Pokémon Go, the CIA, and thirty billion images: The invisible price of a free game - critical summary review

Pokémon GO: the largest invisible army of artificial intelligence

Thirty billion. That is the number of images Niantic collected from Pokémon GO players around the world. And most of those people had no idea that, while chasing a Pikachu around the corner from their house, they were building one of the largest visual databases in the history of artificial intelligence.

The revelation came in March of two thousand and twenty six, when Niantic Spatial ... the AI company spun off from Niantic's technology division ... announced a partnership with Coco Robotics, a delivery robot startup operating in cities across the United States and Europe. The idea is straightforward: use that entire archive of images to guide autonomous robots along sidewalks with a level of precision GPS has never been able to offer.

But to understand how we got here, we need to go back a bit.

In two thousand and sixteen, Pokémon GO launched and became a phenomenon that transcended the gaming world. Within sixty days, five hundred million people had installed the app. At peak popularity, two hundred and thirty million were playing every month. People leaving home before sunrise, crossing parks, wandering unfamiliar streets ... all to catch virtual creatures that appeared on their phone screens, overlaid onto the real world.

What few people realized is that the game was not just entertainment. It was a data collection machine.

Every time a player visited a Pokéstop, battled at a gym, or completed a mission, the phone recorded far more than the capture of a digital creature. It recorded the exact position of the device, the camera angle, the direction of movement, the speed, the time of day, and the lighting conditions. Each phone was essentially a portable weather station ... except instead of measuring rain, it was measuring the visual world around it.

In two thousand and twenty, Niantic accelerated that collection. It added a feature called Field Research, which encouraged players to scan statues, monuments, and points of interest with their cameras in exchange for in-game rewards. The idea seemed harmless: help map the world and earn virtual coins. But each scan generated detailed three-dimensional models of the real environment.

The result was something no professional mapping company could replicate. Millions of different people scanning the same locations, from different angles, at different times, under different weather conditions. A plaza photographed in the morning sun, in the afternoon rain, at night in the snow. Up close, from a distance, from above, from below. That diversity gave the database a richness that professional cameras mounted on cars, like those used for Google Street View, simply cannot match.

Brian McClendon, the chief technology officer at Niantic Spatial, put it in perspective in an interview with MIT Technology Review. The company has over one million locations around the world where it can position someone with a precision of a few centimeters ... and more importantly, know which direction that person is looking.

The technology now has a name: Visual Positioning System, or VPS. It works like this: instead of relying on satellite signals like GPS, the system analyzes what a device's camera sees and compares it against its database of images. If GPS is like trying to locate someone by shouting in a crowd ... where the voice bounces off walls and gets confused ... VPS is like recognizing that person's face in a high-resolution photograph. Far more precise.

And precision matters. In urban areas with tall buildings, GPS can be off by fifty meters or more. That puts you on the wrong block, on the wrong side of the street. For a pedestrian in a hurry, it is inconvenient. For an autonomous delivery robot that needs to stop right at the restaurant door or on the customer's sidewalk, it is a disaster.

That is where Coco Robotics comes in. Founded in two thousand and twenty in Los Angeles by Zach Rash, a recent UCLA graduate, Coco is now one of the largest urban robotics platforms in the world. Its electric sidewalk robots have completed over five hundred thousand zero-emission deliveries in cities like Los Angeles, Chicago, Miami, and Helsinki. They operate through Uber Eats, DoorDash, and Wolt, serving more than three thousand restaurants and merchants.

John Hanke, the CEO of Niantic Spatial, summed up the connection between both worlds in a memorable way: getting a Pikachu to run realistically along a sidewalk and getting a delivery robot to navigate safely along that same sidewalk are, at their core, the same problem.

But the story of Niantic does not begin with Pokémon. It begins with spy satellites.

Before creating Niantic, John Hanke co-founded a company called Keyhole in two thousand and one. The name was a direct reference to American military reconnaissance satellites. Keyhole built a three-dimensional Earth visualization software that caught the attention of two very specific investors: NVIDIA and In-Q-Tel ... the venture capital arm of the CIA, the Central Intelligence Agency.

The CIA and the National Geospatial-Intelligence Agency were not just investors. They were clients. They helped adapt Keyhole's software for use in the Iraq War in two thousand and three. Television networks CNN, ABC, and CBS used the tool to show simulations of the Baghdad bombings. Google noticed the potential and bought Keyhole in two thousand and four. The software was reborn as Google Earth, and then came Google Maps, Street View, and the entire geolocation infrastructure we use today.

Hanke stayed at Google leading the entire Geo products division. In two thousand and ten, he created Niantic as an internal Google lab. In two thousand and fifteen, he spun it off as an independent company. In two thousand and sixteen, he launched Pokémon GO.

