Newsletter Archive
Listen to Our Podcast
Dear Aventine Readers,
While there is plenty of news to keep our attention riveted to the here and now, today we’re looking at a longer-term challenge: how will economies adapt to populations with fewer young people and more older people? It’s a fascinating and extremely broad question, which is why — at least for now — we are focusing on the role AI-driven technology might play. Will it accelerate productivity enough to compensate for a smaller workforce? Will some economies benefit more than others? We spoke to five experts in economics, demography and development with a wide variety of thoughts on the subject. Read on to find out more.
Also in this issue:
Thanks for reading and have lovely weekend,
Danielle
Subscribe
Subscribe to our newsletter and be kept up to date on upcoming Aventine projects
Boosting Productivity Amid Depopulation
As fertility rates around the world drop and people live longer, long-held demographic patterns are shifting, reshaping populations across the globe and raising pressing questions about how growth can be maintained while populations shrink.
Already more than two-thirds of people on the planet live in countries with fertility rates below 2.1 — the number of children per family required to maintain a steady population — according to the United Nations 2024 World Fertility Report. At the same time, life expectancy has increased to a global average of around 70 years, and closer to 80 years in Europe and the Americas — over double what it was less than 200 years ago — thanks to unprecedented advances in health care and living standards that have taken place over the last century.
Combined, these two phenomena are reshaping the world’s demographic makeup. Not too long ago the population breakdown of most countries resembled a pyramid, with large numbers of young people at the bottom and a decreasing number of ever-older people toward the top. Today, age distributions increasingly resemble rectangles, in which there are similar numbers of people in all age groups. Over time, unless fertility rates increase, the base will become narrower than the middle — which has already happened in some nations. Those demographic shifts, unaddressed, will have profound effects, with a growing elderly population requiring support from a dwindling number of working-age adults.
There could be upsides to this scenario: A shrinking population would require less food, water, energy and other resources, decreasing demands humans put on the planet. But the impact on economies could be devastating. “Absent action,” A recent study by McKinsey Global Institute reported, “younger people will inherit lower economic growth and shoulder the cost of more retirees, while the traditional flow of wealth between generations erodes.” What might this look like? As the economics writer Noah Smith summed it up: “In the 1990s and 2000s there were more than 5 working-age Americans (age 15–64) for every elderly American (64+). By 2021, there were fewer than 4. That means that the economic burden of supporting each elderly American is now shared among only 4 people instead of 5. And in other rich countries, it’s even worse. In France, there are only 3 working-age people for every elderly person. In Japan, there are only two.”
There is no silver bullet for this problem. Instead, there are several levers a society could pull to encourage continued economic growth. Governments could shape the demographic mix, through adjustments to immigration rules, say, or tax incentives to help increase fertility rates. They could try to increase labor intensity, a measure of the amount of hours worked per person, by raising the retirement age or relaxing rules around how many hours employees can be expected to work. Or they could attempt to boost productivity — the value of work done per hour — which is typically achieved by encouraging more effective use of technology.
Directly shaping the demographic will be difficult. In a recent The New Yorker article, Gideon Lewis-Kraus described efforts around the world to increase birth rates: “The world’s most lavishly pro-natalist governments spend a fortune on incentives and services, and have increased the fertility rate by approximately a fifth of a baby per woman,” he wrote.
It may also be hard to increase labor intensity. McKinsey’s report, for instance, points out that in regions in which the share of working-age people is already in decline, women and those over 50 years of age have already greatly increased their participation in the workforce, so there might not be much more to gain.
That leaves productivity. Can it be increased enough to counteract the effects of an impending demographic jam?
Historically, technology has enabled humans to significantly increase productivity — from the steam engines that drove the Industrial Revolution to the robots that automated the manufacturing of cars. And the potential impact of our changing demographics comes at a time when many proponents of artificial intelligence believe that the technology will dramatically boost human productivity in knowledge work. Other technologies such as robotics, autonomous vehicles and quantum computing, could also have a significant impact on how much humans can accomplish.
But technology’s role in raising productivity isn’t as straightforward as just turning an R&D switch to stimulate innovation. The last 20 years have seen huge, potentially productivity-boosting, technological shifts, with the mass adoption of smartphones, cloud computing and e-commerce, among countless other technologies. Yet productivity growth has fallen in many Western countries, including in the U.S., which saw average productivity growth of about 1.5 percent per year between 2005 and 2024, compared to a long-term average of about 2.1 percent. That paradox is an open research question among economists, and the phenomenon is influenced by many factors, including how technology is adopted, which areas of the economy it affects and how it influences spending.
