Why Being Sort-Of Human Won’t Cut It

Design with AI in mind.

Originally published in ad:tech.

In 1970, Masahiro Mori published his celebrated essay on robotics, “The Uncanny Valley.” In it, he argued that if you’re going to make a robot, you either want to make it absolutely human-looking or not very human-looking at all. If you merely get close to human, you venture into an uncanny valley and creep people out. That’s why humanoid robots like Asimo don’t look very human, and we find it hilarious rather than frightening when they fall down stairs.

This is also an important lesson—and warning—for brands. As they confront a new world of AI, chatbots, and voice-enabled systems, they face a serious dilemma: Consumers want companies to be more human, but we must tread carefully. So, how do we do this?

Design for characters, not experiences

It seems the simple answer might be to follow the lead of interfaces such as Alexa, whichreflect humanity rather than act human. But in 1996’s The Media Equation, Clifford Naas and Byron Reeves of Stanford demonstrated that we tend to treat computers and other media as if they were real people. We attribute gender and personality traits to them and allow them to affect our emotions. While some have argued that AI services don’t need to be human, people will personify them whether we want them to or not.

As a result, brands can’t develop AI services solely using best practices for product design. Instead, they need to incorporate elements of character design. They need to decide if a brand is a single personality or a set of characters that hand over to each other (or a human) at the appropriate time. And they need to make the characters believable and relatable. Above all, this requires serious real-world, qualitative testing. We have to build the characters and then take them out for extensive test drives to see what works and doesn’t.

Consider context

Our expectations of people (and by extension, characters) are heavily dependent on context. You don't want the person on the end of an emergency call to crack jokes, or a customer-service agent to tell you they’re having a bad day. You need AI that understands if it’s morning or night, whether you’re at work or play, and what exactly was on your mind the last time you talked.

A truly human AI service gets this. For example, a good chatbot should be able to pick up conversations where it left off. It knows if it’s raining or whether you’re taking a plane to Kansas today. It can also remember interactions, so it understands the context of what’s happened before. In order to achieve this, we need to build in a rich context that influences how a character responds.

Make it rich and believable

In Pixar’s Toy Story, there’s a scene in which Wheezy the penguin needs to cry in order to push the story along. But the Toy Story world has strict rules. Wheezy is a plastic toy that cannot shed tears. Instead, the movie’s creators set the shot behind a pane of glass in the rain. A raindrop runs down in front of his face and has the same effect.

Whenever you build a believable AI-world model, it must have rules that are strictly followed. Just as Disney and Pixar (which was, after all, founded by Steve Jobs) sweat every last detail, we must too. Everything communicates character, from how fast the system responds to the tone of a voice to how transition animations communicate character. And all of it must seem as if it’s coming from the same place.

Be inclusive

Accessibility used to be the thing you did once you'd completed the design and build. It was a box to tick after the fact. But with AI, that’s a little like designing a physical building first, and then thinking about where the accessibility ramps go after. Modern design needs to have accessibility at its core because we don't know who will be using it and how.

With AI, we need to stop thinking about the current trend of "mobile first." We must become platform agnostic. We need to design the service first, and then build the interfaces on top of it.

Be real, not efficient

Currently, we design products and services that reflect an obsession with efficiency and productivity. Everything must be faster, quicker, and better. But if you think about the most memorable interactions with friends you've ever had, they were probably not the most efficient. They were the ones that were most delightful. So, it’s critical that in our search to build lifelike AI, we abandon our usual tendencies and make something engaging first, and productive second.

In summary, today we’re shifting from a world where people need to be computer literate to a world where computers need to be people literate. If you offer products and services with captivatingly lifelike conversational behavior, you’ll be more personal than personalized. Your brand will be more human than human, which is exactly where it needs to be.

Join the conversation