Today, Google released TensorFlow Privacy, an open-source tool that will help keep your data anonymous, even as AI learns from it. The now-public code is based on differential privacy. That's what allows Gmail's Smart Reply to guess what you're going to say by collecting data from other people's emails, and at the same time, keeps Smart Reply from revealing any juicy secrets people have typed before.
Differential privacy is not new. In fact, it's fairly common. Essentially, it makes sure AI cannot encode information that is unique to you and could therefore reveal your identity. Instead, AI only learns from patterns that show up en masse. What thousands of people type into Gmail might become a Smart Reply auto response, but the personal data you enter will never show up in a stranger's email.
By sharing TensorFlow Privacy, Google hopes developers will add this type of security to other machine learning tools -- and maybe even improve on it. To encourage adoption, Google promises TensorFlow is easy to use. It only requires "some simple code changes" and hyperparameter tuning. You can access the tool on GitHub, and if you want to dive deeper, Google has also released a technical whitepaper.