Wake Me Up When the Chatbots Write Their Own Scripts
“I don’t wanna be a poet /’Cause I don’t wanna blow it”
ONE OF THE characteristics of calling into a contact center that consumers consistently loathe is the feeling that the agent they’re dealing with is just reading a predesigned script. It feels robotic and completely tone deaf to our needs. How many of us have fumed when a tech support engineer asks if we’ve rebooted the device? “Yeah, I have no idea how I got on this turnip truck, but I guess I’m about to fall off it, right, pal?”
But then companies seemingly said, “Oh, you don’t like the script? Wait until you get a load of our new extreme version. I’m sure you’ll like that much more.” Over the past few years, most of us have been subjected to chatbots nominally designed to help us solve customer service issues on our own. But chatbots today are nothing but 100 percent pre-scripted dialogues, with little to no flexibility, and with personalities less palatable than my endless string of dad jokes. People use chatbots, but rarely do they love them, and the scripted feeling contributes to our antipathy.
For better or worse, however, there are clear glimmers of how chatbots will eventually gain the ability to “go off script.” Natural language generation technology will allow companies to set artificial intelligence to the task of summarizing documents or to write news stories. In fact, that last use case has already become close to commonplace. The Washington Post used a natural-language-generation-like tool called Heliograf to create news stories as far back as 2016. Not a chatbot engaging in interactive dialogue, mind you, but baby steps toward that.
The autonomous direction in which chatbots will go, however, was really brought home to me by a recent Google research project called Verse By Verse. It purports to be “an experimental AI-powered muse that helps you compose poetry inspired by classic American poets.” The user selects a poet or several poets in whose style the work will be created, then chooses some parameters such as form (quatrain, free verse, etc.) and rhyming scheme. The user then feeds in seed text to give Verse By Verse some idea where it is supposed to take the poem and the poetry spills forth.
To me this seemed like a fun way to see what classic American poets could do as songwriters. Daddy Yankee heard through the voice of Emily Dickinson? Metallica by way of Robert Frost? Frost’s poems “The Vanishing Red” and “The Trial by Existence” already have Metallica-esque titles, so why not? The myriad possibilities were tantalizing. In my experiment, I decided to stick with my tried-and-true Prince and see what Walt Whitman would have done with “1999.” Here’s what I got:
“I was dreamin’ when I wrote this,
Forgive me if it goes astray.
A strange tears will ripe for my heart!
So soon I cease into your rage,
Return to peace with my own blood,
Exult in my west, my old race!”
Umm, sure? Other than the fact that the words are nonsensical gibberish and seemingly bear little relation to the Prince lines I fed the system, sure—it’s Prince by way of Walt Whitman. It does feel distinctly Whitmanesque, and that’s the key for the future of chatbots. When this technology progresses to the point of consistently creating coherent text, companies will be able to train the chatbots on a particular brand voice. The chatbot will be able to have free-flowing conversations in that voice. That would be a huge step toward creating engaging customer experiences. We’re still quite a way from that reality, but as Prince said in one of his lesser songs, “I’ve seen the future and it works.”
Ian Jacobs is a principal analyst at Forrester Research.