‘Pokémon Go’ Players Unknowingly Trained AI Map With 30 Billion Images

ByMF Jones

This is absolutely nuts. 143 million people who thought they were catching Pokémon were actually building one of the largest real-world visual datasets in AI history.

Nearly 10 years ago, Pokémon Go turned the real world into a digital scavenger hunt and we all had loads of fun capturing Pikachus and Charizards in parks and alleyways and buildings across the globe.

Well, Niantic Spatial, part of the team behind Pokémon Go, have now disclosed that photos and AR scans collected through Pokémon Go have produced a dataset of over 30 billion real-world images.

The company is now using that data to power visual navigation AI for delivery robots in partnership with Coco Robotics.

Which means we weren’t just walking around with our phones capturing virtual creatures. We were scanning landmarks, shopfronts, parks, and pavements from every angle, at every time of day, in lighting and weather conditions that standard photography would never capture. We documented the physical world at a scale no mapping company could have replicated on the same timeline or budget.

Niantic systematically collected these images, data point by data point, across 8 years, while we played the game, completely oblivious.

Ultimately, hundreds of millions of players who played Pokémon Go around the world built what might be one of the most valuable AI training datasets in the world, and we had no idea we were doing it.

Pretty incredible. I mean, I’m sure it was all there in the terms and conditions when we downloaded the app, but a) no one reads all that and b) even if they did, I don’t think anyone could have imagined that we were making this level of contribution just by playing the game.

They basically made us all work for free! Still, we had a great time doing it. Fair enough really.

What next? OnlyFans collecting images of women from every angle to train up AI girlfriends for the rest of us? Wouldn’t put it past them.

Similar Posts