When Hurricane Ian swept through Florida in late September, leaving a trail of destruction with high winds and flooding. But a week after the storm passed, some people in three of the worst-hit provinces saw an unexpected beacon of hope.
Nearly 3,500 residents of Collier, Charlotte, and Lee counties received a push notification on their smartphones offering $700 in cash assistance, with no questions asked. A Google algorithm deployed in conjunction with GiveDirectly, a nonprofit organization, estimated from satellite imagery that those people lived in severely damaged neighborhoods and needed help.
GiveDirectly is testing this new way to target emergency relief in partnership with Google.org, the charitable arm of the search and advertising company. The individuals bidding money were users of a benefits app called Providers, which manages food stamp payments. By targeting posts using Google’s AI software, GiveDirectly was only able to provide assistance to people living in areas devastated by Ian, faster than manually sorting the roles of the app’s users.
This is GiveDirectly’s first time using this technology in the US, but it has previously tested a similar idea in Togo in the months after the pandemic crippled the global economy. There, households were offered help based on signs of poverty detected by image algorithms from UC Berkeley researchers and clues from cell phone bills.
The Florida project was made possible by a mapping tool called Delphi, developed by four Google machine learning experts who have worked with GiveDirectly for six months since the end of 2019. The software marks communities in need after disasters such as hurricanes by overlaying live maps of storm damage with poverty data from sources including the U.S. Centers for Disease Control and Prevention. The storm damage data is provided by another Google tool called Skai, which uses machine learning to analyze satellite imagery before and after a disaster and estimate the severity of damage to buildings.
“You now have a map that shows where is socio-economically vulnerable and where is damaged,” said Alex Diaz, who leads the AI for Social Good team at Google.org. assistance.”
The algorithms driving Skai’s damage assessments are trained by manually tagging satellite images of several hundred buildings in a disaster-stricken area that are known to be damaged. The software can then quickly detect damaged buildings throughout the affected area. A research paper on the underlying technology, presented at a 2020 academic workshop on AI for disaster relief, claimed that the automatically generated damage assessments match those of human experts with an accuracy of between 85 and 98 percent.
In Florida, GiveDirectly this month sent a push notification with an offer of $700 to any Providers app user with a registered address in the Collier, Charlotte, and Lee County neighborhoods, where more than 50 percent of them, according to Google’s AI system. the buildings were damaged. So far, 900 people have accepted the offer, half of which have been paid. If every recipient accepts GiveDirectly’s offer, the organization will pay out $2.4 million in direct financial aid.
