Pokémon GO Players Fuel Niantic's Ambitious AI Endeavour
Niantic, the creator of Pokémon GO, is building an AI model aimed at mapping and predicting real-world environments.
Known as a "Large Geospatial Model" (LGM), this AI leverages vast amounts of location data provided by Pokémon GO players and other Niantic app users to train its system.
The company announced on 12 November that this development focuses on achieving what it calls "spatial intelligence," an innovative way for computers to perceive and navigate the physical world.
Niantic is quietly using your Pokémon Go data to train large-scale geospatial Al models. Niantic is making geospatial Al models to help computers navigate real spaces, and it's using your Pokémon Go data to help. Niantic quietly announced its Al plans in a corporate blog post and… pic.twitter.com/B3Z6jWBK0m
— ConsiouSean (@ConsiouSean) November 22, 2024
How Niantic's Model Works
Niantic's approach relies on large-scale machine learning to connect millions of real-world scenes.
This model is trained using images submitted by players, collected through its Visual Positioning System (VPS).
These photos are combined to build detailed 3D maps that reflect various conditions, including seasonal changes, lighting variations, and perspectives that vehicles or drones cannot access.
To date, Niantic has amassed over 10 million scanned locations worldwide, with 1 million new scans added weekly.
The AI model analyses this data to comprehend places even when full scans are unavailable, allowing it to predict the appearance of locations.
For instance, by recognising parks, churches, or homes, it can make educated guesses about similar environments elsewhere.
Niantic likens this approach to how ChatGPT understands language—using extensive data to identify patterns and generate meaningful responses.
Why Pokémon GO Players Are Key to This Project
Pokémon GO is more than a game; it has become a tool for gathering data that powers Niantic's ambitious AI model.
As players interact with their surroundings, they contribute geospatial data through actions like scanning PokéStops, gyms, or specific landmarks.
This information is crucial for Niantic's VPS, which determines a phone’s exact location and orientation with centimetre-level precision.
Pokémon Playgrounds, an experimental feature in Pokémon GO, is one example of this integration.
It allows players to leave virtual Pokémon in real-world locations for others to interact with and photograph.
This not only enhances gameplay but also feeds the AI with valuable pedestrian-level insights.
What Sets Niantic's VPS Apart?
Unlike traditional mapping systems that depend on vehicles or aerial imagery, Niantic's VPS gathers data from ground-level perspectives, often covering places inaccessible to cars.
This unique approach provides detailed insights into public spaces and private areas alike.
Moreover, Niantic has trained its model using 50 million advanced neural networks and over 150 trillion parameters.
By compressing thousands of images into lightweight neural data, the system becomes capable of analysing environments at an unprecedented scale.
Applications of the Large Geospatial Model
While gaming is the most apparent use of Niantic’s LGM, its potential extends far beyond.
The company envisions applications in augmented reality (AR), urban planning, logistics, and even remote collaboration.
AR glasses, for instance, could use the model to blend digital objects seamlessly with the physical world.
Similarly, city planners might utilise it for spatial analysis, while businesses could streamline delivery routes by understanding pedestrian patterns.
Niantic sees this project as the foundation for spatial computing, bridging the gap between the digital and physical realms.
As AR devices grow in popularity, the LGM could reshape how humans interact with technology in their everyday lives.
Privacy Concerns and Data Usage
Niantic’s reliance on player data raises questions about privacy.
According to its privacy policy, the company collects location data, names, and email addresses, among other details.
However, it does not clarify how this information is processed beyond its use in games.
For children, a separate policy exists, with a portal for parents to manage their child’s profile.
Anton Dahbura, Executive Director of the Information Security Institute at Johns Hopkins University, remarked that companies leveraging customer data for such purposes are becoming "the new normal."
Niantic maintains that its data collection offers unique value by focusing on pedestrian perspectives, setting it apart from other mapping technologies.