Pokémon Go has been quietly harvesting player data to train AI

Do you remember playing Pokémon Go back in the day and all the surrounding hype? All the fun adventures of catching Pokémon around your neighborhood with friends and family? Niantic certainly does.

Niantic is the developer behind Pokémon Go, and it has recently announced that it’s building a new “Large Geospatial Model” (LGM) using millions of scans taken by Pokémon Go players on their smartphones (via IGN). How does the LGM learn from these scans? Well, that’s thanks to Niantic’s new Visual Positioning System (VPS).

“Over the past five years, Niantic has focused on building our Visual Positioning System (VPS), which uses a single image from a phone to determine its position and orientation using a 3D map built from people scanning interesting locations in our games and Scaniverse,” Niantic said in a press release.

How does Niantic’s LMG work exactly?

Lots of data and lots of training

You’ve likely heard of a large language model (LLM) model before, like ChatGPT for example. Its AI model is trained on vast amounts of existing text so it can produce pretty normal sounding and coherent text when the user prompts it to. An LGM operates similarly. It’s trained on what real world places look like, such as a church or park, and can use that data to produce information on places it hasn’t seen yet. Niantic says the LGM will help computers perceive and interact with physical spaces in new ways, and can be used in AR glasses, robotics, content creation, and autonomous systems.

So how does Niantic get all this location data? Well, its games, of course. Pokémon Go is still a popular mobile game, and it has users traveling and pointing their phone at places trying to catch Pokémon. Those location scans get sent back to Niantic, though its unclear how long it’s been collecting them for. The advantage of Niantic’s Pokémon Go is that it gets location data other services like Google Maps can’t.

To map out and create spaces with AI, you need more than just the perspective of a car on the road. You need the perspective of a pedestrian at street level looking at places and seeing different things, and that’s exactly what Pokémon Go players unwittingly deliver. I haven’t played Pokémon Go in years, but it’s wild to think I may have contributed my own data to creating an AI product like this. I guess the terms of service I didn’t read had a lot in it.

Related

This app just brought your ’90s Pokémon obsession back to life

The Pokémon company has launched Pokémon TCG Pocket, which lets players collect and battle Pokémon cards virtually.