The maker of the wildly popular mobile game Pokémon Go is making a sharp turn toward a new frontier in AI research.
Pokémon Go players have spent years voluntarily uploading location data from their smartphones coupled with images and video of their surroundings. Using this raw data, Niantic, which develops Pokémon Go and other games, is aiming to create the first ground-level geospatial model of the earth, a model that can see, reason, and interact with the physical world around us.
Toward that goal, San-Francisco-based Niantic this week announced a nearly $4 billion deal that will spin off most of the gaming studio and return the remaining company to its roots in mapmaking.
“Humans take spatial intelligence for granted. We walk in one direction and we don’t need to see the other direction to be able to turn around and go back, even for places we’ve never been to before,” Niantic senior VP engineering of the Niantic Spatial Platform, Brian McClendon, told Barron’s in a December interview. “For a computer this is a huge challenge.”
As of November, Niantic had 10 million scanned locations worldwide, concentrated in North America, Europe, and Japan.
Niantic hopes to make a large geospatial model that can be the foundation for a range of location services which require detailed ground-level mapping.
In a Wednesday LinkedIn post, Niantic CEO John Hanke said, “Today’s LLMs represent the first step toward a future where a variety of expert models collaborate to reason and understand complex problems, and many of those problems will require deep and accurate knowledge of the physical world.”
McClendon and Hanke led the teams that built Google Maps, Google Earth, Street View and other Google location-based services.
They are expert mapmakers—Niantic spun out of Google in 2010 to take its mapping expertise to gaming.
Mobile gaming studio Scopely will buy most of Niantic’s games, including Pokémon Go, for $3.5 billion, leaving $350 million in cash behind for Niantic shareholders whose haul would total $3.85 billion if the deal is approved by regulators.
The new company, Niantic Spatial, will be capitalized with $200 million from its predecessor, plus another $50 million investment from Scopely.
Niantic Spatial will need every penny of that for this project, and it still may not be enough: The amount of data collected and put into the model are far more than even the largest language model. Even in its early form, if Niantic put all of its scanned locations together, it would be about 100 times the size of the largest language model. For now, the company will limit itself to small portions of the full data set. Niantic will have high and growing computing needs, and it may require technological breakthroughs that dramatically reduce AI computing costs for it to succeed.
And the model will never be finished. “One of the big challenges with a 3-D map like the one we are building is it’s never complete because the world is dynamic and frequently changes,” McClendon told Barron’s.
The transaction is a barometer of where tech has come in the 15 years since Niantic was founded. Back then, mobile gaming looked like a huge opportunity. It was. Today, that opportunity is brewing in AI.