerful argument:

People have been predicting that machines will eliminate human jobs for generations. They have always underestimated our ability to create new work.

There is considerable evidence behind that argument.

It would be foolish to ignore it.

But there is a danger in turning a historical observation into a universal law.

The fact that previous technologies did not permanently eliminate enough human work to destabilize the employment system does not prove that no technology ever can.

History tells us what happened before.

It does not guarantee that every future technological change must produce the same result.

The Concern Never Completely Disappeared

It would also be wrong to say the United States simply forgot about automation.

In 2016, the Obama administration published Artificial Intelligence, Automation, and the Economy. It explicitly warned that AI could automate tasks that had long required human labor and disrupt the livelihoods of millions of Americans. It discussed education, retraining, stronger worker protections and policies intended to spread the benefits of technological change more broadly.⁷

The current administration is also studying the labor effects of AI.

Its 2025 America's AI Action Plan calls for an AI Workforce Research Hub to examine AI adoption, job creation, displacement and wages and to conduct scenario planning across different levels of AI impact. It also calls for AI education, skills development, apprenticeships and rapid retraining for workers affected by AI displacement.⁸

In April 2026, the Department of Labor and National Science Foundation announced funding for state and territorial AI-readiness hubs intended to help workers and employers adapt.⁹

So it would be inaccurate to say government is doing nothing.

That is not my concern.

My concern is whether what we are doing is proportional to the risk we are now facing.

The Risk Is Not the Same as It Was in 1964

The people worrying about automation then were extrapolating from relatively narrow machines and primitive computers.

They were imagining what technology might someday become capable of doing.

We no longer have to imagine quite as much.

By 2026, AI systems can write, code, summarize, translate, analyze documents, generate images and increasingly use tools. Businesses are adopting generative AI rapidly, even though deep workflow replacement and economy-wide labor displacement remain much earlier-stage phenomena.¹⁰

Large-scale employment effects have not yet appeared across the entire economy. That distinction is critical.

But early labor-market signals are no longer entirely hypothetical either. Recent research has found deterioration concentrated among some younger workers in occupations more exposed to generative AI, while other evidence still shows little economy-wide displacement so far.¹⁰

None of that proves that mass technological unemployment is coming.

But it establishes something much more modest:

The technology about which people speculated in the 1960s has become capable enough that the risk deserves to be taken more seriously, not less.

And yet much of our institutional response still seems built around an assumption inherited from previous technological transitions:

There will be another job.

“We Will Retrain Them” Contains an Assumption

Retraining is a sensible response when technology changes the kinds of workers the economy needs.

A factory closes.

Healthcare needs more workers.

Train some former factory workers for healthcare.

Clerical positions decline.

Cybersecurity positions increase.

Create a pathway between them.

Software changes accounting.

Teach accountants to work with the new software.

That is difficult enough in practice.

But conceptually, the solution is straightforward.

A worker's old job disappears.

The economy needs that worker somewhere else.

Move the worker.

Much of today's AI workforce policy follows that model.

Teach people AI skills.

Expand apprenticeships.

Identify growing occupations.

Improve job matching.

Retrain displaced workers rapidly.

Those measures may be extremely useful.

But they contain an assumption that deserves far more attention:

What if the economy does not need an equivalent number of workers somewhere else?

That is a different problem.

Retraining can solve a skills mismatch.

It cannot by itself solve an insufficient demand for human labor.

Those are not the same thing.

There Is a Large Difference Between Worker Transition and Economic Transition

Suppose AI eliminates one million jobs but helps create 1.2 million different jobs.

We have an extremely difficult worker-transition problem.

People need new skills. Some need to relocate. Some will earn less. Education systems have to change. Employers have to find workers. Government may need to help bridge the transition.

But the basic employment system survives.

Now consider another possibility.

AI allows the economy to produce the same or greater output while permanently requiring substantially less human labor.

Now retraining is not enough.

We have to ask different questions.

What happens to labor income?

What happens to consumer spending?

What happens to tax revenue tied to wages and salaries?

What happens to Social Security and other systems financed partly through payroll taxes?

What happens to mortgages and household debt when income falls but debts do not?

What happens to the businesses whose customers depended upon the income from those jobs?

What happens to the value of productive assets if increasingly efficient businesses eventually discover that too few customers can afford what they produce?

And what happens if all of those effects begin reinforcing one another?

Those are not questions about teaching somebody to use AI. They are questions about how an economic system functions.

This Is Where the 1960s Look Surprisingly Modern

The strange thing about rereading the automation debate of the 1960s is not that those people correctly predicted our future.

They did not.

They underestimated the economy's ability to adapt to the technologies of their time.

But they were willing to ask a broader question:

What if ordinary adaptation is not enough?

Their federal commission considered an income floor.

It considered government as an employer of last resort.

It considered relocation.

It examined the relationship between technological change, employment and income.

And outside government, The Triple Revolution was explicitly asking what would happen to consumer purchasing power if the connection between jobs and income weakened.

