You were training AI while catching Pokemon

Summary

Niantic's Pokémon GO, which became a global phenomenon, unknowingly helped build one of the largest visual data sets in history. The data, collected from millions of players' AR scans and photos, now powers highly accurate geospatial AI systems used for everything from delivery robots to urban planning. However, the ethical concerns arise from players' unaware contributions to a commercial AI model, raising questions about consent, data ownership, and privacy. The video highlights how gamified data collection can work without explicit consent, drawing parallels to other data-driven tech companies and their practices.

Takeaways

• 📱 Pokémon GO reached 500 million installs within 60 days and had 143 million active players at its peak in 2016.
• 🤖 Niantic Spatial, the AI company spun off from Niantic, now powers delivery robots using data collected from Pokémon GO players.
• 📸 Players unintentionally created one of the largest real-world visual data sets by taking millions of photos while playing the game.
• 🌍 Niantic's visual positioning system can pinpoint a location to within centimeters using photos instead of GPS.
• 🏙️ The data set is highly valuable because it includes multiple angles, times of day, weather conditions, and detailed metadata for each location.
• 🎮 Players were incentivized to scan real-world locations through in-game AR missions, generating training data without knowing its full commercial use.
• 💰 Pokémon GO and its game division were sold to Scopely, owned by the Saudi Royal Investment Fund, while Niantic retained the AI and spatial technology.
• 🔒 Niantic's terms of service allowed broad use of user-submitted data, often without clear disclosure of commercial applications.
• 🚗 The collected data supports a wide range of applications, including AR glasses, autonomous vehicles, urban planning, and potentially surveillance.
• ⚠️ The script highlights ethical concerns about consent, privacy, and repurposing user-generated data for commercial AI models.

Q & A

WhatQ&A Development for Script was the initial success of Pokémon GO in terms of user engagement?

Pokémon GO saw 500 million installations within 60 days of its launch, and at its peak, 143 million people were actively playing, walking around public places like parks and sidewalks while using their phones to catch digital creatures.

What significant technology did Niantic develop using the data from Pokémon GO players?

Niantic developed a visual positioning system that can pinpoint users' locations to within a few centimeters. This system uses images of buildings and landmarks, captured by players, to determine precise locations, without relying on traditional GPS.

How did Pokémon GO players unknowingly contribute to the creation of a massive geospatial dataset?

Players generated a large amount of visual data by taking photos of landmarks, which were then tagged with metadata such as GPS coordinates, camera angle, movement speed, and weather conditions. This data was used to train a spatial AI model.

What makes Niantic's data collection unique compared to other mapping services like Google Street View?

Unlike Google Street View, which uses fixed-height cameras mounted on vehicles, Pokémon GO's data came from millions of players capturing pedestrian-level images from diverse angles, times of day, and weather conditions. This allowed for continuous data collection for 8 years.

How did Niantic gamify data collection for players?

Niantic gamified data collection through AR scan tasks, where players were rewarded with in-game items for scanning real-world locations to create 3Pokémon GO Data InsightsD models. This incentivized players to contribute data without realizing its broader use.

What are the ethical concerns surrounding the data collection model used by Niantic?

The ethical concern is that players were not fully informed about how their data would be used. They thought they were simply playing a game, but their data was actually being used to train a commercial AI system. The consent model was not transparent about the potential commercial applications.

What companies and countries are involved in the ownership and use of the data collected from Pokémon GO?

After Niantic sold its game division to Scopely in 2025, the ownership of Pokémon GO transferred to Scopely, which is partly owned by Saudi Arabia’s Savvy Games Group. Meanwhile, Niantic retained its spatial AI business, which is now used by companies like Coco Robotics for autonomous delivery robots.

How did Niantic's privacy policy handle the data collected from players?

Niantic’s privacy policy allowed them broad rights to use the data collected through Pokémon GO and pass those rights to other companies. The AR scan feature was presented as voluntary, but it was embedded in the gameplay and incentivized through rewards, making it unclear how much data was actually being contributed.

What is a geospatial model, and how is Niantic's AI model similar to a large language model?

A geospatial model predicts locations based on visual input, similar to how a large language model predicts the next word in a sequence. Niantic's AI model uses a vast dataset of images from various angles, weather conditions, and times of day to accurately predict locations from photos.

How could the data collected by Niantic be used for applications beyond gaming?

The data collected by Niantic has a wide range of applications, including for augmented reality (AR) glasses, autonomous vehicles for navigation in urban areas, construction site mapping, urban planning, and even surveillance by law enforcement or intelligence agencies due to its high level of accuracy.