Which matters more for the economy — babies or robots?

Two sweeping visions of the future have been unfolding, each producing grim — yet seemingly contradictory — predictions for the fate of humanity.  

On the one hand, we’re learning that the birth rate is falling all over the world, leading to aging societies and a global population set to decline this century. If trends continue on their present path, demographers warn, there won’t be enough people to work to support society. The extreme labor shortages would lead to stagnation, poverty, and ultimately — in the most dire scenarios — the collapse of civilization itself.

On the other hand, there are repeated warnings that artificial intelligence could take most, or even all, jobs. Anthropic CEO Dario Amodei recently predicted that AI would eliminate 50 percent of entry-level white-collar jobs within the next five years. Though other AI leaders are more skeptical about such sweeping automation, the International Monetary Fund did find that between 2010 and 2021, the US regions that adopted AI most quickly  saw larger drops in employment rates, with men and workers in manufacturing and service jobs hit hardest. 

What happens if we’re short on both workers and jobs? Can both be true at once? And if they cancel each other out, does that mean we don’t need to worry? 

Many researchers studying these topics simply do not engage with one another — whether because of disciplinary silos that reward specialization, or timeline mismatches that make collaboration feel irrelevant. Demographers think in decades while technologists think in years, business leaders navigate quarterly earnings, and economists toggle between immediate policy concerns and long-term growth models.

The reality is also that researchers are operating under extraordinary uncertainty. We don’t know yet whether AI will complement workers or replace them, whether displaced workers will retrain successfully as in past transitions, or how aging populations will drive policy responses. This makes it easier to focus on more narrow predictions than to attempt forecasts that span multiple unknowns. 

But these conversations are closely intertwined, and, even under uncertainty, there are clues to what we can and can’t know.

Population growth vs. economic growth

About five years ago, Joseph Davis, Vanguard’s chief global economist, started fielding questions from investors that he didn’t quite know how to answer. With the economy changing in unfamiliar ways — from an aging workforce to booming tech stocks — how should they think about where to put their money? Should they be bracing for long-term inflation? Should they just follow the momentum and buy into tech giants like Amazon and Nvidia?

Davis, responsible for guiding Vanguard’s 50 million investors, couldn’t find anyone systematically studying how tech and population trends might interact — so he decided to do the research himself. The effort resulted in a working paper focused on how demographics, budget deficits, and globalization have shaped the US economy over the last century. “It was a humbling experience,” Davis, who recently turned that research into a book, told me. 

One of his clearest conclusions is that long-run economic progress does not depend primarily on population size. Using a model built on 130 years of economic data, he finds that changes in population growth have almost no meaningful correlation with GDP or inflation. Instead, the biggest gains in living standards have come during periods of major innovation — like the electrification of the 1920s or the rise of personal computing in the 1990s — regardless of population trends. 

Davis pointed to historical periods — like the Renaissance and the Roaring Twenties — when population growth was actually slowing, yet economic output surged. “Population growth slowed during the 1920s — we cut immigration by 90 percent. But growth accelerated anyway,” he said. 

By contrast, eras with strong population growth but weak economic productivity, like the 1970s, produced little real progress. 

“Demographics matter,” he told me. “It’s just that it can’t be looked at in a vacuum.”

The fear that aging societies are destined for decline is widespread — but it’s not well supported by the evidence. Davis noted that aging can be linked to increased long-term investment in technology and infrastructure, pointing to countries like Japan and Germany. These nations show that, while shrinking working-age populations can strain public budgets through rising health care and pension costs, and make it harder for businesses to find workers, they don’t inherently lead to economic disaster. 

Dean Spears, the co-author of After the Spike, a new book on population decline, also argues that concerns about aging societies lacking enough workers may be overstated in an era of technological change. 

“Aging isn’t what we emphasize in our book, because we don’t think it’s the most important thing,” he told me. “If AI is able to make output per worker greater…then with fewer workers, you could make up for the fact that there are fewer workers per population.”

