Zoom in the Google Maps software on your computer, and finally, you’ll see the outlines of the houses. You should credit artificial intelligence for helping to build scenes like that — and that’s not everything AI’s done with the game. Over the last few years, the company has turned more to machine learning to automatically keep tabs on the changing geography of the world and then update how it reflects cryptographically. So it is very fascinating to think that, can maps be generated automatically? If yes then who? No?
In reality, Google reached an inflection point around 2015 when it learned that it had to adjust its approach to keep its maps up-to-date, according to two Google Maps employees who talked exclusively to Popular Science. Andrew Lookingbill, Head of Innovation at Google Maps, defines the experience as an “epiphany.”
Holding maps up-to-date in more than 200 countries is hard — so the team needed to switch from only creating maps to something more strategic. “We had to start designing the computer that produces the world,” says Lookingbill.
The way this happens is by machine learning algorithms that are strong enough to take pictures — like those generated by streetcars or satellites — to derive the details they need from them and then update the map. This knowledge is likely to be details such as the name of the lane, the house number, or the shape of a building seen from above. Google has prided itself on this topic before: the 2017 blog post outlines their attempts to develop an algorithm that can interpret street names in France and notes that algorithms like this might change addresses on the globe.
Say somebody is constructing a new house and a street-view, vehicle cruises. “It may end up being searchable on our charts without a human being in the inner circle, or having something to do with it,” says Lookingbill. This process — of AI studying imagery and upgrading the map — is what he considers “the first step towards our maps being self-healing.”
Creating building diagrams is one mission, he says, where better AI has rendered it easier. A machine learning algorithm will look at the images of the satellite and then create the outline of the building on the Google map. Thanks to that, “we were able to increase the number of buildings we built across the planet,” says Lookingbill. Everything has occurred over the span of a year. “For a sense of scale,” he says, “all of the previous buildings that we’d have, have taken us a decade to model.” Google reflects on that in a blog post published today, which outlines how the previous algorithm generated the building as appearing “fuzzy” (the post also discusses the basic steps and data sources that go into their mappings).
Other research, still in its “born” phase, includes utilizing AI to construct new routes on the map from the imagery it analyzes. The “route reconstruction,” says Lookingbill, includes them “actively attempting to work out the structure of the roads that we don’t already have on the globe, based on images.”
For artificial intelligence algorithms to do stuff like building diagrams or draw new paths, the company utilizes images like top-down satellite data; to retrieve details like street addresses, house numbers, and business titles, the company depends on on-street observation.
And, most specifically, the tech environment has machine learning algorithms which are used to train data to carry out tasks at a super-human stage, at times. That may involve things worldly like Yelp, which uses AI to evaluate and arrange the pizzas and tacos that its customers access. And AI doesn’t do just things like recognizing the images: it can also play and win a host of things like poker or even a Rubik’s cube. I think you can do a lot of other things.