New website shows you how much Google AI can learn from your photos

Software engineer Vishnu Mohandas decided to leave Google in more ways than one when he learned that the tech giant had briefly helped the US military develop AI to examine drone footage. In 2020, he quit his job working on Google Assistant and also stopped backing up all of his images to Google Photos. He feared that his content could be used to train AI systems, even if they were not specifically related to the Pentagon project. “I have no control over the future results this will enable,” thought Mohandas. “Shouldn’t I be more responsible now?”

Mohandas, who taught himself to code and lives in Bengaluru, India, decided to create an alternative photo storage and sharing service that was open source and end-to-end encrypted. Something “more private, healthier and more trustworthy,” he says. The paid service he developed, Ente, is profitable and claims to have more than 100,000 users, many of whom are already part of the privacy-obsessed masses. But Mohandas struggled to make it clear to a wider audience why they should reconsider relying on Google Photos, despite all the conveniences it offers.

Then one weekend in May, an intern at Ente came up with an idea: to give people a sense of what some of Google’s AI models can learn by studying images. Last month, Ente launched https://Theyseeyourphotos.com, a website and marketing stunt designed to use Google’s technology against itself. People can upload any photo to the site, which is then sent to a Google Cloud computer vision program that writes a stunningly detailed three-paragraph description. (Duck asks the AI ​​model to document small details in the uploaded images.)

One of the first photos Mohandas tried to upload was a selfie with his wife and daughter in front of a temple in Indonesia. Google’s analysis was comprehensive and even documented the specific watch model his wife wore, a Casio F-91W. But then, Mohandas says, the AI ​​did something strange: It discovered that Casio F-91W watches are often associated with Islamic extremists. “We had to adjust the prompts to make it a little more wholesome but still scary,” says Mohandas. Duck began asking the model to provide short, objective results – nothing dark.