When Machines Won’t Learn
If you believe the evangelists, “new advances in AI” will allow businesses to restructure their support and sales operations, and maybe even their business models.
Via a chatbot interface to Messenger, an AI program would take care of work a human would do. Mark Zuckerberg implied as much when he introduced Facebook’s bot platform only last April to great fanfare, in a move hailed as part of an AI arms race.
The “Fourth Industrial Revolution” of white-collar automation was almost here.
But reports that Facebook has scaled back the fashionable emphasis on machine learning and AI after encountering a high failure rate have turned out to be true.
Judging from the changelogs and updates to the Messenger APIs, the social network has dialed down its desire to see businesses implement complex conversational AI-powered chatbots, and instead seems content with straightforward software agents that respond to written requests and commands. These simpler programs, connected to its instant-messaging service, are proving more reliable than machine-learning-driven bots, which have come unstuck and confused in use with normal people.
British entrepreneur Syd Lawrence thinks Facebook has been wise to move away from an emphasis on machine learning. “The whole AI hype is horrifically off the mark,” he told us. “Whereas these things can be really useful and powerful. But they’re certainly not AI. We’re way away from AI being really useful.”
Lawrence, CEO of The Bot Platform, says: “The whole conversation around AI and even the term chatbots is all horribly wrong. I really hope we can put AI chatbots to rest. No one wants to talk to a chatbot.” If that sounds odd from a company at the bleeding edge of the chatbot hype, then it needs an explanation.
The bots Lawrence’s team creates are really akin to a Unix daemon, or in Windows, a Service: they’re a background process or agent. In fact, we all use these often without knowing it, Lawrence tells us, and they’ve never stopped being useful. Now they’re slightly more useful.
“One of the bots using our platform is performing almost 10,000 per cent better than their mobile website for one common task, and we’ve customers that have their bots performing 30,000 per cent better than an email mailing list,” says Lawrence.
A real-life walkthrough of a bot on the revamped Facebook Messenger shows what he means.
The revamped bot UI has a persistent menu, and can now run “headless”. Imagine bypassing a bank or a utility’s voice menu and instead using a set of menu buttons instead. It’s so much faster to navigate, you won’t want to go back to “press 5 for an customer service representative”.
Hopefully this will put “fake AI” to rest, says Syd in a Medium post describing the changes.
Just be clear about you’re trying to do, be aware of what the limitations are – specifically, typing on mobile is hard – and stop claiming “that it’s AI”. Facebook is now careful to call the agents “bots”, not “chatbots”, in addition to offering a less annoying customer interface than IVR, aka voice menus.
“Disabling free text completely might sound counter-intuitive, especially for a messaging platform, but it still lets you do outbound messaging, and control the conversation. You can still present the user with buttons. Not surprisingly, people prefer to press the buttons rather than type.”
So perhaps the future isn’t virtual robot buddies replacing humans, but simple interactive mobile UIs replacing another technology: the 1990s-spawned nightmare of voice menus. ®