How Kik Predicted the Rise of Chatbots, by Backchannel
Adam Rifkin stashed this in Chatbots
Are Chatbots the next big user interface?
Like its bot-platform competitors, Kik is cobbling together bits and pieces from the half-century history of chatbots, the annals of artificial intelligence, and the “internet of things” as it tries to solve a big problem: Users want to do more and more with their phones, but they’re downloading new apps less and less often.
Chatbots, according to the narrative of the day, are a new “conversational interface” replacing apps (which replaced the browser, which replaced the desktop). Since so many of us are already spending much of our time messaging with human friends, the argument goes, we should not have to leave that environment to do things like summon an Uber ride, move money between banks, or order food. We should tell a bot to do these things, and the bot should make it so.
But when I recite that spiel to Ted Livingston, Kik’s affable founder, he pushes back — not because he thinks the idea is hype, but because he thinks it’s too narrow. “We look at chat, very simply, as a better way to deliver software,” he says. “So to us, customer service and selling items on the web are interesting applications of chat — but they’re only a small subset of the type of experiences we’ll be able to deliver.”
In the bot gospel according to Kik, chat is the new distribution mechanism for all things digital — the portal through which we will access businesses, news, entertainment, games, and personal services, as well as one another. “I think the future is that every interaction will be powered by chat, in the same way that every piece of information is now on the web. It will become the operating system for society.”
Livingston’s iconic example is ordering hot dogs and beer from your stadium seat. Forget waiting on line and missing a big play. Forget downloading the ballpark’s app that you’ll almost never use again. Just scan the code next to your seat and message your order to the stadium bot, right from the Kik chat window where you’re already messaging with your friends. When you’re done, it will disappear, until your next visit. “Bots are to apps as dating is to marriage,” says Josh Jacobs, president of Kik Services, who works out of Kik’s new Los Angeles office.
Thanks to WeChat, this interaction model is already ubiquitous in China, where the bots that represent businesses are known as “official accounts.” But it took Kik a few years and false starts to arrive here, including a push to weave partners’ web content directly into chats. “We’re sitting in the next room, right over there, and we’re thinking, people don’t want to build web apps, we’ve tried to make it simple for years, it’s not working,” says Livingston. Then, he recalls, his cofounder and CTO Chris Best had an idea: What if apps were chats themselves? “It was this lightbulb moment that connected a million things.”
Two years ago, Kik started inviting other companies to test their bot presences on its network through what it then called “promoted chats.” In April, Kik launched its new Bot Shop — an in-app storefront where users can connect with new bot services (like a personal shopper from H&M) and games (like Arterra, a post-apocalyptic space adventure). A week later, Facebook announced its own new bot platform and bot store at its annual F8 developers’ conference. Kik execs insist they didn’t plan a David-bops-Goliath moment. But they smile as they say it.
Kik’s bots live inside a chat app, but they don’t really want to chat much with you. It turns out that, for delivering routine information, at least, conversation isn’t very efficient. If you need a daily weather report, for instance, you may not want to wade through a hit-or-miss joke with each forecast. That is exactly what Poncho, a chatbot unveiled at Facebook’s bot launch, provides, and it opened to scathing reviews. (“Frustrating and useless” — Gizmodo; “the slowest way to use the internet” — The Verge.).
For Kik, the important word in “conversational interface” isn’t “conversation,” it’s “interface.”
“If you go to our bot shop, almost none of them use the 26-char alphanumeric keyboard,” Livingston says. “They ask you a question, but then they give you four-five-six buttons as suggestions, and you choose one. What adventure do you want to go on? A B or C?” Almost like … a phone tree? “Right — no big deal. But the friction to trying that experience is almost zero.”
Interviewing Livingston is similarly modular, like navigating a choose-your-own-adventure story: He keeps helpfully presenting you with options to select. “There’s the past — I can tell you, going back, how we got here. There’s the present — why do I think bots are interesting? Then there’s the future — like, open or closed, what do I think will be exciting about bots, what could be worrisome, all that stuff. So where do you wanna go?”
The annals of artificial intelligence are shaped by two competing visions.
In “weak” or narrow AI, the system simply masters a task — it can play chess but not banter with you. With “Strong AI” (aka “artificial general intelligence”), a la Hal 9000 or Samantha from Her, the system is adept enough with language and situational awareness to learn and improvise and even impersonate a sentient being.
