On May 28, 2015, Google launched Google Photos as a standalone service, announced by Anil Sabharwal, head of the product. The headline consumer hooks were free, unlimited storage of high-quality photos and videos and automatic organization. As the announcement put it, “Google Photos automatically organizes your memories by the people, places, and things that matter. You don’t have to tag or label any of them.”
The feature that made it feel like magic was visual search powered by machine learning. Google promised that “with a simple search you can instantly find any photo - whether it’s your dog, your daughter’s birthday party, or your favorite beach.” Under the hood, this relied on the deep convolutional neural networks that had transformed image recognition after AlexNet won the ImageNet challenge in 2012. The same technology that classified benchmark images could now recognize the contents of an ordinary person’s camera roll.
Google Photos grew quickly, passing a billion users within a few years and becoming one of the most widely used demonstrations of computer vision in daily life. It let people who had never heard of a neural network search a lifetime of pictures by typing a word.
For business readers, Google Photos is a clean example of a research breakthrough reaching mass consumers within a few years. It also previewed the risks: an image-recognition system trained on imperfect data can make offensive mistakes at scale, a lesson the product would learn publicly soon after launch.