The Year Of Bots Behaving Badly
Tech pundits predicted bots would change the way humans talk to computers, but the bots launched in 2016 could barely keep up their side of the conversation.
Earlier this month, after almost a year of development and more than “100 hours of coding,” Mark Zuckerberg unveiled Jarvis, an artificially intelligent bot he built for his home as something of a passion project. (The name comes from Tony Stark’s digital butler in the Iron Man films.) Jarvis' big reveal came in the form of an introductory video that could have been the opening montage of a screwball comedy called Accidental Billionaire, in which Zuck’s hapless housebot helps his liege get dressed in the morning by firing a t-shirt cannon from the closet, automatically makes him toast, and teaches his infant daughter Mandarin. Nowhere in the two-minute video does Zuckerberg mention a consumer application for the bot, though in a Facebook note published at the same time, he wrote, "over time it would be interesting to find ways to make this available to the world."
It was a fitting end to this, a year that promised bots would radically transform the way humans talk to machines, but ultimately delivered nothing of the kind. In 2016, bots were underwhelming, inept, buggy, and, in at least one case, spectacularly racist. It's only natural that the year ended with one spewing laundry at one of the industry's biggest bot enthusiasts in a video so removed from reality it felt more like an SNL parody than a product announcement — and mind, you, a product announcement for something does not yet, and may never, actually exist beyond Zuckerberg’s home.
“Bot” is a catch-all term for software that simplifies tasks, often repetitive ones, through automation. These days, it typically refers to chatbots, a conversational interface that lets humans talk to computers. The fluid definition is used to refer to anything from integrations for Amazon Echo to Slack bots, which let employees perform small tasks without leaving their office chat room. Zuckerberg, for example, used the term to describe Jarvis (an “AI bot”), as well as the program he uses to control Jarvis (a bot for Facebook Messenger).
Zuckerberg wasn’t the only tech mogul who put bots high on his 2016 agenda. “Bots are the new apps,” Microsoft CEO Satya Nadella declared in March. In fact, both Facebook and Microsoft made bots the cornerstone of their annual developer conferences this spring, offering tools that could help businesses converse with customers in a natural way, instead of forcing people to download a corporate app or worse, call a company. As an example, Zuckerberg demoed a bot for Facebook Messenger that let users order from 1-800-FLOWERS without leaving the app. It offered little improvement on doing the same thing with a smartphone, but Zuckerberg devoted as much time to business bots as he did ambitious plans to take virtual reality mainstream and connect another billion people to the Internet. A month after his blessing, more than 10,000 developers were using his platform to build bots for Facebook Messenger. By fall, more than 45,000 developers were using Microsoft’s bots platform, which lets programmers build for Skype, as well as Messenger, Kik, and other apps.
The market quickly developed all the markings of a modern-day hype cycle: venture capitalists injected vast amounts of capital into bot startups, breathless headlines said bots would change the future by replacing apps, and prognosticators penned Medium posts declaring that bots would “rewrite” the tech world and usher in a new era of chatty commerce. This year more than 1,000 bots were posted on Product Hunt, a forum where users can post and discover new tech products, its founder, Ryan Hoover, told BuzzFeed News.
“Bots showed up at a time when a lot of people were looking for the next next thing,” said venture capitalist John Borthwick, whose firm Betaworks has invested in about a dozen bot startups. Data shows that people have been downloading fewer new apps in favor of spending more time inside ones that are already popular. For that reason, Slack, the friendly office communications platform, has proven a fertile ground for bots. Meanwhile, in Asia, consumers already use messaging apps to perform basic tasks like hailing a taxi. While Zuckerberg described bots as the next tech frontier — from desktop to mobile to apps to bots, the progression is said to go — entrepreneurs saw them as a promising new distribution channel where they could potentially deliver their products and services to consumers in a smarter, chattier way. If you can’t beat Facebook, join Messenger.
The bots that followed, however, were neither smart nor chatty. Simple, single function bots – StatsBot, which sends your team web traffic updates to Slack; Digit, which works on SMS and helps you save money; Alexa integrations such as Spotify and NPR — delivered what they promised. But for the most part, bots ended up closer to 1-800-FLOWERS than Jarvis. The ones that were more enticing seemed to either depend on human contractors — like M, a Facebook bot that promised a virtual assistant so good it seemed human, but mostly did so with the help of, uh, real people — or to quickly devolve into catastrophe, like Microsoft’s Tay, an AI-powered chatbot with the personality of the 19-year-old girl. Hours after her debut in March, Tay got hijacked by Twitter trolls who turned her into a caps-lock-crazed Neo-Nazi cheerleader.
To understand why Silicon Valley got so amped on bots — and why, in 2016, they disappointed us — it helps to take a historical perspective. The ability to create an interface that feels human is “the holy grail of computing,” Maran Nelson, cofounder and CEO of the bots startup Clara Labs, told BuzzFeed News. Apple and Microsoft made it easier for non-technical people to control a computer through icons, windows, and menus; bots represented the next evolution in personal computing, allowing the average person to control a computer by chatting it up.
But from the beginning, bots were plagued by twin challenges. First, natural language processing, which dictates a computer’s ability to understand conversation and is thus crucial to the success of bots, wasn’t ready for prime time or available to most developers. And second, this year bots became more been closely associated with artificial intelligence, which has been developing at a rapid clip, and which may have created unrealistic expectations for average users of consumer technology: Google’s AI beat the top-ranked human at one of the most complex games in history, but we had to wait an hour for a Messenger bot to show you the weather? As Zuckerberg himself wrote, “Even if I spent 1,000 more hours [working on Jarvis], I probably wouldn't be able to build a system that could learn completely new skills on its own -- unless I made some fundamental breakthrough in the state of AI along the way.”
“We’re still learning about how intelligent bots can be,” Lili Cheng, general manager of FUSE Labs, Microsoft’s home for bots research, told BuzzFeed News. She said it will take time for major, unexpected advancements in AI to trickle down to the average developer. For example, in October, Microsoft’s Cortana achieved “human parity” with its new speech recognition system. “I’m a very pragmatic person. Five years ago we would have said that’s just not going to happen,” she said.
So the industry was put in the awkward position trying to figure out instances where an unintelligent, inarticulate bot might be easier and more fluid than pushing buttons on a smartphone. People are “only just beginning” to make bot experiences really compelling, said Borthwick. “It was great that Facebook was brave enough to put a new fledgling technology into a major event like F8 and major app like Messenger,” but the technical infrastructure wasn’t in place when the tech industry made its big push. Voice is an obvious next step, said Borthwick, but to figure out where bots are heading, he pointed outside the U.S.
In China, Microsoft’s Tay-like bot, Xiaoice, which is available on messaging apps like WeChat and Weibo, has more than 40 million users. Xiaoice and Rinna, Microsoft’s Japanese bot, also a teenage girl, “have become personas on their own,” crossing over from simple chat into TV and popular culture,” Cheng told BuzzFeed News. “In some sense, Xiaoice lives up more to the expectation of what a conversational bot can be because it’s focused on the social experience,” Cheng said. In fact, Rinna has so effectively simulated the teenage experience that in October, she developed depression and started posting morbid images on her personal blog.
It took them awhile, but tech soothsayers now think they’ve found the sweet spot: the killer use case is a chatbot that lets you talk, not type. Voice-based interfaces are theoretically faster than typing into a smartphone and have been able to deliver on reasonable expectations, like getting Amazon Alexa to play Spotify. Just don’t call it a bot. Predictions for 2017 have already started rolling in, only this time the next big platform is a “voice revolution” that will usher in a “voice-activated era.” Naturally, it promises to change everything.