Please, Facebook, don't make me speak to your awful chatbots

Facebook Glass

Have you heard? Apps are dead: chatbots are the new apps. And they will soon be doing everything, from taking your pizza orders to scheduling your meetings. This is the future and it’s going to be terrible.

The rise of the chatbot has been foretold for some time but only in the past few weeks with Facebook’s Messenger bots, chat app Kik’s bot store and the rise of subversive artbots have they really hit the public consciousness.

So how did we get here? In many ways, it’s the perfect convergence of almost every tech trend of the past five years.

Trend one: the hybrid command line

Remember Peach? Probably not. At this stage, four months after launch, it seems unlikely that even Peach’s developers remember Peach. The app, a social network from the co-founder of Vine, tried to live somewhere between a Twitter-style social network and a WhatsApp-style messaging platform. It was fun for about three days, and then it died a quick death as everyone got bored.

But Peach did have one thing going for it: a pseudo-command-line, that let users type simple commands in without needing to plough through menus. Typing “g” in the compose box brings up a gif search; typing “c” lets you share a calendar event.

The same practice used, with more success, by business-focused chat app Slack. In both cases, the commands let the app offer a rich set of features, without becoming over-cluttered or difficult to use, and they let power-users become rapid and efficient.

Trend two: natural language parsing and voice recognition

It’s been almost five years since the iPhone 4S introduced Siri into our lives, and voice control is now a mainstream feature on smartphones. Every major phone platform offers its own voice-controlled assistant, while Amazon will even ship you the Echo, an obelisk that sits on your living room table listening to everything.

These personal assistants still aren’t great, even after half a decade of improvement, but they are sometimes good enough. That is actually pretty impressive because it means they have dealt with two very difficult, largely unrelated tasks. They can understand what you say, and they can understand what you mean. The former is voice recognition, and the latter is natural language parsing.

Trend three: the Facebook Platform

Facebook is already utterly dominant as a social network. But its attempts to grow beyond that have been a more stop-start affair. From 2009, companies could build “apps” on Facebook’s website, which led to the runaway success of Farmville in particular. But the Facebook-hosted apps failed to take off beyond games, and were lost in the switch to mobile.

More recently, Facebook managed to branch out in a different way, offering up its own platform as a way for other developers to authenticate users. Facebook gets knowledge of what their users are doing in other apps, and the apps get to trawl through Facebook for data on their new users. But that too isn’t enough for Facebook: time spent in other apps is time spent out of Facebook, and that’s time when you can’t be shown adverts. So the latest attempt is to keep you inside Facebook for good.

Instant Articles do that for the news articles you used to click while Facebook Video replaces all those YouTube links. For the sites that need people to leave Facebook, it’s terrifying. But for Facebook books its much more convenient.

So how do these come together?

When Mark Zuckerberg introduced Facebook’s chatbot-based vision of the future this month, it was the obvious next step to all these things: third-party bots which plug into Facebook’s Messenger service.

The new bots build on the command line model of Peach and Slack, not by forcing you to learn intricate commands to type in the midst of your messages with friends, but by allowing you to quickly and easily send a few words to the right chatbot to fulfil complex commands.

They take the lessons of Siri and Google Now, by using Facebook’s own AI engine to parse the natural language you speak to the them in and performing complex tasks in response.

And they are built into Facebook’s walled-off version of the internet. Now you don’t even need to click on an instant article to get the headlines: you just ask the Wall Street Journal’s chatbot, and it will tell you what’s happening.

There’s just one problem: the chatbots suck.

The bots really suck

One day, chatbots will be widely used, and a useful solution to a number of everyday problems. But not today.

The problems with existing chatbots begin with how they actually work. Almost uniformly, the initial examples of Messenger bots are disastrous: unable to parse any instruction that doesn’t fit their (entirely undocumented) expectations, slow to respond when they are given the correct command, and ultimately useful only for tasks which are trivial to perform through the old apps or websites.

CNN’s bot is the worst of a bad bunch, but serves as an example of how low the bar is set. It can only do two things: interpret short phrases and do a keyword search (so if you say “Clinton”, you get a few headlines about Hilary Clinton).

Actually, it does one other thing: the instant the bot gets any message at all from a user, the user is subscribed to a daily roundup of emails.

How can you unsubscribe from this? If, and only if, you reply with the word “unsubscribe” and no other text. Though the bot responds to you in a chatty fashion, it’s unable to understand anything more natural sounding than “unsubscribe”.

Give it any other command at all and it replies with a literal ¯\_(ツ)_/¯. Cute, but not the fifth time you try to guess how to unsubscribe unsuccessfully.

