Stanford scientists use Artificial intelligence to map poverty

Stanford scientists have taken the artificial intelligence route to map poverty in areas which were previously devoid of data. The researchers will combine satellite images and machine learning to help organizations and policymakers to distribute funds and evaluate policies more efficiently. The low-cost method would assist governments and charities to eradicate extreme poverty; finding people who live below the poverty line of $ 1.25 a day. While there are plenty of initiatives going on in Africa but the latest developed poverty map will give direct aid to some of the most deprived areas.

The idea behind using satellite images for work came when the researchers found significant errors and gaps in the current data. The other fact being that there is a dearth of local survey information about poverty levels in individual African villages but plenty of satellite images are available. So the task was for scientists to come up with ways to use these images to extract valuable insights.

How will satellite find deprived areas?

Since the satellites have eye-view of the planet, so places on Earth that appear brighter at night tend to be developed, while in deprived areas luminosity levels are lower. Previously World Bank had used night lights to measure regional inequality and income inequality in Africa. Inspired by this method Burke and team of scientists have developed an algorithm that is better in comparison to all existing methods.

How the model work?

To make this model work, researchers used three inputs into a computer, daytime high-resolution imagery, nightlife luminosity data and actual survey data. For this project, five countries were chosen from the continent- Nigeria, Tanzania, Uganda, Malawi, and Rwanda.

A two-step process known as transfer learning was used to train the algorithm. Firstly scientists combine data on night light and daytime images. Once the computer maps things like roads, urban areas, farmland, etc., then the model look for correlations. After this, the model ingests the actual survey data and compares it’s against correlations. Then the model takes all this map to make predictions about the missing data.

Will the tool see success?

Two major reason to opt for this device is that it is nearly costless to scale across countries. As the imagery used for this project are free due to Google Maps. Also, the code is open source so anyone with computer skills can download and modify it.

The second point to be noted is that the latest developed way of tracking poverty can replace door-to-door household survey. These surveys are not only expensive and institutionally challenging, but there are always loopholes in the document.

Ultimately scientist from Stanford believes that governments and organizations will no longer go in the blind path as the tool will evaluate anti-poverty programs to reduce poverty.