“Did Pokémon Go Players Unknowingly Help Create Niantic’s New AI Mannequin?”
“`html
Update (11/22/2024): A spokesperson for Niantic confirmed to Sport Developer that the studio has dedicated the last 5 years to developing AI technologies that power its Visual Positioning System (VPS) required for its Global Geospatial Model. When players agree to share their AR scans, that data contributes to the VPS, and Niantic assured that players’ personal information “[has not] been sold to anyone.”
They referred to the PokéStop scanning feature in Pokémon Go that assumes capabilities, allowing Niantic to “create precise, dynamic 3-D maps of real-world objects and their locations, aiding devices in understanding their environment in AR in real-time.” A blog released on November 13 explores an experimental Pokémon Go feature and explicitly credits Pokémon Go and Ingress players for “contributing a vast array of scans from around the globe, enabling us to deliver a new form of AR gameplay to our audience.”
According to the spokesperson, Niantic’s next move is to scale its small AI technologies into a more comprehensive, global system that “serves an even larger audience worldwide.”
Interesting tidbit: Niantic recently unveiled a Global Geospatial Model (GGM), which utilizes machine learning to “interpret a scene and link it to millions of other scenes globally.” However, according to 404 Media, the GGM was likely constructed by unknowingly enlisting players to do the studio’s work over several years through games like Pokémon Go.
In its blog, Niantic conveyed hopes that the GGM would “establish a shared understanding of geographical locations, and comprehend areas yet to be fully scanned.” It is based on the studio’s Visual Positioning System (VPS), which allows players to “orient themselves in the world with centimeter-level precision” and search for or map digital content in their actual location, even after leaving.
Both Pokémon Go and Ingress are augmented reality (AR) titles that gather geolocated images as players explore areas to catch Pokémon or discover player-created artwork. By Niantic’s own admission, its VPS has scanned over 10 million locations globally in the past 5 years, and “receives about 1 million new scans each week,” collected from its games—one of which has been quite popular for nearly a decade, implying that players may have unwittingly contributed to the development of its GGM.
Similar to a Large Language Model (LLM), Niantic’s model aggregates data from real-world locations, aiming to utilize this information to “enable computers to not only perceive and understand physical spaces but also to interact with them in innovative ways.” What distinguishes its data from something like Google Maps or Street View is the perspective: as Niantic points out, the information is “sourced from a pedestrian viewpoint and includes areas inaccessible to vehicles.”
As of July 15, 2024, Niantic’s privacy policy states it utilizes geospatial technologies and player recordings to “construct a 3D understanding of real-world locations, aiming to offer new types of AR experiences to our users.” Participation is optional and can be turned off by players at any time; however, it seems to be a “key component” of Niantic’s objectives for AR and its model.
“The transition from LLMs to LGMs is another phase in AI’s evolution,” the blog concluded. “The future operating system of the world will rely on the integration of physical and digital realities to create a foundation for spatial computing that prioritizes people.”
The gaming industry seeks to engage with genAI, for better or worse
Companies such as NVIDIA and OpenAI have faced criticism for their methods in utilizing whatever resources they could to develop their AI technologies. In August, 404 published a report alleging that NVIDIA conducted extensive scraping of copyrighted material from YouTube for their AI tools and reportedly dismissed concerns by asserting that their practices were “in full compliance with the letter and spirit of copyright law.”
Violating consent has been a significant point of contention regarding generative AI and similar technologies for some time: voice actors have expressed concerns about how their voices were utilized by modders creating content for popular games, voicing comparable unease that audio companies are pressuring them into allowing their performances to be used for.
Certain developers showcase advanced genAI technology in non-intrusive performing methods as a means of alleviating the development burden and uncovering innovative solutions to existing challenges. Nonetheless, top-tier executives speculate the technology may also be leveraged for concepting new gaming concepts or other creative potential.
Game Developer has reached out to Niantic for clarification on the acquisition of its geospatial data, along with transparency regarding the usage of that data. We will provide updates once a response is received.
About the Author
Contributing Editor, GameDeveloper.com
A native of Kansas City, MO, Justin Carter has contributed to several websites including IGN, Polygon, and SyFy Wire. In addition to Game Developer, his articles can also be found at io9 on Gizmodo. Don’t inquire about how much gum he has consumed, as the response will likely exceed what he is willing to admit.