Joshua Blumenstock has spent much of his career focused on easing the pain of poverty, project by project, in countries such as Togo, Afghanistan and Bangladesh. But with the advance of AI-driven machine learning, the Berkeley economist has set his sights on a more holistic goal: calculating what it would cost to eradicate extreme poverty worldwide.
The answer, according to a new working paper: $318 billion per year, or 0.3% of global gross domestic product, would be enough to bring hundreds of millions of people out of extreme poverty. That's an enormous sum, and yet Blumenstock and his co-authors found that it's dwarfed by the amount humanity spends every year on alcohol or cosmetics.
"Our hope is to try and get to a more realistic estimate of the cost of eradicating extreme poverty through direct cash transfers, so that that lack of realism is no longer an excuse for not taking action," said Blumenstock, co-director of UC Berkeley's Center for Effective Global Action (CEGA). "We now have a number that's as accurate as possible. And it's really not that big, which is encouraging."
"The numbers tell us it isn't crazy to set our sights on a big, ambitious goal," said co-author Paul Niehaus, professor of economics at UC San Diego. "That's partly because extreme poverty has already fallen so much in recent decades … but also because advances in data science have made it possible to design shovel-ready policies that get the most help to the poorest people."
The research is innovative not just for the estimated cost, but also for the use of AI-driven machine learning statistical analysis to solve such a profound human challenge. That powerful technology can help policymakers identify the world's poorest people based on detailed surveys of living conditions, spending habits and other granular details of assets and poverty. The application of machine learning devised by the researchers allows for unprecedented precision in assessing how much aid would be needed to bring people up to $2.15 in daily income, the benchmark generally used by global bodies to define extreme poverty.
The working paper, "What Would It Cost to End Extreme Poverty?" was published online in December by the National Bureau of Economic Research.
Blumenstock is a professor at the UC Berkeley School of Information and the Goldman School of Public Policy. In addition to Niehaus, other co-authors are CEGA data scientist Leo Selker and Roshni Sahoo and Stefan Wager, both at Stanford University.
In all, the researchers studied poverty in 23 low-income countries for which detailed economic data is available at the household and individual level. Extreme poverty rates in those countries, accounting for about half of the world's poorest people, could be reduced to 1% of the population or less with an annual investment of $170 billion per year, they concluded.
They then scaled that to the rest of the world's poorest countries, concluding that an annual investment of just 0.3% of global annual economic output - $318 billion - could all but eradicate such poverty. By comparison, they wrote, that's somewhat more than was spent on foreign aid in recent years. But it's a fraction of the annual global spending on alcohol, which amounts to 2.2% of the world's total economic output, or on cosmetics, which totals 0.6% of total output.
"This potentially makes ending poverty outright far more affordable than it has ever been in the past," they write on the CEGA website.
A longtime goal, with no clear path
The elimination of extreme poverty has long been a goal for development agencies and religious and civil society organizations. Ending poverty by 2030 is listed No. 1 in the U.N.'s Sustainable Development Goals, and the goal has been recognized in the foreign policy of some Western nations, too. Major financial bodies track the persistence of global poverty; in 2022, the World Bank raised the benchmark for extreme poverty to $2.15 per person per day.
Even if it's a moment where the cards might seem stacked against us, I think that we like to be optimistic.
Joshua Blumenstock
Effective domestic policy in low-income countries, in concert with economic growth and aid programs, have dramatically reduced extreme poverty - from 41% of people worldwide in 1981 to 8% in 2024.
But improvement has slowed. In recent years, voters in the U.S. and other nations have rebelled at paying for foreign aid. How, then, to drive continued progress?
Many experts have called for a universal basic income, with policy designed to assure that no individuals or families fall beneath a set floor. But that's an inexact way to provide aid. Because the uniform payments would give some households more than they need to rise above the poverty line, the approach can be very expensive, Blumenstock and his co-authors write.
Now, with the evolution of advanced machine learning tools, policymakers have new options for calibrating the aid sent to each household, to more precisely match its needs.
Even countries handicapped by persistent poverty are now gathering fine-grained data, using lengthy surveys to assess household income, land ownership, family size, education, food consumption, characteristics of home construction and other details.
Making aid more precise, more efficient
The new research shows that, with the rapid evolution of AI-driven machine learning, policymakers and aid agencies can do deep statistical analysis based on existing survey data to more precisely assess need and target aid.
Blumenstock cites the example of Togo, a small country in West Africa that has worked closely with CEGA. Its Harmonized Survey on Households Living Standards from 2019 collected rich information on the living conditions of roughly 6,000 households. Based on those results, Togolese policymakers could directly calculate which of the surveyed households suffer extreme poverty and how much aid would be needed to bring them above the line.
A key challenge, however, is to figure out the poverty and needs of the millions of households that did not participate in the survey. Machine learning allows policymakers to identify a handful of readily observed criteria - such as the size of a home, the number of rooms, the roofing materials, or even how the home appears in satellite imagery - that together provide a good indication of the household's underlying need.
Those indicators can then be used to determine how much support should be allocated to the many around the country that did not participate in the survey.
"Statistical machine learning helps policymakers figure out how to make more nuanced decisions about how much each household should get," Blumenstock said. Relatively precise payments can then be transferred to each household using phone-based banking software to get them above the extreme poverty line.
Lon&Queta via Flickr
Evidence indicates that the infusion of cash into desperately poor communities would drive economic growth, the authors write. And compared to the universal basic income approach, targeted aid would save hundreds of billions of dollars.
Political challenges, and cause for optimism
The new working paper comes with caveats, however, largely focused on cost.
Most low-income countries cannot afford to make these payments without outside support. At the same time, the authors acknowledge that the U.S. and other governments have sharply cut foreign aid since 2023.
Adding to the strain: The targeted aid payments they detail would not replace existing aid programs, but supplement them. And the projected cost of ending extreme poverty would rise significantly if global organizations agree with the World Bank's decision last year to raise the benchmark for extreme poverty to $3.00 per person per day.
But an important value of the study is simply showing that the cost of eradicating extreme poverty could be relatively manageable if policymakers use the new tools to make more precise, targeted aid payments.
The authors calculate that, for an American earning the median annual income of $45,000, the cost of ending extreme poverty this way works out to about $135 per year. They also explore scenarios where billionaires make investments to wipe out poverty in individual countries, easing human pain and establishing a noble personal legacy.
"Even if it's a moment where the cards might seem stacked against us," Blumenstock said, "I think that we like to be optimistic."
Learn more:
- Go to the project homepage to read an overview, a non-technical executive summary and frequently asked questions.
