Facebook Can Map More of Earth In a Week Than We Have in History

Facebook Perform Better/Quicker Mapping the Earth in less than a Week

A blog post in the social network has announced that its AI system had taken two weeks in building a map which covers about 4% of our planet. It is 14% of the Earth’s land surface with 21.6 million square kilometres of photographs taken from space, processed and traced in a digital representation of buildings, settlements and the roads they tend to show. Facebook has said that it could perform better and much quicker possibly mapping the complete Earth in less than a week.

The goal of data-science team of Facebook is to create maps in helping the social network plan how to deliver internet to the users who are presently offline. It is an uncertain starting point though whatever one may think regarding Facebook’s internet colonialism, the drones of the company would be capable of beaming Wi-Fi to the disconnected till they know where they are. The model was competent in mapping 20 various countries after training on just 8000 human-labelled satellite photos from an individual nation. This overwhelming together with Facebook’s data science team had not even attempted to go fast.

Open Street Map Project

The company has stated that it has enhanced the process where it could do the same mapping within a few hours. Presuming that it had the photographs, it could map the Earth in around six days which is something that humanity has not managed to do yet. Facebook had mentioned in its blog post that they had processed 14.6 billion images with their convolutional neural nets, naturally running on thousands of servers at the same time. Utilising its AI, Facebook had imitated how humans tend to make maps in the 21st century.

It is now being made by a project known as Open Street Map that tends to use volunteer labour in tracing satellite photographs by hand, picking out the roads and houses. The subsequent maps have been utilised all across the world often for calamity response, with the system capable of building maps of a complete unmapped area within a few days. Facebook’s AI has the capability of doing the same within seconds. Mapping on the scale which its AI system had demonstrated would probably take decades for human team of extent. It is more data than people or their organisations that are built to manage.

All Neural Networks Scale Like Facebook’s Mapping AI

Map-making AI of Facebook is just one of the possibly thousands of narrow AIs, those which are trained to aim on individual task, tossing through human tasks around the planet presently, faster and on bigger scales. The CERN particle physics laboratory towards Geneva, Switzerland is utilising deep learning to discover patterns in the mass of its collision data and pharmaceutical companies are utilising it to discover new drug ideas in data sets which no human can align.

Alison Lowndes of Nvidia who tends to help organisation in building deep-learning systems has said that she now works with everyone, such as doctors, governments, researchers, parents, retailers as well as mysteriously meat packers. What is appealing is that all neural networks tend to scale like Facebook’s mapping AI. The actual power of narrow AI is not in what it could do since its act is almost not good as that of a human. Maps, which Facebook’s AI tends to produce is not as good as those which come from a company like custom map developer Mapbox.