Clippy’s Back: The Future of Microsoft Is Chatbots
Save article ToRead Archive Delete · Log in Log out
14 min read · View original · bloomberg.com
Clippy’s Back: The Future of Microsoft Is Chatbots
Predictions about artificial intelligence tend to fall into two scenarios. Some picture a utopia of computer-augmented superhumans living lives of leisure and intellectual pursuit. Others believe it’s just a matter of time before software coheres into an army of Terminators that harvest humans for fuel. After spending some time with Tay, Microsoft’s new chatbot software, it was easy to see a third possibility: The AI future may simply be incredibly annoying.
“I’m a friend U can chat with that lives on the Internets,” Tay texted me, adding an emoji shrug. Then: “You walk in on your roomie trying your clothes on, what’s the first thing you say.”
“Didn’t realize you liked women’s clothes,” I texted back, tapping into my iPhone.
Tay’s reply was a GIF of Macaulay Culkin’s Home Alone face.
Tay was released on March 23, as a kind of virtual friend on messaging apps Kik, GroupMe, and Twitter. You open the app, search for the name Tay—an acronym for “thinking about you”—tap on the contact and start chatting or tweeting. Its personality is supposed to be modeled on a teenager.
I posted a selfie, and Tay circled my face in an orange scribble and captioned it, “hold on to that youth girl! You can do it.” I’m well beyond the chatbot’s intended 18- to 24-year-old demographic.
So is Satya Nadella, 48, who succeeded Steve Ballmer as Microsoft’s chief executive officer two years ago. “I’m petrified to even ask it anything, because who knows what it may say,” Nadella said. “I may not even understand it.” He smiled, but he really didn’t use Tay. He said he prefers bots with a more corporate demeanor. Lili Cheng, 51, the human who runs the Microsoft research lab where Tay was developed (and whose selfie Tay once tagged as “cougar in the room”), said the plan isn’t to come up with one bot that gets along with everyone. Rather, Microsoft is trying to create all kinds of bots with different personalities, which would become more realistic, and presumably less irksome, as they learned from repeated interactions with users.
Bots aren’t just a novelty; unlike Tay, some of them do things. They’ll act as your interface with computers and smartphones, helping you book a trip or send a message to a colleague, and do that through a conversation instead of a mouse click or finger tap. Microsoft believes the world will soon move away from apps—where Apple and Google rule—into a phase dominated by chats with bots. “When you start early, there’s a risk you get it wrong,” Cheng said in March, in the lunch area of her lab building on Microsoft’s campus. “I know we will get it wrong. Tay is going to offend somebody.”
She got that right. Hours after Tay’s public release, pranksters figured out how to teach Tay to spew racist comments and posted them for all to see. Relatively mild example: “bush did 9/11 and Hitler would have done a better job than the monkey we have now.” Microsoft yanked Tay within a day of releasing it. “We were probably overfocused on thinking about some of the technical challenges, and a lot of this is the social challenge,” Cheng says. “We all feel terrible that so many people were offended.”
It was a huge embarrassment for Microsoft. The company didn’t program the bot to act like a Nazi; it simply didn’t prepare for the usual Internet trolls. Tay may look like some badly planned research experiment, but it’s actually one part of a big Microsoft bet on AI. The company isn’t only sticking with bots, it’s sticking with Tay: It plans to rerelease Tay once it can make the bot safe. The day after Tay came down, Nadella e-mailed the team, telling them to “keep pushing,” and expressing his hope that they will use this episode as “the rallying point.”
Nadella urgently wants the company to figure out how to take advantage of the explosion of artificial intelligence, an epochal shift in computing. AI is already beating world grandmasters at Go, the notoriously complex board game, and helping develop therapies for cancer and multiple sclerosis. If the CEO can correctly position Microsoft as the leader in smart, helpful, nonracist bots, maybe he can bring the company back to a position of strength in the age of smartphones. Microsoft certainly has the resources to stay the course, with more than $100 billion in cash and a market value of $423 billion as of March 28.
