Slashdot Mirror
Facebook is Using AI To Map Population Density Around the World (theverge.com)
Facebook has always been a company with global ambition, but few projects illustrate this better than its ongoing attempt to map the world's population density using AI. From a report: The company first unveiled this work back in 2016 when it created maps for 22 nations. Today, it expanded that with new maps that cover the "majority" of Africa. "The project will eventually map nearly the whole world's population," Facebook said in a blog post. As Facebook explains, creating maps like this is a challenging job for humans. Although we have high-resolution satellite imagery that covers pretty much every corner of the globe, turning this into useful information is a time-consuming process. To create population density maps, for example, humans have to label each building in the images, then cross-reference this with census data.
This is particularly tricky in the African continent where census tracts can cover regions as large as 150,000 square miles but contain just 55,000 people. Luckily, this sort of task -- tedious but simple -- is perfect for AI. To automate this process, Facebook's engineers used data from open-source mapping project Open Street Map to train a computer vision system that can recognize buildings in satellite imagery. They then used this to remove the vast majority of the satellite data that showed unoccupied land.
This is particularly tricky in the African continent where census tracts can cover regions as large as 150,000 square miles but contain just 55,000 people. Luckily, this sort of task -- tedious but simple -- is perfect for AI. To automate this process, Facebook's engineers used data from open-source mapping project Open Street Map to train a computer vision system that can recognize buildings in satellite imagery. They then used this to remove the vast majority of the satellite data that showed unoccupied land.
Why facebook?
Data is their business. Presumably they have the raw data necessary to come up with this map, maybe they will even publish it as a freebie, but for them it'll probably just be a little exercise in machine learning, to see what else they can glean from the piles of other data they have. Maybe map this population data to affluence, sexual preference, age, political leaning, spending patterns... and that good stuff they can sell.
If construction was anything like programming, an incorrectly fitted lock would bring down the entire building...
From the description, it seems like it's counting buildings rather than people. While buildings certainly imply human habitation, there are so many exceptions (farming/retail/industrial/utility buildings, abandoned homes, seasonal homes, homeless people, nomads, migrants, refuges, homes with more or less occupants than average [and how do you determine average]) that the results would be next to useless.
Support Right To Repair Legislation.
Whatever they are going to use the information for, it certainly won't be to benefit you, or me, or anybody but Facebook.
You live and learn, or you don't learn much.