Debuting last week at Facebook’s F8 conference, the initial batch of Messenger bots have not been well-received. Spencer Chen struggled to order flowers because the bot didn’t understand cancellation commands, while Mat Honan was provided a weather forecast for an entirely different city.
Buzzfeed’s Katie Notopoulos tried to purchase shoes, and came upon a realization:
Some people hate to shop. They want to look nice, but hate having to pick out their own clothes and would be grateful to be shown just three options from a trusted store. […]
However, this is NOT the customer who is buying $400 women’s shoes. The $400 women’s shoes customer most likely enjoys shopping. The same for $400 men’s shoes! The kind of shopper who is going to buy the fairly expensive kind of apparel that Spring offers is not the same customer who wants to use a bot that will show them three options.
And Darren Orf over at Gizmodo summed up the platform’s status:
But most people will likely try out these rough-cut bots and decide they’re not worth the hassle. Chatbots leave you with that same itch in the back of your mind that it’s easier to get the weather or send flowers the old-fashioned way. They’ll get better, but it’s going to take time. Right now, chatbots are a robotic wild west, and for the foreseeable future, you’re better off sticking with civilized society.
These issues are familiar to anyone who used Siri after it first launched in 2011 or, indeed, anyone who has used virtually any natural language processing software. Beyond speech recognition problems, it’s clear that virtual assistants and bots like these continue to struggle with grasping human intent. They’re only programmed to pick up on key words and phrases, and to try to follow the progression of the conversation. But the moment that something is interjected — like when Chen asked the flower delivery bot whether it was available in Canada — it doesn’t follow the flowchart, and the conversation breaks down.
If Apple is on track to debut the Siri API at WWDC, it would be helpful to third-party developers if it would assist with the interpretation and contextualization of the conversation instead of requiring developers to build their own artificial intelligence infrastructure.