Let’s face it. Most of us aren’t so lucky as to have a voice like Scarlett Johansson’s. No no, not that kind of voice, although sure, her husky, breathy voice is enticing. What I’m referring to, however, is her literary voice – the Spike Jonze screenplay for Her that made Scarlett’s character say all the right things at the right times and just right.
Set in an unspecified near future, Her is the story of a man who falls in love with his operating system, Samantha. While the plot of Her pivots on an unlikely human-AI romance, it is also a reflection on the complexity of building an AI voice whose ability to converse and emote is nearly indistinguishable from a human’s.
That’s also why I’m here. Recently our team at Two by Fore Interactive built the first ever real estate agent Chatbot (at least that we know about). Since then we’ve created a whole bunch more for agents and teams alike. As the team’s copywriter (my official title is The Lady Who Does the Word Stuff), I was tasked with finding the Chatbot’s voice – creating a kind of (slightly less seductive) Interactive Samantha. Although present-day tech is not quite at Her-level (the AIs aren’t going to ascend to another dimension any time soon), our Chatbot nevertheless marks a new era for personalized integrated automation.
Our Chatbot connects to the Messenger account on a Facebook Business Page, and works to get you more meetings with qualified leads while you sleep (or meet with those aforementioned leads).
Creating a Chatbot that people actually want to interact with is kind of like creating a likeable character for a novel or film. There is a narrative to follow (the conversation workflow), a specific world where the character exists (the larger context of the brand), and a teleological plot (what you want the leads interacting with the Chatbot to do). Nobody wants to interact with a boring and ineffective bot, just like they don’t want to read a boring book or watch a boring film.
A Chatbot’s tone and linguistic style is crucial, but there is a lot of behind-the-scenes knowledge involved as well. I had to think: what exactly do we want our Chatbot to do? What is the best way for it to do that? A Chatbot should both anticipate clients’ needs, and balance expectations of how a bot would sound. One way to maintain user engagement while also not ever leaving leads hanging is by repeating information back in order to confirm what has been said. Once the bot has asked a series of questions, it’ll say “did I get that right?” This works to warm and secure leads.
In an article about Her and the real-life possibilities of AI for Wired, Vlad Sejnoha writes that the power of language “relies on the assumption that speakers and listeners are intelligent and knowledgeable about both the social and the physical world.” The nuance involved in creating a Chatbot linguistics is due in large part to this. Despite technological advances, there’s a limit to how much a bot knows and can reason. What a Chatbot can do, however, is serve as a funnel. By asking structured questions, a Chatbot can appear much smarter than it actually is.
Joe Toscano of Google writes: “Your bot should reflect your brand. If your brand were a human being, how would it speak? What kind of language would it use?” When thinking about our Chatbot’s communication style, I decided to make our bot self-reflexive, in order to both acknowledge its current limitations as a technology and to bring some flair and humour to its language. So when a user goes off-script, the bot doesn’t glitch out unattractively, instead it says something like, “Sorry, despite being a superior species, I’m still acclimating to human jargon. Can we try again?”
Writing the voice of a Chatbot is complicated so that the experience of interacting with one isn’t. Like a good advertising campaign, a good bot has personality. People love to anthropomorphize – that’s just Being a Human 101, we’re always looking for a reflection of ourselves. Despite the knowledge that a Chatbot isn’t a sentient person typing back to you in real time, we imagine that it is – which is part of what makes writing one so fun. And it also happens to be what makes those interacting with the bot slightly fall for it, even if not as strongly as Theodore fell for Samantha.