In March of two thousand and twenty five, Niantic's gaming business was sold to Scopely for three point five billion dollars. Scopely is an American mobile game developer, but it belongs to Savvy Games Group, which is in turn controlled by Saudi Arabia's Public Investment Fund. Pokémon GO, along with other games like Pikmin Bloom and Monster Hunter Now, passed into the hands of a Saudi conglomerate.

At the same time, Niantic separated its spatial technology division into a new company: Niantic Spatial, led by Hanke himself, with two hundred and fifty million dollars in initial capital. That is the company now commercializing the mapping built from players' images.

This is where things get more delicate.

On the side of the defenders, there is a legitimate argument. The game informed users it was collecting mapping data. When the scanning feature launched, a clear notification appeared in the app telling players they were contributing to augmented reality data collection. The terms of service, accepted by everyone upon installation, granted the company broad rights over submitted content. From a legal standpoint, everything appears to be in order.

Some even argue this is one of the most ingenious business strategies ever conceived. You create a global game based on a beloved franchise, millions of people participate voluntarily, and the byproduct of that fun is a technological asset of immeasurable value. No personal information was exposed or sold beyond what Google Street View already shows. And the end use ... helping delivery robots find their way ... seems quite harmless.

On the side of the critics, the problem is not legal. It is ethical. Agreeing to terms of service and understanding what you are consenting to are very different things. How many of the five hundred million players who installed the app actually read the terms? And of those who did, how many imagined their photos of plazas and monuments would be used a decade later to train autonomous navigation systems sold to commercial companies?

This pattern is not new. Google ran the reCAPTCHA system for years, those tests that ask you to click on traffic lights, bicycles, or crosswalks to prove you are not a robot. A study by the University of California at Irvine estimated that, over the course of thirteen years, users spent eight hundred and nineteen million hours solving those challenges ... the equivalent of over six billion dollars in wages that were never paid. The classified images fed Google's computer vision systems, including those used by Waymo, the self-driving car division valued at forty five billion dollars.

The logic is similar: you think you are proving you are human, but you are actually training the machine to see the world like a human does.

And there is an additional concern with Niantic Spatial. A company executive has publicly acknowledged that the Large Geospatial Model they are building could, in the future, be sold to governments and militaries. A system that identifies with centimeter-level precision where a photo was taken, based on the buildings and landmarks visible in it, has obvious applications for surveillance and intelligence. And the data now exists within a complex corporate ecosystem, with connections running through Saudi Arabia's sovereign wealth fund.

It is also worth remembering what happened with Waze. The navigation app built by volunteers who reported traffic conditions was later used by law enforcement agencies to obtain information about driver locations. Data created for one purpose, repurposed for another.

Scopely and Niantic maintain that player data is stored exclusively on servers in the United States and that the company has never sold and will never sell data to third parties. Players' precise location data is kept for a limited time and used only for essential game operations, such as anti-cheat enforcement. And public space scans are anonymized before being passed to Niantic Spatial.

But today's guarantees are not necessarily tomorrow's guarantees. Companies change hands. Privacy policies get rewritten. And a database of thirty billion geotagged images with centimeter-level precision is not going to simply disappear.

The partnership with Coco Robotics is just the first declared step. Delivery robots equipped with cameras will generate even more visual data that feeds back into the system. Every food delivery made by a sidewalk robot improves the map. Niantic Spatial openly talks about building a living map of the world, one that continuously updates with new data.

If it worked with Pokémon GO, it can work with any augmented reality app. Smart glasses, urban games, navigation tools ... anything that points a camera at the real world and runs software capable of recording what it sees.

The future of autonomous navigation will probably not be defined by satellites in space. It will be defined by images taken from people's pockets.

What to do with this information

If you play Pokémon GO or any augmented reality app, it is worth checking which camera and location permissions are active. Disabling scanning tasks is a concrete option for anyone who does not want to contribute to the collection. It does not affect the core game experience.

If you work in technology, privacy, or regulation, this case is an important study on how generic terms of service can provide legal cover for future uses of data that the original users would never have anticipated. The gap between formal consent and informed consent remains enormous.

If you are an investor or follow the tech market, Niantic Spatial deserves attention. The company is sitting on a data asset with no equivalent in the market. Robotics companies, augmented reality firms, and autonomous navigation startups will need high-precision visual maps, and Niantic has nearly a decade-long head start in collection.

If you are simply someone who uses the internet, the lesson is broader. Every time a service is free and a bit too fun, it is worth asking: what am I giving away in return? The answer is almost never just your attention.