Spears doesn’t think that aging is irrelevant, as fertility rates and the age breakdown of a population shape budgets, taxes, and public services. “If you’re the finance minister,” he said, “it certainly matters.” But he sees aging as a policy challenge, not an existential threat. The long-term trajectory of a society, he said, will depend far more on productivity, innovation, and how well a society’s systems and programs actually work. 

Neil Thompson, the director of MIT’s FutureTech research project, agrees. “Changes in AI capabilities and what they mean for both the ability to augment productivity of human labor and to fully automate some tasks are happening so much faster and will have so much bigger effects than demographic changes,” he told me. 

So, will AI make us more productive?

The question, then, is whether AI will actually boost productivity enough to offset a shrinking population.

Davis, of Vanguard, ran thousands of economic simulations, and the results kept coming back split. While the long-term effects are hard to predict, his simulations point to two futures over the next decade — a “tug-of-war” between the productivity gains AI could deliver and the fiscal strains posed by aging populations and rising public debt. In the first, which he gives a 45–55 percent probability, AI becomes a “general-purpose technology” like electricity, driving substantial productivity growth. 

The confusion surrounding how AI affects productivity extends far beyond academic circles.

In the second, with a 30–40 percent chance, AI proves incremental — useful but not transformative enough to counteract rising deficits and an aging workforce. In this scenario, the bleaker forecasts of demographers — that a shrinking  number of workers will cripple the economy — are more likely to be true. “I wish the odds [for growth] were higher,” Davis told me, adding that much of it will depend on other policy choices governments make, especially when it comes to deficits.

That same uncertainty is reflected in differing views between two leading economists. Daron Acemoglu, who won the Nobel Prize in 2024, estimates AI will automate only about 5 percent of work tasks profitably over the next decade, producing modest GDP gains. Without active policy intervention, he warns, AI will primarily replace workers rather than augment them.

Erik Brynjolfsson, a Stanford Univeresity economist, is more optimistic, believing AI could potentially push annual productivity growth about a full percentage point if it amplifies rather than replaces human work.                                                                                                                                                                                                    

A growing recognition of uncertainty

The confusion surrounding how AI affects productivity extends far beyond academic circles. Anthropic just launched a research program to study AI’s economic impact — a tacit admission that even AI developers don’t fully understand what they’re unleashing. 

The US Bureau of Labor Statistics (BLS) also only recently began incorporating AI impacts into its employment projections. In an analysis published in February, the agency takes a deliberately cautious approach, emphasizing that technological change doesn’t automatically translate to job losses. Some roles may shrink, particularly those involving highly standardized tasks like insurance claims processing, while others could grow due to new AI-driven demands or the continued need for human oversight.  

Even experts and advocates deeply versed in related fields acknowledge the limits of current understanding. “I’m not really a labor economist,” Spears said when I asked about links between AI’s economic impact and falling birth rates. Lyman Stone, a demographer and director of the Pronatalism Initiative at the Institute for Family Studies, told me he has not looked specifically into questions of workforce automation and depopulation. 

Malcolm Collins, a pronatalism advocate and former tech entrepreneur also lacks a clear idea of what these colliding trends might mean. “It might be that governments can still make the math work just by having lots of people, or it might be that AI really does replace all jobs and it becomes irrelevant how many people exist within a country,” he wrote by email. “I want to believe that humanity will always have some sort of differential role as an economic actor, but I will be honest that is only hope on my part, and I see no reason why AI would not replace almost all human jobs.”

Yet even as AI advances, many of the fastest-growing occupations in America remain distinctly human-centered. The BLS, for example, projects 21 percent growth for home-health and personal-care aides between 2023–’33. McKinsey estimates that AI could automate tasks equivalent to 11 million full-time US jobs by 2030, but surging demand in care work, green technology, and STEM fields still leaves net hiring needs of around 4 million workers. 

Those fastest-growing jobs, it turns out, are often the hardest for machines to replicate. There’s still a lot of separate discussion for now but the conversations won’t stay separate forever. Eventually, economic and demographic debates will have to converge.