We already use machine learning to outsource our labor, our love, and our memories, so it’s not surprising that Google is using machine learning to automate a daily communication we really wish we didn’t have to deal with so often: our office email replies.
On Wednesday, Google announced that they were expanding their Smart Reply feature to all Gmail users. Introduced in 2015 to Google’s Inbox app, Smart Reply suggests automated responses to emails, which, we’ve noticed, are far friendlier than the replies we would normally tap out.
Rich with exclamation points and cheery tone, Google’s technology can really reduce office friction.
Now, Smart Reply will suggest three responses after reading your email. You can use them immediately or edit them. If your email is about scheduling a meeting next week, Smart Reply may suggest: “Let’s do Monday,” “Monday works for me,” or “Either day works for me.” (This beats our usual responses of “k” and “sure.”)
Google’s neural networks are not just good at answering yes/no questions, they are learning how to pick up emotional tone, the most complex part of human language for a machine to understand.
According to Google, using algorithms to write your emails is gaining popularity. 12% of replies in Inbox on mobile are currently machine-generated, Google’s Greg Bullock said.
— Google (@Google) May 17, 2017
How smarter email replies work to read your mind
The technology behind Google Google engineers said that if someone emailed you about “that interesting person at the cafe we like gave me a glance,” Smart Reply would weigh each part of information that sentence reveals.
That could mean knowing that “we” indicates you and the sender, or considering whether or not “glance” is a positive or negative gesture.
It’s still a work in progress, but Google believes that the “system can find responses that are on point, without an overlap of keywords or even synonyms of keywords.” Currently, Smart Reply is only available in English.
For scheduling mundane tasks, go ahead and outsource your thoughts to a machine. I would not be hurt if I received generic responses ghostwritten by Google about my colleague’s meeting. But for harder, complex negotiations between colleagues—write it out yourself. Although that’s my personal opinion.
Other users are conceding to Smart Reply’s intuitive power to suss out what we mean.
Writer Ben Crair noted that Smart Reply was perceptive about what he wanted to say after a breakup. He had written a messy email of contradictory feelings to his ex-girlfriend, while Google had seen through him and simply suggested he reply: “I miss you so much.”
Google’s senior research scientist told Crair that Smart Reply learns how to notice patterns and reply to every scenario—from colleague requests to breakups—by reading through anonymized past emails Google keeps. (Does Google scanning your emails sound creepy? That’s part of the terms of service we sign off on when we sign up for Google. You can’t opt out of that unless you quit Google, but you can opt out of Smart Reply by turning the option off on mobile; your emails will still get scanned, but you won’t have the ready replies pop up to send to people.)
Alan Turing’s famous 1950 test said that for a machine to be intelligent, a human should be unable to distinguish whether or not the conversation had come from a human or a machine. As Smart Reply’s existence shows, whether we like it or not, we’re getting there.