By Cade Metz for Wired
I JUST ASKED a bot to tell me how much I spent on coffee this past month. And a few moments later, it replied. “Let’s see,” it said. “You spent $199.83.”
Certainly, I drink an awful lot of coffee. But not that much. After asking the bot for a bit more information, I can happily say I spent at least some of that nearly two hundred dollars on chopped fruit and the occasional pastry. The bot didn’t actually explain this, but it did give me list all payments to coffee shops, which made things clear. “Thank you,” I told the bot. And it responded with a big yellow smiling emojicon.
This bot is called MyKai, and it’s the brainchild ofKasisto, a startup that spun out of the Silicon Valley research lab that helped create Siri, the talking digital assistant that comes with the Apple iPhone. Dovetailing with popular messaging apps like Facebook Messenger and Slack, MyKai is meant to provide a simpler way of watching your money—or even moving it. This bot can bring up your latest bank balance or tell you how much you’re paying in fees or instantly wire money across the ‘net. “We’re doing things you can’t do as quickly in other ways,” says Dror Oren, Kasisto’s vice president of product.
People in Asia are already using MyKai, and beginning today, you can too. Because it’s focused on banking—and banking alone—it works pretty well. But it’s also flawed. And it’s a bit weird. Unsettling, even. All of which makes it a great way of deconstructing the tech world’s ever growing obsession with “chatbots.”
In recent months, some of the Internet biggest names have put their weight behind this big idea, including Facebook, Microsoft, and Google. And countless others are pushing the same notion, including startups like Kasisto and the New York-based GoButler. The idea is that we’ll soon use messaging apps—like Messenger, Slack, Telegram, and Skype—to communicate not only with our friends and family, but with all sorts of Internet services. Without leaving our favorite messaging app, the thinking goes, we’ll hail rides from Uber, book plane flights from United, and, yes, check in on our bank accounts—all in a wonderfully conversational way.
But here’s the rub: artificial intelligence hasn’t quite progressed to the point where it really carry on a conversation—where it can always give us what we’re asking for. If you’ve used Siri—or any other smartphone digital assistant—you know this. And though people use messaging apps to interact with Internet services in Asia—particularly China—the rest of the world isn’t necessarily geared towards this kind of thing. At least not yet.
I must admit: I felt uneasy installing MyKai. I’m hardly unfamiliar with the latest in AI and other smartphone tech, but as I linked an independent, ostensibly intelligent service to my bank account, I couldn’t help but question the decision. That’s not necessarily rational. For the most part, MyKai just reads bank balances and statements. But this is my money we’re talking about. I’m giving up the password to my banking app. And MyKai goes on to ask if I’m interested in connecting to Venmo, so I easily send money to people too. With a startup—or indeed any company—all this gives you pause. But it’s also where the Internet is moving. So I installed MyKai, linking it to my Facebook Messenger app, and I started chatting.
MyKai is remarkably adept at understanding what I’m asking—and that’s largely because it’s focused solely on banking. When I ask “How much money do I have?” or “How much did I spend on food in May?,” it understands. But when I ask who won the Spain-Italy match at Euro 2016, it suggests I take another tack. The thing to realize about today’s chatbots is that they can be reasonably effective if they’ve honed to a particular task—and that they break down if the scope gets to wide. Piyush Gupta, the CEO of DBS, a big Singapore bank that offers MyKai through its own app, says that bot works far better than other natural-language services the bank has tried—services originally designed for a much broader set of applications.
Kasisto doesn’t necessarily use the latest in natural-language AI. Whereas Google and Facebook and others have advanced the state of the art with deep neural networks—networks of hardware and software that can learn tasks by analyzing vast amounts of data—the startup takes a more pragmatic approach. Or at least an approach that’s more pragmatic for the here and now. Basically, Kasisto relies on algorithms that carefully define how the service will work, as opposed to algorithms that learn on their own from data. That’s not the best way to tackle a more comprehensive style of natural language understanding designed to handle any conversation—the style Facebook and Google are reaching for—but it can work quite well for a banking bot.
That said, natural language understanding is only part of what’s needed to make MyKai work. Once it grasps what I’m saying, it must act on that. This breaks down a little because the MyKai is dependent on the info supplied by my bank. It doesn’t realize that some of that $199.83 went to fruit and pastries because it’s just looking for the word coffee in each payment record. The bigger picture is beyond the bot because it’s beyond my bank data. But I can ask MyKai to list all my coffee-shop transactions, and then, just from the prices listed, I can infer. Kai doesn’t always answe my questions with complete precision. But it’s gives me some decent info. I’m now thinking I should cut back on all the coffee shop espresso.
That’s fun, and kinda useful. But the bigger question is whether this bot is something people will use on a regular basis—not just for a day. In China, so many people use WeChat, a popular messaging app, to interact with other online services. This is how they hail a car or make a plane reservation. But that’s China, where the Internet evolved in very different ways. We Americans are already wedded to apps, and it’s unclear whether a messaging service could really save us that much time and energy—or, even if could, whether we would make the transition. Inertia is a powerful thing.
DBS, now the largest bank in Southeast Asia, is also working to embed its chat-happy banking app in outside messaging services like WhatsApp and WeChat, and Gupta sees this as the future. He’s trying to compete with Alibaba, the Chinese company that has already brought so many banking tasks online, but he also believes banking bots can take off in the U.S. too. “Banking is going to be embedded in other things you want to do with your life,” he says.
Even in China, it should be said, people don’t really use chatbots. When they juggle various Internet services through WeChat, they send pre-defined codes or press certain software buttons. They don’t carry on conversations with bots. The chatbot idea is still a bit of mystery. But that means it could go either way. MyKai doesn’t seem useful enough to replace my banking app. But it’s evolving, with Kasisto also working to with banks like DBS to deeply integrate it with their systems. And I’m certainly intrigued. That could be the caffeine talking. But maybe not.
First appeared at Wired