Pop culture these days is awash in tales of AI run amok, from machines that act like humans to humans in love with machines. The reason seems clear enough: We’re anxious about a world in which machines have superseded us. But it isn’t just armchair philosophy. Designers are already grappling with whether a robot should ever really act like a human.
I don’t mean robots in the futuristic sense, like you might find in Westworld or 2001, but rather robots that we’re increasingly dealing with everyday. I mean the robots that live in Google Allo, Facebook Messenger, and a host of other new services that use chat-based assistants to serve up everything from airplane reservations to outfit advice. You can dub this the new skeuomorphism: The attempt to make bots mimic full-blown chat partners. Just as skeuomorphism in interfaces was about replicating things like trash cans and leather calendars to better help people grok their software, the new skeuomorphism is about mimicking how people talk, in the hopes that we’ll be willing to talk to machines.
Paul Adams is one of the designers fighting against that new skeuomorphism. “Bots shouldn’t act like humans, or pretend to be human,” he says. “Which is different from Facebook and many others.” Adams is the vice president of product at Intercom, a startup that makes chat bots like the ones you find more and more companies using to handle customer service on their websites. Intercom is among the most successful: The company claims over 13,000 customers; has raised $116 million in venture funding; and says 200 million conversations occur ever month using its platform. Today, Intercom is launching Educate, a new chat-assisted customer service app that happens to reveal subtle lessons for how myriad others should approach AI in interface design.
For Adams, who was formerly a global design lead at Facebook, the question of bot skeuomorphism isn’t one of morals—Adams isn’t decrying it on grounds that human-like chatbots are betraying the materials they’re made of, as a modernist designer might. Rather, it’s an issue of of usability. Intercom’s own usability studies have found that when people interact with chatbots, the ones that act human turn out to be maddening. The reason is simple: When a bot makes someone think it’s a human, the user will lob questions at it like any other real person. But when bots fall short in those interactions, users quickly grow frustrated. They feel tricked, and want to flee the system altogether.
The obvious solution is to let bots behave like bots. Intercom has done that before—a bot will announce itself and make clear the limits of its capabilities. But where Educate extends that system is in a layer of filtering: It uses AI to figure out if a customer’s question is simple and common, or rare and difficult. After making that decision, it either shunts users to a bot or a human.
The bot works as you’d expect, fielding queries. And when a human service agent does come online, he or she is aided by AI: When the conversation progresses, the backend service scans for queries it recognizes. If it finds one, then the AI assistant will suggest articles from its library of self-help guides that the human should share with the customer. (Educate uses a Medium-like CMS for creating customer-support answers.) Thus, Educate doesn’t use AI to replace humans with conversational bots. Rather, the AI is used to identify when a person is needed—and when people are needed, the AI augments the person’s capabilities, saving time and effort that might otherwise be spent on repetitive tasks. “We’re using machine-learning to create better answers, but there’s no bot getting in the way,” says Adams. So, for example, a human service agent who has discovered that a customer’s problem isn’t actually a new one won’t have to spent minutes explaining the solution anyway. Instead, they’ll be able to address these with a couple clicks and move on.
The reason that so many companies like Intercom are now investing in creating chatbot-powered customer-service is that their customers associate that kind of one-on-one interaction with friendly, human-to-human service that’s been lost in the internet era. And what they’re taking advantage of is a new paradigm for computing that revolves around chatting—which might be a more familiar and scalable interaction paradigm, given how much chat applications of one sort or another are dominating our daily life. As Adams puts it, the internet is re-architecting itself around people. “Messaging is a paradigm that’s here to stay,” he says. “People will interact with each other through messaging as the primary channel.”
Today, we live in a world where chatbots help companies engage with customers in a high-touch way that’s been lost on the internet. In the meantime, the internet is steadily being remade to be just as accessible as your chats with friends. If that’s not sci-fi, then what is?