More than 60 years later, with vastly more capable technology, much of today's response still begins with:

How do we prepare workers for the jobs of the future?

That is an important question.

It just may not be the only one.

Another Country Is Beginning to Ask the Broader Question

Britain offers an interesting comparison.

In June 2026, the United Kingdom created an Artificial Intelligence Economics Institute (AIEI) jointly under its Treasury and Department for Science, Innovation and Technology.

Its mandate is broader than worker retraining.

It is supposed to study AI's effects on productivity, labor markets, companies and income distribution, build models of economy-wide effects, and run scenarios to help government prepare for different possible AI futures.¹¹

It is new.

It may succeed.

It may fail.

It may produce excellent reports that politicians ignore.

Creating an institution is not the same thing as solving a problem.

But its existence recognizes an important distinction:

AI may create an economic transition, not merely a workforce transition.

That is the distinction I want to see taken much more seriously.

Preparation Cannot Stop at the Border

Even a well-prepared United States cannot address this problem entirely by itself.

Modern economies are deeply connected through trade, capital, technology and supply chains. If one country restrains labor-replacing AI while competitors accelerate, companies inside the more cautious country can still face automated foreign competitors, cheaper imports and pressure to move capital or production elsewhere.

National policy can reduce harm and improve preparation. It cannot create an economic firewall around a global technological transition.

The United Nations has now created a Global Dialogue on Artificial Intelligence Governance built around the premise that AI's opportunities and risks cannot be managed effectively by countries acting entirely alone.¹²

That creates another problem beyond the scope of this article: meaningful management of the largest AI risks may eventually require international coordination.

Unfortunately, the same international competition accelerating AI also makes that coordination extraordinarily difficult.

Preparation Should Rise With Risk

We do not know how many jobs AI will ultimately replace.

That uncertainty should be stated clearly.

Perhaps AI will create enough new industries and occupations to absorb most displaced workers, just as earlier technologies did.

Perhaps humans and AI will prove highly complementary.

Perhaps labor shortages caused by aging populations will absorb much of the displacement.

Perhaps productivity will increase wages enough to support strong consumer demand.

Those are all plausible possibilities.

But there are less comfortable possibilities too.

And the stakes are different now because the technology is different.

In 1964, a commission worried about what automation might someday become.

In 2026, companies and governments are spending enormous sums specifically to make machines more capable of performing cognitive and physical tasks that currently require people.

The question therefore should not be whether we can prove today that the Paradox of Automation will happen.

We cannot.

Has our preparation increased in proportion to the seriousness of the possibility?

I am not convinced that it has.

The Warning Does Not Have to Be Right to Be Worth Hearing

There is another trap in thinking about future risks.

Suppose someone identifies a danger.

People respond to it.

The danger never fully materializes.

Years later, people may conclude that the original warning was foolish.

But sometimes a warning looks wrong because people listened to it.

And sometimes a reckless decision looks intelligent because the bad outcome happened not to occur.

A driver who speeds through a dangerous intersection without crashing has not proven that speeding through the intersection was safe.

He has proven that he got through it once.

Good planning requires separating the quality of the decision from the luck of the outcome.

That principle becomes more important as the consequences grow.

We do not need certainty that a bridge will collapse before inspecting it.

We do not need certainty that a bank will fail before stress-testing it.

And we do not need certainty that AI will eliminate a large share of human labor before seriously examining what would happen if it did.

The purpose is not to predict disaster. The purpose is to avoid discovering, after the fact, that we understood the risk decades earlier and still failed to prepare for it.

In 1964, America could reasonably say:

We do not yet know what these machines will eventually be capable of doing.

Sixty-two years later, that answer is becoming less comfortable.

The machines are here.

They are improving rapidly.

And we are deliberately trying to make them better.

The question is no longer whether technological progress should continue.

It is whether our capacity to think through its consequences is advancing anywhere near as quickly as the technology itself.

Endnotes

Lyndon B. Johnson, “Letter to the President of the Senate and to the Speaker of the House Proposing a National Commission on Automation.” March 4, 1964.
Ad Hoc Committee on the Triple Revolution, The Triple Revolution, 1964.
National Commission on Technology, Automation, and Economic Progress, Technology and the American Economy, 1966.
ERIC record for the National Commission report, Technology and the American Economy.
U.S. Government Accountability Office, “Unemployment Insurance: Transformation Needed to Address Program Design, Infrastructure, and Integrity Risks.”
Congressional Budget Office, budget and economic outlook materials, 2026.
Executive Office of the President, Artificial Intelligence, Automation, and the Economy, December 2016.
The White House, America's AI Action Plan, July 2025.
U.S. Department of Labor, announcement of AI-readiness hub funding, April 2, 2026.
Stanford Institute for Human-Centered Artificial Intelligence, 2026 AI Index Report — Economy.
United Kingdom Government, “Introducing the AI Economics Institute.” June 2026.
United Nations, Global Dialogue on Artificial Intelligence Governance.