AI Can Now Find You In Photographs

(Scypre.com) – Artificial intelligence (AI) has emerged as a pivotal force in the technology sector, captivating the attention and investments of companies and institutions eager to enhance its capabilities.

At Stanford University, three graduate students—Silas Alberti, Michal Skreta, and Lukas Haas—delved into the realm of AI through their project conceived during the course “Computer Science 330, Deep Multi-task and Meta Learning.”

Inspired by their shared interest in the popular Swedish game GeoGuessr, the trio sought to develop a system that could surpass human proficiency in geolocating photos. Skreta described the game’s premise, where players are placed in a random location on Google Street View and tasked with pinpointing the location on a map.

Thus, they embarked on creating their software, Predicting Image Geolocations (PIGEON), leveraging the image-analyzing system CLIP from OpenAI, which learns about images by processing text related to them.

To train PIGEON, the students compiled a database of around 500,000 Google Street View images and incorporated a classification tool to organize the images based on their global positions. Remarkably, PIGEON achieved a 95% accuracy rate in identifying the correct country of a given photo and consistently narrowed down the location to within 25 miles.

The success of the program lies in its ability to discern intricate details, such as variations in weather, soil composition, and other elements that often elude human observers.

In a head-to-head challenge against geoguessing expert Trevor Rainbolt, PIGEON emerged victorious, marking a significant milestone as the first AI program to outperform Rainbolt.

While the project showcases the potential benefits of AI, concerns have been raised by Jay Stanley, a senior analyst at the American Civil Liberties Union (ACLU). He acknowledges the program’s utility in identifying locations from photos but warns of its potential to unveil sensitive information, such as the whereabouts of individuals seeking privacy, exemplifying the powerful impact of AI.

Stanley prompts contemplation on the broader implications and ethical considerations, pondering what tech giants like Google could achieve in this domain.