Pokémon Go has been quietly harvesting participant knowledge to coach AI

Key Takeaways

Niantic is creating a “Massive Geospatial Mannequin” (LGM) from knowledge collected by gamers in its video games like Pokémon Go.
The mannequin learns from scans utilizing a Visible Place System (VPS), which may use a single picture to find out its place on a 3D map.
It is unclear how lengthy Niantic has been gathering this location knowledge for.

Do you keep in mind taking part in Pokémon Go again within the day and all the encompassing hype? All of the enjoyable adventures of catching Pokémon round your neighborhood with family and friends? Niantic actually does.

Niantic is the developer behind Pokémon Go, and it has recently announced that it is constructing a brand new “Massive Geospatial Mannequin” (LGM) utilizing hundreds of thousands of scans taken by Pokémon Go gamers on their smartphones (through IGN). How does the LGM be taught from these scans? Nicely, that is due to Niantic’s new Visible Positioning System (VPS).

“Over the previous 5 years, Niantic has targeted on constructing our Visible Positioning System (VPS), which makes use of a single picture from a cellphone to find out its place and orientation utilizing a 3D map constructed from individuals scanning attention-grabbing areas in our video games and Scaniverse,” Niantic stated in a press release.

Associated

AI is running out of gas

Developments in AI have been shifting at what looks like gentle pace, however there are indicators it starting to sluggish.

How does Niantic’s LMG work precisely?

A lot of knowledge and many coaching

You have probably heard of a giant language mannequin (LLM) mannequin earlier than, like ChatGPT for instance. Its AI mannequin is skilled on huge quantities of current textual content so it may well produce fairly regular sounding and coherent textual content when the consumer prompts it to. An LGM operates equally. It is skilled on what actual world locations appear like, reminiscent of a church or park, and might use that knowledge to provide data on locations it hasn’t seen but. Niantic says the LGM will assist computer systems understand and work together with bodily areas in new methods, and can be utilized in AR glasses, robotics, content material creation, and autonomous techniques.

So how does Niantic get all this location knowledge? Nicely, its video games, in fact. Pokémon Go continues to be a well-liked cell recreation, and it has customers touring and pointing their cellphone at locations attempting to catch Pokémon. These location scans get despatched again to Niantic, although its unclear how lengthy it has been gathering them for. The benefit of Niantic’s Pokémon Go is that it will get location knowledge different providers like Google Maps cannot.

To map out and create areas with AI, you want extra than simply the attitude of a automotive on the street. You want the attitude of a pedestrian at avenue stage locations and seeing various things, and that is precisely what Pokémon Go gamers unwittingly ship. I have not performed Pokémon Go in years, nevertheless it’s wild to suppose I could have contributed my very own knowledge to creating an AI product like this. I suppose the phrases of service I did not learn had rather a lot in it.