Pokémon Go Maker Used Billions Of Images To Train An AI Map Of The World

Pokémon Go players have long suspected that developer Niantic was selling the map data those players were creating in the course of playing the augmented reality monster catcher. It turns out they were right, and now we know that one of the clients Niantic is selling this information to is using it to train AI map models used by delivery robots. Not even Pikachu is safe from contributing to the AI dystopia.

Niantic Spatial, the AI-focused offshoot of the Pokémon Go company that was formed last year when Niantic was acquired by Saudi-owned Scopely, has revealed it has used Pokémon Go’s map data to create a map model that can pinpoint a person’s location within centimeters. This would be used by delivery robot companies like Coco Robotics for more accurate navigation in places where GPS is less reliable, like cities with higher rates of signal interference. Niantic Spatial has been using data collected from Pokémon Go to build a visual positioning system that uses images and videos instead of just coordinates from a GPS. The model has been trained on over 30 billion images captured in urban environments, specifically around hot spots like gyms where players would have taken photos from many different angles and at different times of day.

“We had a million-plus locations around the world where we can locate you precisely,” says Niantic Spatial CTO Brian McClendon. “We know where you’re standing within several centimeters of accuracy and, most importantly, where you’re looking.”

John Hanke, CEO of Niantic Spatial, says partnering with Coco Robotics is the beginning of a much larger vision to create a virtual simulation of the world that changes as the world does, and gathers more map data from more robots using the system.

We like to joke that Pokémon Go’s first summer was basically the closest the world ever came to peace, but even those good memories must be soured by the capitalist hellscape, as it turns out everyone was unknowingly doing unpaid labor for an AI company creating something that sounds like a tool for a surveillance state. The foregrounding of the use of this data to help delivery robots makes the whole thing sound like it could be a net good for people who have had trouble with getting their lunch delivered, but it fails to mask the more sinister implications of a company creating a hyper-detailed map of the world and selling it to anyone it pleases.