Update: It looks like the Smart Reply feature is rolling out starting tomorrow.
Google’s Inbox app for Gmail is one of the best things to happen to personal email management since email was invented. Using Google Now’s power, it can automatically create calendar events, sort out your junk and priority emails and suggest reminders. Now it’s about to get a whole lot smarter…
Making use of the company’s deep neural network tech (commonly known as AI) Inbox by Gmail will start giving suggested responses whenever you receive an email. That means that, instead of typing out entire sentences, it’ll offer three suggestions it thinks you’re most likely to use based on the email you received. The update will hit the app later this week and joins a whole host of other AI-powered moves made by Google recently, including YouTube thumbnails, better voice search and nightmarish picture editing, among others.
In its blog post about the update, Google explains how the technology works, stating that as the email comes in, one network encodes the email by consuming the words one at a time and then produces a vector, or list of numbers. This vector essentially captures the message of what the email means. A second network then uses this ‘thought vector’ and creates a grammatically correct reply one word at a time. To the user, it should feel fast, intuitive and natural. But the process of building this capability wasn’t without its challenges:
One challenge of working with emails is that the inputs and outputs of the model can be hundreds of words long. This is where the particular choice of recurrent neural network type really matters. We used a variant of a “long short-term-memory” network (or LSTM for short), which is particularly good at preserving long-term dependencies, and can home in on the part of the incoming email that is most useful in predicting a response, without being distracted by less relevant sentences before and after.
Of course, there’s another very important factor in working with email, which is privacy. In developing Smart Reply we adhered to the same rigorous user privacy standards we’ve always held — in other words, no humans reading your email. This means researchers have to get machine learning to work on a data set that they themselves cannot read, which is a little like trying to solve a puzzle while blindfolded — but a challenge makes it more interesting!
Since testing, Google’s engineers have been surprised by how good the neural network-powered Smart Reply system is at getting it right. If you have the Inbox app installed on your smartphone, you’ll be able to download the AI-infused update from next week. If you haven’t used it yet, you can grab it from the Play Store for free.