POKEMON GO Uses Player Photos to Teach AI Model

As someone who’s been captivated by the magical world of Pokémon since I was a child, watching this new evolution of technology is nothing short of mind-blowing! The idea that the very app I use to catch digital monsters could be contributing to something as groundbreaking as an AI that auto-completes real-world locations is simply astounding.


In one way or another, we’re all immersed in the digital age, and this includes the realm of Pokémon GO as well. As reported by IGN, the developers of Pokémon GO, Niantic, are currently working on an AI system that can automatically identify real-world locations based on limited information. This is made possible through data gathered by players of Pokémon GO. Essentially, this AI learns what real-world places look like and uses that knowledge to predict the appearance of unseen locations. In essence, it’s similar to ChatGPT, but instead of text, it operates within the physical world. Here’s a statement from Niantic about their innovative technology:

Picture yourself standing at the backside of a unique church. Let’s consider a nearby AI model that has only encountered the main entrance of this church, making it unable to identify your exact location. Since it hasn’t seen the back before, it won’t be able to pinpoint where you are. However, on a broader scale, our AI has inspected countless churches across the globe, each captured by different local models in various locations worldwide. No two churches are identical, yet many display shared features. An LGM (Large Geospatial Model) serves as a tool for accessing this dispersed knowledge.

The existence of these environments within Niantic’s apps, particularly Pokémon Go, is made possible because users are continually scanning their surroundings. Niantic reports that there are currently approximately 10 million scanned locations spread across the globe, with an additional million accessible through their VPS service. They receive around a million new scans every week, each containing hundreds of distinct images. For now, these resources are mainly utilized for the enjoyable experience provided by Pokémon Go. It remains to be seen what possibilities this technology might hold for less entertaining purposes in the future.

Read More

2024-11-21 03:02