Emerging Tech

Facebook is using A.I. to create the world’s most detailed population maps

It’s probably fair to say that Facebook hasn’t had a banner year when it comes to good publicity. What just a few years ago seemed like a benevolent and cool tech giant is often now cast in a negative light, largely due to the lingering effects of controversies like the Cambridge Analytica scandal. But Facebook is working hard to show that that its expertise in A.I. has a valuable role to play as a force for good in the world.

With that in mind, A.I. experts and data scientists at Facebook have today shown off the world’s most accurate population density maps yet created. Building on work that dates back to 2016, the company has unveiled maps covering the majority of the African continent. Eventually, it says that this will expand to map nearly the whole world’s population. This will allow humanitarian agencies to determine how populations are distributed even in remote areas; opening up new opportunities for healthcare and relief workers to be able to distribute aid where necessary.

“[Our population density maps] are 3x more detailed than any other population map available,” James Gill, a software engineer at Facebook, told Digital Trends. “On the technology side, we have to address data infrastructure on a massive scale, developing spatial algorithms that work with interesting computation graph topologies, novel labeling techniques, and apply state-of-the-art research in machine learning … On the impact side, our work informs life-saving service delivery like vaccination delivery and natural disaster response. So it’s very exciting both from a technical perspective, as well as from an impact perspective.”

Facebook stressed that no Facebook user data has been, or will be, used in the project. The population maps are instead based on census and satellite data, which contain no personally identifiable information. To create its dwelling-spotting A.I., the company built a tool that can take extremely high-resolution satellite data (50 cm of land per pixel) and then cross-reference it with census data to build up an accurate picture of how many people live in an area. While this could be done manually using satellite images, it would be incredibly time-consuming and tedious. Facebook’s algorithms can achieve the same thing with an accuracy level of 99.6 percent.

The results are interesting from a computer science perspective. From a human perspective, however, they could be life-saving in a humanitarian situation. See, this is kind the of Facebook we want to see more of!

Editors' Recommendations