Google, Quietly Building a Much Smarter email App

There are more ways to communicate online than ever before.
Yet millions of people still use email to interact with one
another.
You may use Slack to talk with work colleagues and
Facebook Messenger to chat with friends, but chances are you
still go — often begrudgingly — back to your email inbox
throughout the day.
There is no shortage of email apps to choose from, and many
of them, like the recently shuttered Mailbox, have made it
significantly easier to manage our hectic inboxes.
But none are quite like Inbox by Gmail. Yes, that Gmail —
the email service from Google.
On the surface, Inbox may look like just another cool email
app that happens to come from a giant tech company. But
don't be fooled — Inbox is actually an ambitious effort to
rethink how email will work a decade from now.
Writing your emails for you
Inbox has all of the basics you would expect from Gmail, like
the ability to bookmark (or pin) emails, move them to folders,
and batch delete. It also has power user features you may
know from other email apps, like the ability to snooze emails
for later.
But Inbox primarily serves as a playground for the Gmail
team to experiment with artificial intelligence, or as Google
calls it internally, machine learning.
Google uses machine learning to power everything from how
it sorts search results to how it filters the barrage of spam that
tries to get your attention.
But when it comes to Inbox, the most complex example of
machine learning is a feature Google announced in November
2015 called Smart Reply. By analyzing the contents of an
email, Smart Reply gives you three quick responses it thinks
you might want to choose from.
And it's the kind of thing that feels like magic when it works.
"I wasn't sure if it was even possible," Greg Corrado, a senior
research scientist at Google who worked on Smart Reply,
told Tech Insider. "But it was such an exciting opportunity
that we couldn't help but try."
Corrado, who belongs to the elite group of researchers who
make up the Google Brain team, had already worked on
machine learning used in spam classification and Gmail's
Priority Inbox, a feature that sorts important messages from
less important ones. He said that Smart Reply was born out of
the team wanting to know if machine learning could be used
to respond automatically to emails in a human-sounding way.
The feature ended up taking a year to be built.
When you open a new email in Inbox, Google's computers
summarize its text into a machine readable format called a
"thought vector," he explained. Each vector is informed by a
field of scientific research called sentiment analysis, or the
idea that a computer can understand the general tone and
intention of a piece of text.
The end result is an email from a colleague about vacation
times creating a thought vector with three possible responses
that are suggested in the app.
Google could have produced anywhere from one to 500
possible responses per email, but the team designed the
feature to be as simple and helpful as possible. "We really
wanted something that wouldn't try to say the same thing in
three different ways," Corrado said.
The natural extension of Smart Reply would, of course, be
generating responses longer than one line and then eventually
entire conversations. Could Smart Reply one day analyze an
entire email thread, learn the context of a conversation, and
write paragraph responses for you? Corrado laughed when
I posed the question.
"I think that if we could build such a system that would
amount to passing the Turing Test," he said, referencing a
scientific test developed by Alan Turing in the 1950s that
determines whether a computer is capable of communicating
like a human being. "We're not anywhere close to that."
So for now, it's baby steps.
The future of email
There is perhaps no person better qualified to talk about the
future of email than Alex Gawley, who oversees Google
Calendar, Inbox, and the now 12-year-old behemoth that is
Gmail.
"The world has changed a lot since 2004," he said,
referencing the year that Google introduced Gmail. He
pointed out that the most popular web browser at the time was
Internet Explorer 6. Firefox had not been released.
Smartphones as we know them now were years away.
The idea that became Inbox started being discussed internally
at Google a couple of years before it was announced in
October 2014. "We felt that there was an opportunity to
rethink [email] and think about what might an inbox for the
next 10 years look like," Gawley said.
The Gmail team then set out to build a completely new email
experience with mobile and machine learning at its core.
"There's this transition from tools to assistants and computers
doing a lot of the work for you," he said. "So how can we
build an inbox that really did a lot more of the work for you?"
His team observed that Gmail users were spending lots of
time manually bundling and moving emails into different
folders and assigning different tags. So Inbox automatically
groups emails, whether they're newsletters or work updates,
into bundles to simplify the management process.
The team also saw that 80% of searches in Gmail
were looking for a fact in an email rather than the email
itself. So the group designed Inbox's search tool to get you to
an important fact or something you want to act on, like the
gate number for your flight. "You care that your flight
departure was changed, not the email underlying it," Gawley
explained.
The same principle is applied to a feature in Inbox called
Highlights, which automatically lifts the important
information out of an email and displays it above the rest of
the message. It's particularly useful for Google Calendar
invites because Inbox shows you all of the event info at a
glance and lets you accept or deny the invitation without even
opening the email.
It's the "inbox that helps you get back to everything"
ethos Gawley described that informs how the app integrates
with Google Reminders, a smart to-do list. Not only can you
add a reminder while looking at an email and flip through
them at the top of your inbox, but Reminders will even
occasionally suggest a reminder for you based on the
contents of an email.