A Stanford student project called PIGEON (Predicting Image Geolocations) has showcased the impressive geolocation capabilities of artificial intelligence (AI). The project, inspired by the online game GeoGuessr, aimed to identify locations in Google Street View images. Using an AI system trained on 500,000 street view images, the students achieved remarkable performance. PIGEON correctly guessed the country 95% of the time and could usually narrow down locations to approximately 25 miles.
The positive applications of such technology include aiding in the identification of locations in personal photos and supporting ecological surveys, among other beneficial uses. However, concerns have been raised regarding privacy implications. The technology could be misused for government surveillance, corporate tracking, or even stalking, highlighting the potential double-edged sword nature of AI advancements.
The Stanford students demonstrated the power of their AI by pitting it against a seasoned human geoguesser, Trevor Rainbolt, who is known for his expertise in the field. PIGEON outperformed Rainbolt in a head-to-head competition, showcasing the potential superiority of AI in certain tasks.
Despite their success, the students are aware of the privacy risks associated with such technology. They have documented their technique in a paper but refrained from making the full model publicly available to address these concerns. The ACLU’s Jay Stanley emphasizes the growing impact of AI geolocation capabilities, noting that companies like Google already use AI for location estimation with a limited catalog of landmarks. He urges individuals to be conscious of the background details in photos posted online, as AI’s geolocation capabilities are likely to advance further in the future.
