Machine learning algorithms trained on mobile phone data to recognize patterns of poverty may help to prioritize aid to the poorest individuals, according to a study published in Nature. This method was used in Togo to distribute millions of dollars of COVID-19 relief aid to those deemed most needy with increased efficiency, when compared with traditional methods of aid distribution.
In response to the COVID-19 crisis, governments and humanitarian organizations worldwide have distributed social assistance to over 1.5 billion people. However, rapidly identifying and targeting those in greatest need remains a problem. Joshua Blumenstock and colleagues developed, implemented and tested an approach to address this issue based on machine learning algorithms that can measure poverty. Togo’s flagship emergency social assistance programme — launched in April 2020, shortly after the first COVID-19 cases appeared in the country — used these algorithms to disburse aid.
Compared to other geographic targeting methods considered by the government of Togo, the authors’ approach reduced errors of exclusion (or proportion of individuals in need left out of the government’s traditional aid distribution programme) by 4–21%. However, when hypothetically compared to other methods based on social registries (of which Togo had none), this AI approach increased exclusion errors by 9–35%.
These results demonstrate the ability of these methods to be implemented at scale in real-world crisis scenarios, according to the authors.
Planetary science: Modelling electrolyte transport in water-rich exoplanetsNature Communications
Robotics: Taking millimetre-scale origami robots for a spinNature Communications