When it comes to conversational AI, science fiction sets a high bar. In the book and film 2001: A Space Odyssey, the sentient computer system HAL 9000 is so advanced that it figures out how to read people’s lips. A boxy tactical robot in Interstellar, TARS has adjustable levels of sarcasm. Samantha, the conversant operating system in Her, is so nuanced, sophisticated, and “human” than its user, Theodore, falls madly in love.
It’s tempting to take these examples as benchmarks for providing customers with chatbot experiences that feel as human as possible, but that’s almost always the wrong approach—one that can quickly veer into the ditch of unintended creepiness. Massive technological barriers have been hurdled and smashed in the pursuit of conversational AI, but HAL, TARS, and Samantha exist on the other side of the uncanny valley.
Beware the Valley of the Uncanny
As a refresher, uncanny valley is the idea that we like it when things exhibit human characteristics, but there’s a tipping point where seeming too close to human elicits an instinctively chilly response. A quick Google search of the term will bring up a host of images and videos of dead-eyed, rubbery robots that inflict primal unease.
GIF source YouTube
"Evolutionary history has tuned us to detect minor distortions that indicate disease, mental or physical problems," Thalia Wheatley, a psychologist at Dartmouth College, told Scientific American back in 2012. By her estimation, aiming for a human-seeming experience is to go up against millions of years of evolutionary history.
What’s tricky about designing human-like traits for machines is that the “almost” in “almost human” is elusive. There’s no clear line that you can tiptoe up to when designing something that’s just human enough. These design challenges are most often associated with humanoid robots, but they apply readily to conversational AI as well.
Chatbots that generate a typing sound after customers respond to questions over the phone, text interactions with a bot that include too many designed “hmms” and “ums,” web-based service avatars that blink their eyes, these are all inhabitants of the uncanny valley of conversational AI.
Use Anthropomorphism Wisely
Attaching human qualities to conversational AI design is an example of anthropomorphism, a design tool that’s loaded with consequence.
“In conversational design, anthropomorphism can give applications human qualities—a powerful tool for making interactions feel like a genuine interaction with another sentient being,” writes Lance Christmann, Head of Experience Design at OneReach.ai, in his article, A New Twist On Two Old Morphisms. “If used successfully, this lever can foster a sense of human connection, trust, and loyalty.”
As Christmann goes on to explain, anthropomorphism should be applied with clear intent. There are diabolical use cases where it can be used unethically to pull emotional levers, but there are also easy missteps that can create feelings of unease or mistrust.
For example, a troubled customer who is trying to get information about unexpected charges to their bank account is likely to be irritated by a chatbot cracking puns or dropping cute slang (e.g. “I’m searching for deets on the requested transaction”). In a similar vein, using background sound effects and vocal affectations to try and make it seem as though that customer is communicating with a real person is likely to add to their unease, when what they really want is a fast, useful response.
A Little Human Can Go A Long Way
Though it’s a very basic bot, Little MOO is a great example of a little bit of human delivering high levels of satisfaction. When customers complete a print order through moo.com (Big Moo), they get follow-up emails from Little Moo with status updates.
“It's Little MOO here - thanks again, we've loved seeing your designs come to life,” a typical email reads. “I'm excited to say Big MOO has printed your order and it'll shortly be dispatched from our warehouse. Yay”
A few carefully selected human traits (a name, a sense of shared satisfaction about a print order, a dash of enthusiasm) forge a strong relationship between Moo and it’s customers. A slight, but deft touch elicits a heavy emotional response.
A more sophisticated bit of conversational AI that delivers big dividends can be found in Maya, a fast and friendly bot that interacts with Lemonade’s customers. This rental insurance company has disrupted their industry by making it easy to purchase policies and donating any leftover money from claims to charities, but it’s Maya who steals the show.
“The biggest thing that pushed me to convert to Lemonade was the utterly charming AI chatbot,” writes Juliette van Winden in a Medium post dedicated to Maya. “Unlike the drag of signing up with other providers, it took me a total of two minutes to walk through all the steps with Maya ... What intrigued me the most, is that it didn’t feel like I was chatting with a bot. Maya is funny and charismatic—which made the exchange feel authentic.”
Exhibiting a balance of human traits and machine-like efficiency, Maya forges emotional bonds with customers and completes tasks in a way that leaves them feeling cared-for, key deliverables for a good customer experience. Maya is friendly, but not too friendly (i.e. she doesn’t ask how a customer is feeling today, but she does offer an appropriate “congratulations” when learning that the user has moved into a new apartment). She puts people at ease while simultaneously cancelling their old policies and creating new ones faster than any human agent ever could
Conversational AI Should be Better-Than-Human
Using anthropomorphism wisely can turn conversational AI from cold and transactional to warm and journey-oriented—maintaining context and brand identity even as interactions with customers carry across multiple channels. Rather than trying to mimic every aspect of human interaction, the best implementations in conversational AI employ small doses of personality on the customer-facing side while sequenced technology running behind the scenes aggregates data and completes complex processes in order to provide a better-than-human experience.
It’s likely that one day, conversational AI will become indistinguishable from interactions with humans (and on that day we will have to grapple with everything that implies). For now, however, the most impactful way to connect customers to the power of conversational AI is by using human traits to build trust and engagement rather than trying to create human-feeling experiences.
Put another way, comfort customers with some human traits, but dazzle them with your ability to solve complex problems.