Yes, this kind of problem will go away in time, Facebook’s natural language parsing will improve allowing developers to more easily understand a phrase like “please, please stop sending me messages” but right now it’s a bad experience.

While some problems will be solved with better tech, others will simply be fixed with more money. Facebook’s own pioneering chatbot, M, is instructive. The bot, which is still only available in the US, acts as a sort of digital concierge. Like older services such as Magic, you can ask the Facebot to help with almost anything you want, and it will do it for you. It will order you food, make reservations at restaurants, or tell you the quickest way to get from A to B. It will even let you send a parrot to your friend’s office. But this chatbot has a secret. It is made of people.

The human touch

Facebook says that the humans behind the service are just there for training purposes: when the AI can’t handle something, it passes it on to a human to do the heavy lifting, and watches what happens, to better learn how to deal with the query on its own.

That might be the long term goal, but in the short term it’s pretty clear that M does a lot of things which Facebook has no hope of automating any time soon, like making a booking at a restaurant that only takes phone reservations.

So the mid-term success for many chatbots will be as a shield, protecting us from the reality that much of the benefits of automation are actually just low-paid grunt work from call centres in developing nations. Be nice to your bots; you never know when they’re actually people.

But technology won’t overcome the more fundamental problem with chatbots: they are misguided attempts to solve a problem that doesn’t exist, built largely because they are popular in east Asia for a completely different set of reasons.

Is ‘WeFacebook’ the future?

If you want to know why Facebook is pushing its Messenger bots so hard, look at WeChat. The Chinese app started as a simple messaging app, similar to WhatsApp, but has grown enormously.

Brands and companies have “official accounts”, allowing them to interact directly with customers. Those accounts started as human-managed customer-service accounts but grew. First came a simple phone-tree-style navigation menu, giving users a bit of control over who they could speak to speak to by texting the number relating to their issue. Then came a simple fixed menu, at the bottom of the screen, letting them have some permanent options always available. And then, gradually, intelligence: they became chatbots.

Those official accounts are a large part of WeChat’s dominance in China. Mark Zuckerberg wishes he could have as big an influence on daily life in the US. But, as WeChat product manager Dan Grover points out, looking at what’s happening inside the conversation is missing the point.

Even in China, even with WeChat’s lead in chatbots, doing anything important or complicated through a conversational interface sucks. Grover uses the example of ordering a pizza: a chat app takes 73 taps on the screen to order a pizza, the vast majority of them on the tiny onscreen keyboard where a single misplaced tap can render a message nonsense. The Pizza Hut app, by contrast, takes 16. And unlike the chatbot, it also conveys a huge amount of useful information in those few taps, letting the user see the full list of toppings, letting them find out the prices of various options, letting them browse a menu, and letting them discover the unknown unknowns – the things they otherwise wouldn’t even know the company could offer.

Where chatbots go to die

This problem alone is the death knell for a huge amount of conversational software. Until it’s good enough to do everything, users pretty quickly learn what a chatbot, or voice-controlled assistance, can and can’t do – and they stop exploring. That means that as the assistant gets better, the user stays ignorant. You can find the last three apps you added to your phone by looking at the icons. Can you name the last three things your chosen brand of voice assistant gained the ability to do? Would you even know how to find out?

The benefit, Grover argues, isn’t in the chat; it’s in the inbox. Almost every app on a modern smartphone is concerned, in some way, with the dissemination of information. But the notification systems on iOS and Android simply aren’t up to the task: the notifications are too divorced from the action itself, too difficult to manage and control, and too binary – either unread, or cleared entirely.

“It’s no stretch to see WeChat and its ilk not as SMS replacements but as nascent visions of a mobile OS whose UI paradigm is, rather than rigidly app-centric, thread-centric (and not, strictly speaking, conversation-centric),” Grover writes.

Some things don’t need to be part of a thread at all. A calculator app wouldn’t gain anything from that system. Nor would a photo-editing app, nor (certain kinds of) games. But for those which do, it’s not hard to see how the WeChat experience improves things.

That’s the end goal that Facebook is chasing, make no mistake. The company wants its app to effectively become the homescreen for your phone: not only the place you go to manage your relationships, but also where you go to hail a cab, check on the weather, book a table or pay a bill. But unlike WeChat, which was building an infrastructure that didn’t previously exist, Facebook needs a carrot to entice us all off the internet and out of our homescreens and into its friendly, cozy, but padlocked walled garden.

That carrot is the chatbots. Right now, the carrot’s rotten, but Facebook will find a tastier one soon enough. Even so, though. I think I’ll pass.

Powered by Guardian.co.ukThis article was written by Alex Hern, for theguardian.com on Friday 29th April 2016 12.16 Europe/Londonguardian.co.uk © Guardian News and Media Limited 2010