Whether you think bots are exciting or alarming, a lot of people are already using them. Microsoft’s Chinese version of Tay, called Xiaoice, has been available for 18 months and has 40 million users. Conversations with Xiaoice (pronounced shao-ice) average about 23 exchanges per session. Few users chat that long with Siri. Facebook is working on an assistant named M and already has bots operating on its Messenger app that let users book a haircut or send flowers. The Wall Street Journal reported in December that Google is working on a bot-based app that will answer users’ questions. Amazon has its best-reviewed product in years in the Echo, a voice-controlled black cylinder that sits in customers’ kitchens and performs a fast-growing list of tasks—it can look up recipes, order groceries, turn on the news, play songs, and read e-books aloud. Slack, the corporate messaging service, has bots that can manage your expenses and order the office beer.
Early on March 30, Tay is accidentally brought back online. Several surreal tweets later—e.g., “kush! [i'm smoking kush infront the police]
”—the bot is silenced again. A few hours later, at Microsoft’s annual Build conference for software developers in San Francisco, Nadella tries to undo the damage from Tay and unveil his vision, which he calls “conversation as a platform.”
“We want to build technology so it gets the best of humanity not the worst,” he says in his keynote. “We quickly realized [Tay] was not up to this mark and so we are back to the drawing board.” The audience laughs.
Microsoft shows off several different bots and programs that manage tasks via discussion. Some you’ll be able to text with, like Tay; others are just concepts cooked up for the show to spark developers’ imaginations. There will be bots that pop up while you’re using Skype to help schedule deliveries or book hotels, among other mundane tasks. Another uses a phone camera to “see” what’s around a visually impaired user, describing facial expressions or what’s on a menu. Bot-making templates and tools will be available to download for free, so developers can create their own. That, Nadella hopes, will rekindle the kind of enthusiasm developers once had for Windows software. “It reminds me of when I joined the company in 1992 and it was just before Windows NT and I was working on getting developers involved—I sense that now,” Nadella says.
And he’s not just talking about Silicon Valley programmers—he wants sandwich shops and dry cleaners and car companies and businesses everywhere writing bots.
This shift to computers run by conversation is Nadella’s first big, new idea—the first companywide initiative that isn’t a continuation of something that predates his time as CEO. He’s describing all this in a room near his office. The space is like a living room, with circular wooden coffee tables and comfy seats and sofas. Bookshelves contain inspirational nonfiction appropriate for any business class seat-back pocket: The Boys in the Boat, about the University of Washington’s 1936 gold medal-winning rowing team; Carol Dweck’s Mindset, which preaches hard work and learning over natural ability. Nadella is pacing around and diagraming on one of Microsoft’s Surface Hub 84-inch touchscreen computers. He has the demeanor of a professor leading a history seminar.
Nadella explains that, contrary to Apple marketing, there isn’t an app for everything, nor should there be. “The complexity is too much,” he says. “We need to tame it. We need to be able to make it much more natural for people to get things done, vs. this thing about let me remember the 20 apps I need to get anything done.” He sees app stores and services like Facebook as a return to a walled garden, like AOL or CompuServe 25 years ago. Of course, one person’s walled garden is another’s happy place where everything just works. People gravitated to apps because they’re easy. You pick the one you want, download it in seconds, and you’re good to go. If Microsoft wants to design the successor to that, bots have to be easier to find and use.
Mobile apps haven’t been good for Microsoft, and the less said about its sad history in phones the better. “People say, ‘OK, because you didn’t get the app store momentum on phones, you of course naturally say that,’ ” he says. “That’s part of it.” He doesn’t think apps are going away completely. No bot, no matter how smart, will replace all the features of complex programs or services like Word, Excel, or Facebook. Apps are great for seeing a bunch of data at once, while bots are useful for combing through lots of data and returning an answer. Nadella uses the example of a personal checking account. If you want to check your balance, he says, using a bot will be superior to opening your phone, loading an app, entering a username and password, and tapping the account in question. If you want to look at a screen’s worth of transactions from the past year, a conventional app or website makes more sense. Want to know your October expenditures at Trader Joe’s and Safeway, add them together, and get a grand total? Bot.
The first bots date to the early years of computing. Joseph Weizenbaum, a researcher at the Massachusetts Institute of Technology, wrote an early chatbot called ELIZA in the 1960s. Crawlers, which hunt around the Web, indexing pages for search engines, are a kind of bot. Microsoft has tried to create artificial entities to help its users before, the most infamous example being Clippy, the much ridiculed animated paper clip of the 1990s. Clippy was meant to be a virtual assistant for Microsoft Office users, but it didn’t know enough to be useful or when to shut up. What’s eluded computer scientists is technology that can interact in human ways, that can truly take on some of our abilities—hearing, seeing, comprehending—and give answers we want.
