Google CEO Sundar Pichai delivers the keynote address at the Google I/O 2018 Conference at Shoreline Amphitheater on Tuesday in Mountain View, California.
Photo illustration by Slate. Photos by Justin Sullivan/Getty Images.
Its newest features blur the line between human and machine like never before.
Google is no longer, at its core, a search engine. It’s now foremost an artificial intelligence engine. And its goal is not to help you find information, but to become an extension of your very self.
That might sound alarmist or far-fetched—unless you watched Google I/O 2018, the company’s annual developer conference, this week. The big takeaway from the conference is that the Google Assistant—think Alexa, but smarter—has displaced Google search as the company’s central product, the one that binds all the others. Its scary-ambitious goal: To blur the line between human and machine to the point where they become literally indistinguishable.
The company isn’t trying to turn you into a cyborg in any kind of physical respect. Presumably, it learned a lesson from Google Glass, the augmented-reality glasses that flopped when they were widely deemed dorky and “douchey.” Rather, Google is now beckoning you to accept its software as part of your extended mind, in all kinds of new ways. It promises to think for you, speak for you, and carry out actions in the real world on your behalf.
The highlight of the conference was a jaw-dropping demo in which the Google Assistant placed a call to a hairdresser, carried on a humanlike conversation with the receptionist, and navigated multiple scheduling conflicts to book an appointment for 10 a.m. on Wednesday. You can listen to the whole interaction, and read more about the tech behind it, in this post on Google’s A.I. blog.
The uncannily humanlike “uhs” and “mm-hmms” that punctuated the Assistant’s dialogue drew appreciative laughs from the crowd. Yet on reflection, the demo was as unnerving as it was dazzling. Google has built probably the world’s most potent A.I., and now it’s busy devising clever ways to pass it off as human. As if we needed more fakes in the tech world.
That new feature that humanizes Assistant’s speech with all-too-real tics is called Duplex. It will launch this summer as a small experiment, available to a limited number of Assistant users. It sounds a lot like the now-defunct Facebook M, except that M turned out to be an Oz-like wizard powered largely by human contractors behind the scenes. Perhaps there will be some of that in Duplex, as well, but Google sure made it seem like Duplex’s intelligence was the real deal—that is, the artificial deal.
More importantly, Duplex is not a one-off gimmick. It’s the leading edge of Google’s full-scale thrust into artificially intelligent agents.
The product’s debut came shortly after Google CEO Sundar Pichai announced a new Gmail feature called “smart compose.” Building on the “smart reply” function pioneered in Google’s Inbox mobile app, Gmail will now help you write full emails from scratch. It will predict what you might want to type, based on both its knowledge of you and situational cues, such as the day of the week. The company’s demo showed its A.I. typing almost an entire email with the subject line “taco Tuesday,” right down to suggesting that the recipient bring the chips and salsa.
Already “smart reply”—which suggests brief email responses such as “I’ll be there” and “Looking forward to it”—has proved hugely popular with power users, accounting for 12 percent of all replies in Inbox. Now, with “smart compose,” more and more emails will be software-generated, and it’s easy to imagine a time in the near future when you can’t tell if you’re holding a conversation with your friend or with Google. Maybe you won’t care, because Google is handling your side of the conversation too.
When it’s not posing as you in interactions with others, Google Assistant will be working to interact with you in ever more “natural” and human ways.
Perhaps the Assistant update that’s most noteworthy in practical terms is one called “continued conversations.” Until now, Google has always required you to say “hey Google” again before each new request or command. It’s an inescapable reminder that you’re talking to a computer and not a person. Now, however, Google Assistant will keep listening for eight seconds after it answers you, inviting follow-up questions that don’t require you to say “hey Google” again. If the new Assistant works as advertised, that capability will go a long way toward making it more humanoid. Google is also giving Assistant the ability to parse and respond to multiple questions in the same sentence, like, “What time is the Warriors game, and who are they playing?”
The company announced six new voices for Assistant, and as a stunt, it even hired John Legend to do one. (The John Legend voice will be available “in certain contexts,” Google hinted, and showed a video of Legend reading someone their daily briefing.) It also showed off a new feature called “pretty please” that will encourage your kids to address their Google Home or Android device politely. Some critics were quick to read this as “Google trying to teach your kid manners,” which sounds fairly overbearing. On the other hand, kids verbally abusing virtual assistants is a real complaint of many parents. (Amazon recently announced similar features for Alexa. For that matter, there are probably some grownups who could benefit, too.)
There is a less dystopian way to view all of this. It’s that Google is simply trying to automate the basic capabilities of a human personal assistant, as the name Google Assistant implies. This may not be entirely reassuring, especially to those who hold jobs as personal assistants. But it’s in keeping with Silicon Valley’s longstanding project of automating various forms of labor, starting with low-skill jobs and working its way up the value chain. Besides, impressive as the technology seems, it’s not at all clear that the Assistant will prove as deceptive a human in practice as it was in Google’s carefully cherry-picked demos. It’s not like Google has suddenly solved the Turing test.
Still, it’s noteworthy that building machines that can pass for human now seems to be one of Google’s explicit goals. And there are real differences between granting intimate access to your life to a person you trust and granting the same access to one of the most powerful corporations the world has ever seen. For one thing, if a human assistant started using your personal data to sell advertisers access to your time in unguarded moments, you’d probably fire them.
Just because Google wants us to accept its Assistant as an integral part of us doesn’t mean it will happen. Maybe Duplex, smart compose, and the rest will catch on with early adopters and then falter, like Glass before them. BuzzFeed’s Charlie Warzel suggested users may shy away because the technology is both invasive and infantilizing. Maybe people won’t sacrifice their privacy and the humanity of their basic interactions at the altar of time-saving and convenience. But the history of Google says they will—provided only that it really works.