How Pokémon GO Players Helped Niantic Build Advanced AI Models

Niantic, the creator of the popular AR game Pokémon GO, has revealed that they were using players’ real-world location data to build its AI project called the Large Geospatial Model (LGM).

Niantic in their blog explained, that when players scan locations during gameplay, they’re helping create detailed 3D maps that AI can understand and interpret.

The Data Behind the Innovation

Since Pokémon GO’s debut in 2016, the game has attracted millions of players worldwide. These users explore real-world locations to catch virtual creatures, generating vast amounts of geospatial and augmented reality (AR) data. Niantic has harnessed this data to train its AI models, aiming to create systems capable of understanding and interacting with real-world environments.

According to the company, the LGM relies on millions of scans from games like Pokémon GO, Pikmin Bloom, and Ingress to enhance its ability to “see” and analyze physical spaces. Unlike traditional AI systems, such as Large Language Models (LLMs) that process and interpret textual data, LGMs deal with spatial intelligence—how physical environments are structured, connected, and experienced.

As of November 2024:

• Over 10 million locations scanned globally
• 1 million active locations in their Visual Positioning System (VPS)
• Players add approximately 1 million fresh scans weekly

The company has already trained more than 50 million neural networks with over 150 trillion parameters, creating an incredibly complex and intelligent system.

Potential Applications

Players contribute to this process through their interactions, such as sharing location data or scanning areas using their smartphones. Niantic uses these inputs to build detailed 3D maps, which power not only AR gaming experiences but also broader applications like robotics, AR glasses, logistics, and urban planning.

A standout feature is Niantic’s Visual Positioning System (VPS), which captures high-accuracy images of real-world locations. By combining billions of scans, the VPS creates maps that are so precise they can identify a device’s position with centimeter-level accuracy.

Privacy Concerns

While the potential of this technology is impressive, it also raises important questions about privacy. Niantic assures players that data collection for AI training is optional. Players must actively choose to scan public locations for their data to contribute to LGM development. Additionally, Niantic’s privacy policy guarantees that personal details, like names and email addresses, are not used for these purposes.

Sources: Niantic Official Blog Post (November 12, 2024)