To grasp the interplay between the world’s shifting demographics, their impact on the economy, and what role technology could play, Aventine spoke with experts in economics, demography and development, who are far from in agreement about how these interdependent forces will play out and what the ultimate effects will be.
There's a lot of hype about AI, but AI is [already] doing a lot. [One example I saw] in China, October 2023, was a warehouse of 1,000 square meters [where] they're using robots just with four people; the manager told me that it used to be 26 people [without a robot]. You can find many examples of [this kind of] automation. So productivity could increase with technological breakthroughs, and many of them we don't know [what they'll be yet]. One thing we do know is that we have faith in human ingenuity ... I don't know exactly what will arrive with this dawn of a new technological era of progress, but we have a lot of potential to increase productivity without having more people."
— Feng Wang, professor of sociology at University of California, Irvine
Proponents [of AI] who are like, ‘We're gonna have a productivity explosion and a growth explosion, economic growth is just gonna accelerate like crazy because of AI’ … I'm skeptical of that … There are elements of [deploying automation] in services that get hard. If my doctor is using AI to read my radiology charts and make a base diagnosis [in order] to make sure they don't miss stuff, great — it's kind of a quality check. But in the end, I want to sit there and actually have [a physician] tell me it'll be OK. And that kind of puts a ceiling on the productivity gains you can actually [achieve]. We've got productivity gains, it's just maybe not as wildly rapid as you could get in terms of, say, automating refrigerator production or car production. I don't think we're ever going to achieve that in services … [At the same time] whether the economy benefits a lot from these productivity gains in services depends on our response in terms of how we choose to spend our money … Like, refrigerators are insanely cheap. I’m a high consumption American, and I have three, but I don’t have, like, 10 … The point being that, in response to the prices coming down of, say, radiology, there's a scope for people to be like, ‘Oh, I'll go get that scan I didn't intend to go get before because it was really expensive.’ But as you automate and make some of these services more productive, they'll naturally start to get squeezed in terms of their share of economic activity, because the prices will be dropping faster than our use of them will be going up.”
— Dietrich Vollrath, a professor of economics at the University of Houston and author of the book Fully Grown: Why a Stagnant Economy Is a Sign of Success
Let's imagine … people want to retire at 60, and the birth rate was low, and so there's a shrinking proportion of the population that's between, say, 20 and 60. In some sense there would be a shortage of labor. But then simple economics would say that when there's a shortage of something, its price changes, and you'd expect wages to be higher than they otherwise would have been. Employers would then find that the mix of what you might call capital and labor switches toward using more capital-intensive means of production [such as machines and robots], as labor becomes scarcer and more expensive … There are naturally going to be a whole lot of economic reactions to that, including more incentives for people to work a bit longer, wages will change, how much labor to employ and how much expensive kit and machines [to buy and use] — all that stuff will adjust … The sort of doom-and-gloom merchants who say this is a crisis, I think, are just not thinking clearly enough about a lot of these adjustment mechanisms, nor are they thinking about the rather substantial advantages of there simply being fewer people. They're focusing on one particular thing, which is, ‘Oh gosh, there aren't enough people to work.’”
— David Miles, professor of financial economics at Imperial College, London a member of the U.K. government’s Budget Responsibility Committee and former Chief U.K. Economist at Morgan Stanley
I actually think that AI is going to drive people increasingly to spend a bigger share of their money on [non-tradable services], because the automated production systems that create shirts or cars or other things will be increasingly robotized and capital intensive, not labor intensive. So you might have pockets of very high productivity, very AI-enhanced production, but actually you're moving a bigger share of your economy towards lower productivity things. … Older people don't need as many things as younger people. You don't need to get a new home, you don't need to buy a car, people are reverting to dematerializing … and what you want to spend your money on is basically experience. And you don't want your massage to be quicker, you don't want people to look after your elderly parent more quickly, you don't want someone to cook a meal for you in a restaurant more quickly. You don't want to have an experience which is quicker, you want it, actually, slower. And that means that a bigger and bigger share of your economy is [based] on [non-tradable] services [that can only be purchased and consumed in the same location as where they are produced], which people don't want to increase the efficiency of. [Those services are] not tradable, you can't do it remotely, you can't outsource — and, incidentally, you can't outsource it to AI.”
