What 1,000 years of history can teach us about making the economy of AI work for everyone

Two economists on why progress requires more than technological innovation alone

It’s tempting to imagine progress strictly as a technological science, where both its history and future are stories of humanity becoming more prosperous one major innovation at a time. But that would be a mistake, according to MIT economists Daron Acemoglu and Simon Johnson in their new book, Power and Progress: Our 1000-Year Struggle Over Technology & Prosperity.

The book traverses a millennium of disruptive technologies, from medieval agriculture and ship design all the way up to the foothills of generative AI. Along the way, they argue that innovation has proven just as likely to cause misery as it has prosperity. On their list of examples is the cotton gin, which turned the United States into the world’s largest cotton exporter, while also deepening a system of enslavement that spread across the American South. When prosperity does win out, it’s usually because citizens organized to demand more equitable arrangements than those enriching a narrow elite.

“A thousand years of history and contemporary evidence make one thing abundantly clear: there is nothing automatic about new technologies bringing widespread prosperity. Whether they do or not is an economic, social, and political choice,” they write.

They are no more sanguine about the future of the next great innovation: AI. In their view, translating innovation around AI into shared prosperity will require countervailing powers, like a labor movement demanding worker-friendly automation, and a civil society movement upholding a more egalitarian vision for the kind of society AI can help create.

“We’re addressing the core problem that no one else has talked about: It’s about the vision that drives how you imagine the future,” Johnson told me. “And a lot of our argumentation and policy proposals are designed to help people understand how easy it would be to change that vision.”

I spoke with Acemoglu and Johnson about what choices steered innovation toward progress in the past, and what specific policies their trek through history suggests we might consider today. A transcript, edited for length and clarity, follows.

Your book argues that technological progress, alone, does not automatically lead to shared prosperity. What else is necessary to turn innovation toward the common good?

Daron Acemoglu: The first key ingredient that we argue for technological advances translating into something resembling shared progress is that their direction should not be just automation, or sidelining humans. That’s critical for keeping humans in the loop and increasing their contribution to production, and an institutional structure that enables them to actually get the returns out of that. And second, vision, which is critical, because we argue there’s nothing inevitable about any of those things. Both technologies and institutions are shaped by the visions of powerful actors as well as other political balances.

You focus on “vision” throughout the book, but it’s sort of a nebulous idea. Are there any historical examples that show how much of an impact guiding visions can have?

Simon Johnson: The reason we start in the medieval period is because we’re quite convinced that the so-called Dark Ages were not particularly dark from the point of view of creativity and innovation. From 1,000 years ago, they were actually very creative, with lots of inventions across agriculture and commerce. But the prevailing vision of that medieval period was one where you had a small elite who argued that they had a divinely endowed power, and took pretty much all the proceeds of that higher productivity and put them into monumental cathedrals.

Those cathedrals did not increase productivity; they didn’t improve public health. They were symbols of oppression. You can argue that it was an alternative vision breaking away from that medieval, religious-oriented, top-down control vision that was necessary, though not sufficient, for the beginnings of the modern era and industrialization.

What struck me about your chapter on the Industrial Revolution was the huge gap between the introduction of new technologies and the beginnings of anything that resembled shared prosperity. You argue that for the better part of a century, many people’s lives just became more miserable.

Simon Johnson: You can argue about when exactly the Industrial Revolution got started. I like the 1720s because that’s when the first big silk mill was built just outside Derby [an English city]. That began putting people into factories with machines that were controlled by an employer. So the 1720s is a good starting point.

Young children were working 18 hours a day pushing coal carts with their heads deep underground. We know that was happening in the 1840s because that was a matter of investigation by royal commission. Because it wasn’t illegal, everyone involved was quite candid about it, and said, “Look, that’s what you need in order for a coal industry to exist.”

So that’s 120 years where you cannot say that these 6-year-old children were living better. Some people argue a little about wages, but living conditions and public health in those cities were dreadful. And after the 1840s, there was a shift in thinking. It wasn’t particularly altruistic, it was more “My god, we have infectious disease rampant in Manchester because there’s no toilets. What are we going to do about it?” Consequently, there was a reimagining of how technology could be applied, including the modern sanitation movement, which was by far the No. 1 breakthrough in the use of technology in the 19th century. And this coincides with trade unions beginning to get organized and pressure on the political elite to allow wages to rise.

