You Don’t Have to Become an AI Engineer

What Actually Happens When Technology Eliminates Work

Discussions about artificial intelligence and employment frequently contain a hidden assumption: AI will eliminate existing jobs, so to maintain employment, we must create an equivalent number of new kinds of jobs. That assumption naturally leads to attempts to imagine the occupations of the future. Perhaps we will need AI engineers, robot supervisors, synthetic-media specialists, virtual-world designers, and jobs with names we haven't invented yet.

Some of these occupations undoubtedly will emerge. But history suggests that we may be asking the wrong question. When technology eliminates enormous quantities of human labor, displaced workers do not necessarily migrate into jobs created by that technology. They frequently just do other things humans already know how to do.

Look Around

Consider some of the occupations employing millions of Americans today. Truck drivers move things. Construction workers build things. Cooks prepare food. Waiters serve it. Mechanics repair machines. Security guards protect property. Nurses care for people. Teachers teach. Salespeople sell. Artists create. Journalists gather and communicate information. Farmers grow food.

These aren't futuristic occupations. A person transported from 1850 would be bewildered by a modern semiconductor fabrication plant or software company. But explain that someone is a cook, builder, farmer, guard, teacher, merchant, or driver, and the underlying activity would be immediately understandable.

The technology surrounding these occupations has changed enormously. The human purposes have not. That distinction matters when we try to predict what AI will do to employment.

The Wrong Model of Technological Change

We tend to imagine technological displacement as a simple sequence: Old job → automation → new technology job A farm worker becomes a tractor mechanic. A factory worker becomes a computer programmer. An office administrator becomes an AI engineer. Sometimes that happens. But it isn't how most technological change has worked.

A better model is: Technology increases productivity in one activity → fewer people are needed to perform it → labor and resources become available for everything else. Some of that "everything else" will be new. Much of it won't be. Agriculture provides perhaps the clearest historical example.

Where Did All the Farmers Go?

Around 1800, roughly three-quarters of the American labor force was engaged in agriculture. Today, depending upon exactly how agricultural employment is defined, something around 1 to 2% is. This was an extraordinary reduction in a category of human labor. If the agricultural share of employment had remained anywhere close to its 1800 level, tens of millions more Americans would be working in agriculture today. 

They aren't. But those missing agricultural workers didn't all become agricultural technologists. Over successive generations, people became factory workers, construction workers, truck drivers, mechanics, salespeople, accountants, teachers, nurses, cooks, managers, entertainers, and thousands of other things.

Some of those occupations were new. Many were ancient. What changed was not necessarily the existence of the occupation. What changed was the amount society could afford to demand of it.

We Got More Restaurants

Consider something as ordinary as cooking. Technology has transformed food production. Agricultural productivity exploded. Transportation became faster and cheaper. Refrigeration reduced spoilage. Industrial food processing expanded. Modern ovens, mixers, dishwashers, and countless other machines reduced the labor required to prepare food. If human wants were fixed, all of this should have dramatically reduced the need for cooks. Instead, wealthy societies developed enormous restaurant industries.

We didn't simply use technology to prepare the same quantity and variety of food with fewer people. We demanded more choices, more convenience, more restaurants, more prepared food, and more experiences surrounding food. Technology reduced the cost of satisfying one level of demand. Human beings responded by creating another.

We Got More Entertainment

Entertainment provides an even more striking example. Recorded music made it possible for one musical performance to be reproduced millions of times. Radio expanded that capability. Television expanded it further. Digital reproduction eventually drove the marginal cost of distributing a song, photograph, article, or video toward zero.

A simple technological-displacement model might have predicted the destruction of entertainment employment. Instead, modern humans consume staggering quantities of entertainment.

Movies, television, streaming video, podcasts, video games, professional sports, recorded music, live music, social media, online video, photography, books, and news all compete for people's time and attention. The technology didn't merely substitute machines for entertainers. It helped create an enormous expansion in the quantity and variety of entertainment demanded.

Productivity Changes Relative Prices

This points toward the more important economic mechanism. Technology doesn't simply "destroy jobs." It changes the relative cost of doing things. Suppose AI eventually allows one lawyer to accomplish work that currently requires five lawyers. The immediate effect is easy to see: less legal labor is required for the same amount of legal work. That is the part everyone worries about. But the next effects are harder to see.

Legal services may become cheaper. People who previously couldn't afford legal assistance may buy it. Companies may perform legal analysis that wasn't previously worth doing. Some lawyers may move into other occupations. Businesses may spend the money they save on something else, while consumers may also redirect their savings. Capital may migrate elsewhere, and new businesses may become economically viable.

The consequences spread through millions of individual decisions. There is no requirement that four displaced lawyers become AI specialists. One might start a company. Another might teach. Another might enter sales. Another might manage a restaurant. Another might work in an occupation that doesn't exist yet.