— Ian Goldin, a professor of globalisation and development at the University of Oxford, former vice president of the World Bank and co-author of the review article “Why Is Productivity Slowing Down?” in the Journal of Economic Literature
I think the puzzle that sits in front of us is that where the absorption [of AI] could be the highest in terms of applying it might be contentious … The more inefficient your economy is, the faster it is possible, in theory, for you to grow. India, Vietnam, Mexico, you know, these kinds of places have the chance to see the biggest benefits from AI, because they are more inefficient, and they could jump up a few levels. [Compare that to] the United States or Germany, where, you know, we're pretty close to the frontier already, so improving your life dramatically [is harder] … So it's not clear that you necessarily have a win-win situation ... you can get a productivity increase that's global, but locally it creates tensions.”
— Manoj Pradhan, founder of the independent macroeconomic research firm Talking Heads Macroeconomics and coauthor of the book, The Great Demographic Reversal: Ageing Societies, Waning Inequality, and an Inflation Revival
Listen To Our Podcast
Learn about the past, present and future of artificial intelligence on our latest podcast, Humans vs Machines with Gary Marcus.
Advances That Matter
Google’s new AI scientist is proving itself in labs. A new research tool developed by Google is designed to help scientists develop hypotheses and accelerate the rate of discovery, and early results from tests performed in collaboration with researchers from Stanford University, Imperial College, London and Houston Methodist hospital suggest it could prove valuable. The software is based on a collection of AI agents that work together. For example, one generates new ideas, another assesses them for correctness and quality, and another ranks those ideas on their novelty, correctness, and testability. Scientists interact with the system using natural language, giving it a germ of an idea, say, or providing feedback on its suggestions. Ultimately, the AI comes up with a research plan containing a list of ideas for scientists to test. In one example, the system proposed how an existing molecule — KIRA6, which has previously shown promise for the treatment of retinal degeneration and diabetes — could be repurposed for the treatment of acute myeloid leukemia. That hypothesis was borne out in lab tests that confirmed that the drug can inhibit a tumor's ability to grow and spread. In another, more complex test, the software was instructed to hypothesize around a set of features that contribute to genetic evolution that causes antimicrobial resistance. The tool proposed a novel molecular mechanism of gene transfer in just days — a hypothesis that had taken researchers years to develop, and which is only now being proven in a soon-to-be-published paper. To be clear, the tool still has limitations: It can struggle to parse non-text data, for instance, and it is still susceptible to hallucinations like other large language model-based technology. Google’s tool joins several other AI models developed for technical disciplines currently in early testing, including OpenAI’s Deep Research, as well as tools being developed by DeepMind and the pharmaceutical company BioNTech. But the use of Google’s model in active research settings is a genuine step forward in using AI to accelerate scientific discovery.
Chinese biotechs are on a drug development tear. The Chinese biotech industry used to be best known for producing large quantities of generic drugs. But, The Economist reports, the country’s biotech firms are increasingly at the forefront of drug development, discovering and commercializing therapies faster than their Western counterparts and doing it more affordably. Some of this is truly innovative research, and some of it is based on a so-called fast-follower approach, in which known drugs are tweaked to make them safer or more effective. Of all the drugs currently in the development pipeline in China, 42 percent are either fast-follower or entirely novel therapies, according to The Economist. The boom is due to a confluence of factors. About a decade ago, China’s drug regulator cleared its approvals backlog and streamlined the rules for clinical trials, aligning them with global standards and making it much faster to get new therapies into testing. Around the same time, many Chinese nationals who had moved overseas for education or employment returned home in response to government initiatives and the rapid growth of the domestic economy, reshoring important skills and expertise. Meanwhile, private funding per year for Chinese biotech increased by more than a factor of 10 between 2016 and 2021. The result is a wave of new drugs that are now being licensed by Western companies for hundreds of millions of dollars each, causing a surge in the total value of drugs licensed in China — almost doubling from 2023 to 2024. So far, trade restrictions being put in place by the U.S. government don’t extend to biotechnology products; if that changes, it would obviously impact this nascent success story.
