Facebook's new Messenger bots are the slowest way to use the internet
Save article ToRead Archive Delete · Log in Log out
4 min read · View original · theverge.com
Today, Facebook announced a bold new bot initiative that lives inside its Messenger application: artificial intelligence-powered personalities at our beck and call for shopping, checking the weather, getting news updates, and more. There's just one problem: the bots are slow... painfully slow. In fact, Poncho the "weathercat," which Facebook messaging chief David Marcus demoed onstage at the company's F8 developer conference today, comes with a disclaimer that it "typically replies within an hour." That's neat, if you're the kind of person who can spend an hour waiting to know whether it's going to rain.
The company's new bot platform comes with a bevy of tools for developers. That means third-parties can create their own AI assistants for us to talk to. The Wall Street Journal's bot, for instance, gives you "the day's top stories, breaking news updates, and instant markets coverage." The words "breaking" and "instant" are a bit ambitious — I sent a message to the WSJ bot and heard back about four minutes later. Once you start talking with the bot, it gets speedier. But a majority of your interactions will involve asking for specific pieces of information and then waiting to see if it's what you wanted.
Bot latency is weird - you're not a person, why does it take 5 mins to respond? Also, why have 'typing' bubbles? pic.twitter.com/rPRiIwuQ1o
— megan quinn (@msquinn) April 12, 2016
The problem is inherent to Facebook's own M bot as well. The service, which debuted in beta last fall, often takes minutes to get back to you and sometimes does so only with a follow-up question to clarify, adding extra time to the exchange. It even utilizes the bouncing trio of dots to indicate it's typing, which is an eerie thing for a piece of software to do. To be fair, M relies on a mix of AI and human components. So some responses naturally take more time while a real person on the other end makes a phone call or dives into online web ordering systems. But for simple questions, these bots' answers are not instantaneous in a way we might associate with our new robo-butlers.
Of course, for Facebook M and Messenger's other bots, this slowness could be an early stopgap. Any amount of delay could be designed to let the platform get up to speed with how users plan on making use of these assistants and whether they can withstand a swell of demand.
Why use bots when our fingers are faster?
Yet it also raises a conundrum for the impending bot revolution (and developer gold rush): why use AI-powered software if it's going to be slower than human fingers on a keyboard? I can check the weather by typing command+t and then "weather" in Chrome in about one second. I don't even need to hit enter, as Google now populates the temperature result as the first drop-down option. Using a bot also involves some back-and-forth for pretty much any request. You sometimes have to prick and prod with follow-ups and clarifications and add extra context to get the answer you're looking for. It's cool to see a request worked out in real time, but the process is not what you might consider efficient.
Granted, bots are not necessarily designed, at least not now, to be faster than us. The real value in the short term is automation. One of Poncho's first questions is to ask whether you want a daily and evening weather report sent to you. The idea is to anticipate our needs and wants and deliver something useful at the right time, without us having to ask. For more complex operations that aren't super time dependent, like shopping or placing a food order, bots can let us jot in a few commands over the course of 10 or 20 minutes while we focus on work.
Over time, these bots should be able to take on more responsibility, as they become smarter, more capable, and better at understanding the nuances of human language. You can even imagine being able to create whole workflows — similar to Photoshop actions or the productivity software Alfred — where we've trained our bots to execute a multi-step series of actions when we type in a special command. For now, however, you're better off typing into Google unless you have time to kill.