When it comes to machine learning, we tend to think big, like Amazon’s logistics and delivery system, or Google's Search function. Little wonder that this is the case – many of us still believe, deep down inside, that Skynet is coming. So, here at the Loves Data office, we were surprised to see an immensely useful (but seemingly diminished) application of machine learning in Google’s new Inbox app. Termed 'Smart Reply,' the function writes a selection of short replies to emails for you. You can then choose a reply option, edit it if needed, and send it off with ease. The whole idea stemmed from the need to address slow typing speeds on mobile. While the new function is a sibling to the much beloved and sometimes overused canned reply function, Smart Reply can do much more than repeat. It learns your communication needs and your personal style producing better and more accurate replies over time. The hope is that one day you won’t need to write short reply emails for setting up meetings or acknowledging receipt.
The development of this feature was no small feat, as the nuances of human speech and writing are extremely complex, with even small differences affecting meaning and tone. Google's Research Blog has a new post about the creation process, and has given some unique and interesting insight into the challenges in the development of this smart little program. Our favourite early development quirk? In an early version, the program had an unusual propensity to reply ‘I love you’ to even the most mundane emails. We’re wishing they hadn’t fixed that one!
Smart Reply is rolling out in the Google Inbox app on both iOS and Android platforms so go give it a try, and see if a robot could write your emails for you.