Or legal demand itself might expand enough that fewer lawyers disappear than the original productivity calculation suggested. This is why technological forecasting is so difficult. We can see the first-order substitution. We can't easily see the millions of second- and third-order responses.

Existing Occupations Can Expand

This possibility receives far too little attention in discussions about AI. Suppose AI dramatically reduces the labor required for accounting, administration, software development, legal research, graphic production, and routine analysis. Why must the workers released from those activities move into technologically advanced occupations?

Perhaps we simply have more teachers, coaches, personal trainers, nurses, physical therapists, construction workers, restaurant workers, salespeople, caregivers, tour guides, entertainers, craftspeople, and entrepreneurs. There are also countless services that are currently too expensive for most people to purchase.

A society with dramatically higher productivity can afford to devote more human effort to activities it previously treated as luxuries. That is exactly what rising wealth has repeatedly done.

Some Human Activities Are Remarkably Persistent

There is another reason many old occupations survive. They involve characteristics that are difficult to separate from physical human existence: location, dexterity, trust, responsibility, personal relationships, physical environments, human taste, social status, authenticity, and human preference. People don't necessarily want every activity performed in the cheapest technologically possible manner.

A machine may eventually be capable of preparing an excellent meal. That doesn't necessarily eliminate restaurants. AI can already generate music. That doesn't mean people cease caring about human musicians. AI can generate text. That doesn't necessarily eliminate the value of talking to another human being. Automation capability and human demand are different questions.

The relevant question isn't merely: Can a machine perform this task? It is: Once machines can perform this task, what will humans still value enough to pay other humans to do? Those are very different questions.

We Don't Need 50 Million Prompt Engineers

This substantially changes the employment question surrounding AI. Imagine, purely for illustration, that AI eventually displaces 50 million jobs or major portions of those jobs. A common response is: Where are the 50 million new AI jobs?

There don't have to be 50 million AI jobs. There weren't tens of millions of new agricultural-technology jobs to absorb America's disappearing agricultural workforce. Labor spread throughout the economy. Some people entered new industries. Others entered very old ones. Existing industries also expanded as productivity and wealth increased. The same could happen with AI.

Perhaps AI creates five million genuinely new kinds of jobs. Perhaps it creates twenty million. Perhaps far fewer. We don't know. But the number of newly invented occupations is not the relevant constraint on future employment. People don't need technologically novel jobs. They need economically valuable things to do. Those are not remotely the same thing.

Most of Us Still Do Very Old Things

This is perhaps the most underappreciated fact about a technologically advanced society. Walk around modern America and observe what people actually do. Someone fixes your car. Someone delivers a package. Someone cuts your hair. Someone prepares your lunch. Someone teaches your child. Someone examines your shoulder. Someone installs your roof. Someone sells you a house. Someone drives a truck carrying your groceries. Someone performs music. Someone watches a building overnight. Someone cleans an office. Someone builds a road. Someone takes care of an elderly parent.

The equipment surrounding these people would astonish someone from 1850. The activities often wouldn't. After two centuries of extraordinary technological progress, human society still employs enormous numbers of people doing fundamentally ancient things. That should tell us something important about AI.

Technology Doesn't Decide What Humans Do Next

Technological progress eliminates constraints. It doesn't dictate how the resources released from those constraints must subsequently be used. Agricultural technology meant fewer humans had to grow food. It did not determine what those humans would do instead.

Industrial technology meant fewer people were required to manufacture many goods. It did not determine what everyone else would value. Computers eliminated enormous amounts of clerical calculation. They didn't determine how the time and money saved would ultimately be spent. AI may follow the same pattern.

It may substantially reduce the amount of human cognitive labor required to accomplish many existing objectives. That is enormously consequential. But it doesn't follow that human beings therefore run out of objectives.

The Question We Should Be Asking

When considering AI and employment, we naturally ask: What new jobs will AI create? History suggests a better question: What will humans choose to do more of when AI makes other things dramatically cheaper?

Some answers will involve technologies that don't yet exist. Some will involve occupations we can't presently imagine. But many may be remarkably familiar. We may build more things, travel more, eat out more, exercise more, receive more personal care, produce more entertainment, teach people more, repair and improve more physical things, create more businesses, or demand services that today only wealthy people can afford.

We may also develop entirely different preferences as our resources increase. We don't know. And that's the point. The farmer standing in an American field in 1800 didn't need to predict software engineering to believe that his descendants could remain economically useful. Nor did all of his descendants become technologists.

Some became cooks. Some became builders. Some became mechanics. Some became salespeople. Some became artists. Some became drivers. Some became doctors. Some became farmers. Two centuries of technological advancement later, humans still do all of those things.

The great mistake in thinking about AI may therefore be assuming that futuristic technology requires futuristic employment. It doesn't. Technology changes what is scarce. Humans decide what becomes valuable next.