Software engineer Vishnu Mohandas decided to leave Google for multiple reasons after learning that the tech giant had briefly helped the US military develop artificial intelligence to study drone footage. In 2020, he quit his job working on Google Assistant and also stopped backing up all his photos to Google Photos. He was concerned that its content could be used to train artificial intelligence systems, even if they were not specifically related to the Pentagon project. “I have no control over the future consequences this will make possible,” Mohandas thought. “So now, shouldn’t I be more responsible?”
Mohandas, a self-taught coder who lives in Bengaluru, India, decided he wanted to develop an alternative photo storage and sharing service that was open source and end-to-end encrypted. Something “more private, healthy and trustworthy,” he says. The paid service he designed, Ente, is profitable and has over 100,000 users, many of whom are already part of a privacy-obsessed community. However, Mohandas had difficulty explaining to a wider audience why they should reconsider using Google Photos, despite all the conveniences it offers.
One weekend in May, an intern at Ente had an idea: give people an idea of what some of Google’s artificial intelligence models could learn from analyzing images. Ente launched last month https://theyseeyourphotos.comwebsite and marketing ploy to turn Google’s technology against itself. Users can upload any photo to the site, which will then be sent to a Google Cloud computer program that will write a surprisingly true, three-paragraph description of it. (Ente asks the AI model to document tiny details in the uploaded images.)
One of the first photos Mohandas tried to post was a selfie with his wife and daughter in front of a temple in Indonesia. Google’s analysis was exhaustive and even documented the specific model of watch his wife wore – a Casio F-91W. But then, Mohandas says, the AI did something strange: it noticed that the Casio F-91W watches were so commonly associated with Islamic extremists. “We had to tweak the prompts to make it a little healthier, but still scary,” says Mohandas. Ente started asking the model to provide brief, unbiased results – no murky results.
The same family photo uploaded to Theyseeyourphotos now returns a more generic result that includes the name of the temple and “partly cloudy skies and lush greenery” surrounding it. However, the AI still makes a number of assumptions about Mohandas and his family, such as that their faces express “shared contentment” and that “the parents are likely to be South Asian and middle class.” He assesses their attire (“suitable for sightseeing”) and notes that “the woman’s watch reads approximately 2:00 p.m., as confirmed by the image’s metadata.”
Google spokesman Colin Smith declined to comment directly on the Ente project. Recommended by WIRED to support pages in this state, files uploaded to Google Photos are only used to train generative artificial intelligence models that facilitate people manage image libraries, such as those that analyze the age and location of people photographed. Business says does not sell the content stored in Google Photos to third parties or operate it for advertising purposes. Users can turn off some analytics features in Photos, but they can’t completely prevent Google from accessing their photos because the data isn’t end-to-end encrypted.