So if I said to you that generative AI is here, and that you and your families will be better off in 120 years, I think people should be fairly unsatisfied with that. Why do we have to wait so long?

Daron Acemoglu: It’s sort of remarkable how consistent this view is among many economists, policymakers, and even the Democratic Party: when you have better technologies, the costs are “transitional.” What that encapsulates is that there’s often an implicit belief that [shared prosperity] is automatic, but it might take time.

The biggest target for Simon and I is that there is nothing automatic about it. But the automatic view gives you a real sense of comfort. Wealth inequality may be horrible, democracy may be in a difficult position today, generative AI may create lots of disruptions, but we’ll work it out.

So if there was nothing automatic about that long period of industrial-born misery eventually turning into shared prosperity, why did it eventually begin to shift for the better?

Daron Acemoglu: Cities and factories created huge amounts of misery, but they also changed how easy it was for people to organize. Once hundreds of thousands of working people were concentrated in cities and workplaces, the demand for representation grew difficult to turn down.

But there were other factors, like the redirection of technological change. The next phase of industrial technology in heavy industry like chemicals and steel opened up new opportunities for investing in human skills. American technologies had to prioritize making unskilled workers more productive, and once that got started it spread around the world.

You make a distinction between two kinds of automation. One, “so-so automation,” just replaces workers outright. The other, “machine usefulness,” either complements their skills or creates new tasks for workers. You argue that we should aim for the latter — what’s an example?

Simon Johnson: The central example is when Ford comes to Detroit and takes on car production. In 1900, the US car industry produced about 3,500 cars a year, mostly artisanal. Henry Ford put car production on the assembly line and increased productivity more than 100-fold. He also, of course, automated many of the jobs that had previously been done by those artisans. However, what he did, along with the managers and engineers and suppliers and consumers, was create this enormous industry. By the end of the 1920s, the US was making between 2 and 3 million cars a year, employing 400,000 people. Most of those people had tasks, which led to jobs that had never been done by any human ever. And by the way, unions became stronger and pushed for wages.

So we are in no way opposed to automation. We are encouraging the seeking out and development of those human-complementary innovations and uses of machines, because it’s that increased demand for labor that is the heart of high wages and shared prosperity.

Labor movements and unions were the basis of those countervailing social powers in 19th-century America. Today, despite some high-profile media organizing and strike waves, union membership remains at an all-time low in the US. What role do you see for the labor movement going forward?

Daron Acemoglu: The future of the labor movement is open. We are convinced you need workers’ voice. It’s not good when AI regulation is discussed by senators and the CEOs of the chief tech companies and nobody else. And there are many things that are wrong. Sectoral unions would be better [than individual workplace unions, as is common in the US], but you might need a broader civil society movement to complement the labor movement, and the labor movement itself needs to find new organizational forms.

You surveyed 1,000 years of history in order to argue that AI will not automatically lead to shared prosperity and that we’ll need specific policy interventions to achieve that. Could you each share one policy that you’ve come to believe should be a part of that debate?

Daron Acemoglu: One that I will put on the table is evening out the taxation of labor and capital. Our tax code creates artificial incentives for firms to use capital instead of labor. You can have bipartisan support if it’s presented the right way: not taxing businesses more, but trying to create more opportunities for labor. Corporate income taxes would be one channel. But first, I would start with removing some of the most aggressive depreciation allowances [a tax deduction that allows businesses to recover the annual cost of property or equipment use] which essentially enable firms to write off a lot of their digital and equipment investments.

Simon Johnson: I’ll suggest two. One is surveillance. In the case of employment, if surveillance actually makes your life more stressful, and makes you more likely to cause an accident or injure yourself, that is something that falls within the sphere of reasonable regulation. We should consider negotiating safeguards on surveillance across G7 allies and other industrial democracies.

The second one is: Show me the new tasks. How do we get more invention in that direction? We know how to do this; we’ve seen it many times since 1940 in the United States. You put some federal government money in.

The interesting thing about federal money is that it’s catalytic; you don’t have to put that much money in. If you look at the Human Genome Project, for example, which was turned down by venture capitalists in the 1980s because they said, “Hey, great idea, but we don’t know how to benefit from it because it will be general knowledge,” it became a government-funded project. It cost about $10 billion, creating an industry that employed 200,000 people, and changed the world repeatedly. So federal money applied in this strategic, purposeful manner, can change everything.

Their book, Power and Progress: Our 1000-Year Struggle Over Technology & Prosperity, is available now.