AI can help track poverty. (Opinions are divided on whether it should.) Assessing poverty, especially on a global scale, is hard work: Labor-intensive in-person surveys are typically required to gather data on things such as how much money a household has to spend, diet, schooling, access to drinking water and so on. That information is then used to calculate metrics that indicate how poor or wealthy different areas are relative to one another, and those results can be used to determine how best to target support. In addition to being expensive and time-consuming, the data can be flawed, underreporting, for example, on families who might not have permanent addresses and could be at the bottom of the income distribution. Additionally, the way metrics are interpreted can be subjective, which can lead to conflicting assessments of who is most in need. As an economical alternative, poverty researchers are increasingly using AI, Nature reports, and it can work surprisingly well. AI analysis of satellite imagery capturing road conditions, green space and the amount of street lighting at night (among other factors) can predict poverty as well as in-person surveys, according to the study. AI can also track mobile phone use, using the frequency of calls or the use of mobile money transactions to identify poverty. This approach, too, has proved to be more accurate than in-person surveys, according to another study published in Nature. So far, the techniques have already been used to direct assistance in Togo, Mozambique and Nigeria, and while the approaches are considered to be faster and cheaper than alternatives, critics worry that they could be abused. The concerns are familiar: that AI can often be inherently biased; that it can encroach on people's privacy; that there isn’t enough evidence yet that it is robust enough as a tool to take the place of existing in-person approaches. For now, though, the reality may be that AI will be used when alternatives are too expensive and time consuming.
Magazine and Journal Articles Worthy of Your Time
Can We Build a Five Gigawatt Data Center? from Asterisk
3,800 words, or about 15 minutes
If progress in building ever-larger AI models continues at its current pace and we don’t figure out ways to spread the training of models across disparate data centers or make models more computationally efficient, predictions suggest that by 2030 the most advanced large language models will need data centers that require five gigawatts of power. As this story explains, that is a daunting prospect: 5GW is about 10 times what today’s biggest data centers require, and about the same demand as the whole of New York City. Building such a data center isn’t just challenging; it might not even be possible for reasons that go beyond the power demands. For anyone with deep enough pockets to fund such an endeavor — which might require capital expense of as much as $100 billion — there are significant problems to solve around the availability of chips, the likelihood of getting appropriate permitting, and, oh yes, finding the equivalent of about five nuclear power plants to keep the thing running.
Can the nuclear industry find a better way to build? from the Financial Times
2,600 words, or about 10 minutes
Speaking of nuclear power: After decades of opposition, the tides are turning for the technology as the demand for clean energy spikes, thanks to the proliferation of data centers and the powering of electric vehicle fleets, buildings and industry. But if nuclear power is to come roaring back in the way that many of its proponents hope, the construction of new plants needs to overcome the difficulties that have plagued recent projects — namely, massive delays and shocking budget overruns. This story suggests there may be hope because the sector is increasingly attempting to streamline the process of building such facilities by standardizing projects as much as possible. For example, two projects in the U.K. led by the French energy provider EDF are separated by almost 300 miles but based on pretty much the same blueprints. The U.S.-based reactor builder, Westinghouse Electric, now refuses to modify its hardware for individual installations. Given how long it takes for nuclear plants to be built, it will be years until we can measure the effectiveness of these new approaches, but the examples show that the nuclear industry realizes that cost is now the greatest barrier to increased adoption.
Adventures in the genetic time machine, from MIT Technology Review
4,100 words, or about 17 minutes
The oldest DNA ever to be studied so far is 2.4 million years old, a sample that came from Greenland and revealed that the land it was taken from was once forest. But as this story describes, genetic samples from the past don’t just promise to help us understand history — they could also shape the future. Scientists around the world are investigating ancient DNA in humans, animals and plants, and in the process shedding light on how exactly evolution got us to where we are. Such projects are unearthing evidence in ancient DNA that hints at the origins of modern diseases like diabetes and multiple sclerosis, which — they hope — could lead to insights about how to combat them. Others are taking genetic variations from plants that existed millions of years ago in hotter climates and incorporating them into modern crops, hoping to enable them to be more robust in the face of rising temperatures. Bringing an extinct species back to life à la “Jurassic Park” is still a ways off — even though it’s definitely on the minds of some researchers — but this story makes it pretty clear that the genetics of the past could have a very modern role to play.