We seem to have technology on the brain this week at Demography is Destiny. Shannon has written about detaching oneself from social media and also about the demographics of those noisy people on Twitter. Now I’m here to inform you that avoiding Facebook might be harder than you think. This is because the media giant is using AI in an attempt to map the world’s population density. Back in 2016 the company created maps for 22 nations. Now it has new maps that cover most of Africa. In the years ahead the project is aiming to map “nearly the whole world’s population”.
So what exactly is the company doing? What exactly can it bring to the cartography and satellite imaging business? There is high-resolution satellite imagery available for most of the globe, but turning it into useful information is hard. The creation of population density maps is labour-intensive and tedious: each building must be labelled and then cross-referenced with census data. This is apparently a job that is perfect for AI – Facebook’s engineers trained a computer vision system to recognise buildings in satellite imagery. It then removed the vast majority of satellite data which had no habitations and was therefore (presumably) unoccupied. To map all of Africa, the Facebook “machine learning systems” analysed 11.5 billion 64×64 pixel images. Researchers from the Center for International Earth Science Information Network (CIESIN) at Columbia University verified the work completed, leading to “unprecedented accuracy,” according to Facebook’s announcement.
But what is this population density mapping for? Facebook has emphasised the humanitarian aspect of the project: the maps it has created will be released free of charge for anyone to use. These can be used to help with disaster relief and vaccination schemes in remote places as well as helping people determine where limited resources can be best put to use. The American Red Cross uses Facebook’s data already for these purposes. For demographic purposes, this sort of information could be used to supplement or check census data that can be inherently unreliable. (Especially if census data is linked to government funds — I wrote about that in relation to Nigeria a few years ago.)
Interestingly though, back in 2016 when the mapping project was first unveiled, Facebook emphasised the commercial implications of this scheme. Knowing where people live helps Facebook know where the unconnected customers might be – those in third world countries without internet who could use Facebook’s help in getting them onto the internet (and onto Facebook). The most effective use of limited resources does not just apply to limited medical resources, but also to limited internet resources. If people are turning away from Facebook in larger numbers in the West, then perhaps Facebook can make good those losses in other countries. Facebook is a money-making scheme after all.
Facebook was not even a thing 15 years ago. Now it is mapping the globe and trying to bring the internet to the masses. I wonder where it will be in the next 15 years and what it will turn its machine learning systems to next?
Marcus Roberts is co-editor of Demography is Destiny, Mercatornet's blog on population issues.