The way Livingston sees it, big companies that have poured money into AI for decades are hoping they can finally see a return on that investment. Thus we have Siri and Cortana and Alexa and Facebook’s M — a whole cast of characters vying to serve as our digital majordomos and intelligently cater to our needs and wishes. They will put a friendly voice in front of the yottabytes of cloud data and logic that sit behind today’s search window, and we won’t ever have to type again.
“It’s a nice story,” says Kik’s Josh Jacobs. “It says, the future is this new kind of technology that requires an army of PhDs to work at your company to compete. Is that true?” To the Google- or Apple-style “one bot to rule them all” scenario, Kik counters with a “let a thousand bots bloom” approach. Mentioning a Kik bot will summon it into a chat, but it won’t pretend to be your friend. “The bot just pops in and pops out,” says Mike Roberts, Kik’s head of mobile development, who oversaw the launch of the Kik Bot Shop. “It doesn’t sit there and listen. You don’t have to worry about what the bot hears, or what it thinks about what else you’re saying.”
Jacobs argues that chatbots will be able to solve a problem for us that’s entirely different from the one search engines have mastered — call it the housesitter-music dilemma. You’re visiting my home; how do you know how to play a song? Am I an Apple Music user plugged into Sonos, a Spotify subscriber piping tracks through Alexa, or what? “As we walk around in our daily lives,” Jacobs says, “we’re coming in and out of networks of available services. And there’s no great interface for detecting, dynamically configuring, and interacting with them, then walking out of their orbit.” For such scenarios, AI might be overkill. It might not even do a great job.
Right now the 50-odd offerings in Kik’s Bot Shop are pretty unassuming. They’re not going to end hunger or usher in world peace; they’ll summon a video clip to make you laugh, start a zombie survival game with your friends, or let you chat with a Teenage Mutant Ninja Turtle.
It’s all admittedly underwhelming for the dawn of a new era. “People ask me all the time,” Livingston says, “show me a bot that points to this future, and I can’t, not yet.” The web looked pretty underwhelming in the mid-’90s, too. But a look at the Kik Bot Shop does make you wonder: What happens when your chat list gets as crowded as your email inbox? Do we really want to invite brands and marketers into the space we use to talk to our friends and family? Isn’t there a possibility, as Buzzfeed’s Nitasha Tiku asked after Facebook’s bot launch, that “users may not want to chat with airlines and online retailers ‘like they’re your friend’”?
Paul Gray, Kik’s director of platform services, says things only get dicey when marketers treat a bot as if it were a broadcast channel. “This is not the best tool for marketers to just blast things out. ‘Oh, great, you can reach users one on one? Let’s send them a message every single day!’” A good bot, Gray says, is one that understands when you’re losing interest.
a) If the AI is really strong/general, why are so many different bots required for these experiences?
b) The people I talk to hate chat bots, e.g. Slack bots. This seems very much like a top down "trend".
c) I do think there is a future for conversational apps, in fact I am working on one now (voice based). But let's not pretend we have artificial intelligence.
Yes, it's a weak mind that couples Chatbots and AI.
Chatbots don't need AI to be useful. And plenty of AI does not use Chatbots.
Slack in general is too chatty so adding more bots there seems like too much.
I like that you're building a voice based app.
Writing my first voice based app has been more fun than I've had with .NET in a while, although the error handling infrastructure and tweaking needed for this application tempered that, eventually. I would like to do more with it. Interesting things I discovered:
- The default setting for CPU utilization for the Windows speech recognizer is 50%. This is why e.g. Siri-type handles most of the recognition on a server. Phones may lack suitable compute bandwidth in many cases.
- I have always thought it was lack of attention to UI that phone-based services make you say 'zero' instead of 'oh'. Now I realize that having to recognize a tiny, relatively undifferentiated syllable like that just tanks recognition, even within a limited grammar. Humans are waaaay better about this than computers, still, and for computers to be decent some contextual information / limiting grammar still really helps. I have been doing all my testing in a noisy environment... the gap is less noticeable with a good signal.
Fascinating that there are all these considerations you don't have in a typing interface.
50% CPU utilization sounds like a lot but the latency of a server round trip sounds intolerable.
I guess we need CPUs to get another 10x more powerful for these types of apps.