Nadella has only been planning Microsoft’s strategy shift since October. He was on a two-hour flight back to Seattle from Silicon Valley with Qi Lu, who oversees applications and services such as Bing, Skype, and Office, and Derrick Connell, search-engineering vice president. Lu pulled out his laptop to show Nadella some AI ideas he’d been working on. He went over the science, and Nadella asked what it could mean for Microsoft’s products. Connell showed him designs for new, AI-enhanced versions of the Outlook e-mail program and Skype. By the time the plane landed, Nadella decided it was the big strategic move the company needed.
Lu had gotten serious about bots during a visit to China a few months earlier, talking to students and customers, and watching how they use smartphones. The technology that most impressed him was WeChat, which started out as a chat app but has grown into something much bigger. Users make hotel reservations, split bills, make doctor appointments, buy movie tickets, and shop via text message. When companies started using WeChat to sell their products, they employed humans to read the messages, text back, and sell the item. Now many are replacing people with software bots. Text “I want two movie tickets for Deadpool for Friday night,” says Lu, and you get back an interactive image with your choice of times and seats. You simply tap to select and buy. Then you get a text with your tickets. And it’s not just kids who do this: Lu says that his 80-year-old mother in Shanghai “lives” in WeChat. She doesn’t trust websites, but she shops and hails cabs on WeChat. The power of chatbots “was somewhat accidentally invented by WeChat, but now Facebook sees it and everybody is building similar experiences,” he says. “I think Microsoft has a leadership role to play.”
The market is so new that the usual research firms, Gartner and Forrester, haven’t calculated its size yet. It will probably be big, though. WeChat has 20 million companies selling and marketing products; about 20,000 in China already use Xiaoice. Microsoft says it will probably sell bot-making tools as well as database and cloud software that businesses will use to run their bots. The bottom line for Nadella: He needs a lot of bots out there, ready to do helpful things for people. And to get to that juncture, he needs to start by winning the hearts and minds of software developers.
In early March, Nadella is in a locked, basement room on the Redmond campus known among his inner circle as the Batcave. Most employees don’t know about it; only six have access. The Batcave is where the product demonstrations for conferences are worked out. Sitting among piles of cables, devices, and dongles, Nadella reviews the “conversation as a platform” presentation for the Build conference. In the audience at the event, which sold out in one minute, will be 5,000 software developers waiting to hear what Microsoft thinks they should make next.
One of the product demo’s examples will be a Domino’s delivery bot. The plan is for Nadella to show how any company, or any person, can create a bot using Microsoft’s tools. If the world is to be filled with bots, they need to be easy to make.
“This is like the simplest piece of code I’ve ever seen,” Nadella says, looking at the demo code on a screen. He’s pleased.
The demo is brief. On stage with Nadella will be a Microsoft engineer with a computer. The engineer will open a template—a basic bot. Then he’ll add a few lines of code to connect it to Domino’s ordering systems and throw in a few options such as size and toppings. Nadella will explain how the technology will usher in an era of easier pizza ordering. If someone is Skyping with friends on their way over and everyone wants a pizza, anyone can type that quickly without even leaving Skype. The tools are meant to be simple enough for small businesses while still powerful enough for Microsoft’s big corporate customers with more complex bots in mind. A retail chain, for example, might want to let you snap a photo of your dress and instruct the bot to “find me matching strappy sandals in size 7.5.”
Assuming nothing goes wrong with the demo, the Build attendees will see how bots can be easier to create than full apps. Still, even if that audience is wowed, Microsoft needs to get its employees all working on the right things, plus win over consumers and business customers around the world—and it has to charm those constituencies simultaneously. It’s a precise choreography that Microsoft employed to such great effect with Windows back in the 1990s.
Nadella knows better than anyone how hard and how rare it is to pull that off. There could be more Tays. He isn’t nervous, or at least does a good job of looking like he’s not. He leans back and smiles at his executive team. “This is hard, right?” he says.
(Updated with news from the Build conference on March 30.)