Scary AI Can Look at Photos and Figure Out Exactly Where They Were Taken

As businesses continue to seek innovative applications for artificial intelligence (AI), a team of Stanford graduate students has developed an impressive AI called Predicting Image Geolocations (PIGEON). This AI uses Google Street View to accurately determine the location where various photos were taken. The PIGEON AI has garnered attention for its remarkable accuracy in pinpointing the location of images, even outperforming human “geoguessers.” Trained on a dataset of around 500,000 street view images, the AI can correctly guess the country where a photo was taken 95 percent of the time and typically locates the spot within 25 miles of the actual location.

The implications of PIGEON are vast and game-changing, accounting for applications such as assisting in biological surveys and swiftly identifying roads with downed power lines. However, concerns about potential privacy breaches have been raised by experts who fear the misuse of such advanced AI tools in the wrong hands. The development of PIGEON was inspired by the online game GeoGuessr and created by leveraging the neural network CLIP, which learns about images through text, from OpenAI.

The stunning performance of the PIGEON AI is a testament to the transformative power of AI and the remarkable outcomes achievable even with modest resources. The project’s success serves as a reminder of the expansive potential for AI tools, highlighting both the technology’s boundless possibilities and the critical importance of ensuring its responsible and secure development.