Making us genuinely nostalgic for those days when all that Pokémon Go users were doing was finding corpses and committing crimes, today it turns out that they've also been training robots how to, like, find and navigate locations in the real world. (Like, we can't help mentally adding, our homes.) This is per a blog post from the game's creator, Niantic—which we first found through 404 Media, which got it from Garbage Day—cheerfully describing how it's used millions upon millions of scans of the world, provided by users just trying to get another damn Bulbasaur, to build what it calls a "Large Geospatial Model." (Kind of like a large language model, except it's built from pictures of the world instead of text. This is probably hideously inaccurate from a technical sense, but you get the gist.)
The post in question gets fairly technical, but the upshot is that it turns out that getting an entire generation of smart phone users to send you a dozen pictures of their environment every single day for years is a pretty good way to collect training data that can be used to teach computers what 3D structures look like. Niantic mostly focuses the conversation around this tech around augmented reality uses—including a recently released Pokémon Go feature that lets you leave monsters out in the world for other players to look at, and applications that allow a computer to accurately fill in missing parts of pictures or provide reverse angles—but the post (which mentions, among other things, tech that allows the company's models to pinpoint, to the centimeter, where in the world a picture has been taken) also makes it clear that the applications for teaching A.I. how to understand and navigate 3D space are, uh, robust.
So, hey, have fun feeding candy to that Snorlax you caught last week when the Synthetic Kill-o-Troopers scale the walls around your compound, navigate your tripwires and snares, and pierce the inner sanctum where The Last Family has been hiding in fear of the ravenous metal hordes. (Or, you know, the slightly less improbable, but still extremely grim, outcomes that grow out of computers getting better at recognizing 3D structures in the real world.) Hope that 20th fucking Rattata you caught today was worth it.
