Find the right balance between humans and bots.
Before you build that A.I. platform, it might be wise to use real people to perfect the process—and to see if you even need it.
Way back in the 1960s, Joseph Weizenbaum of MIT decided to have a little fun. To demonstrate the artificiality of computers— how soulless and very unlike humans they were— Weizenbaum created a little program called Eliza, which he designed to behave like a bored psychotherapist. When a user asked Eliza a question, it would spit back some programmatic response. Weizenbaum hoped to show how impossible it was for computers to imitate humans.
The opposite happened. People were so taken with Eliza, so charmed by its manner, they treated it as one of them, even knowing they were chatting with a computer. Weizenbaum had given Eliza—the first chatbot—just enough of a personality to make “her” seem sufficiently real to believe in. Over 50 years later, it seems like everyone has a startup featuring a chatbot— some piece of A.I. that handles FAQs or customer service or some feature that a human could do but doesn’t. The secret ingredient is personality: making the robot easy to relate to, even when you know it’s a bot.
While it may seem obvious that the more human a robot (or other A.I.) behaves, the more we’ll let it into our lives, you also need to think about the converse. Not only is it wise to make robots seem human—it’s just as important to make humans seem more like robots.
Here’s what I mean: In any startup, there’s tension between doing something manually and building tech to do it automatically. It’s tempting to say, “Let’s write a chatbot program to handle customer service.” But for most startups, it’s probably easier to have a human do it. Especially when testing new features, it’s often more efficient (if a little more boring) to have someone at a keyboard doing the rote work before you spend precious engineering resources building something that people may not actually want. Doing things by hand initially is usually the smarter way to go, until the demand outpaces the manual process. Then, as John Henry learned, automation wins.
But here’s the twist: Your customers often expect a computer to be behind the curtain, and the more your manual process can seem robotic, the more trust your customers will have in you. That seems backward, but think about how we trust automation to do certain things really well, like build our cars or find us low-priced goods. For many services, letting the customer think there’s some monolith doing the computation or the assembly is mighty reassuring. A number of studies have shown that, in this context, technology equals trust.
A couple of years ago, we played both sides of this game at my startup, when we built an app to help people track their medication usage. On the one hand: When people opened the app, we sent them little supportive messages—and we made sure that, even though these messages were automatically sent, they seemed very authentic. We made the robots seem human.
On the other hand: When we sent reports that charted their progress, we knew our users would be more confident if they seemed to be computer generated, rather than something dashed off by someone. Even though these reports required manual work, we made sure they looked like a computer had spit them out. We made the humans seem like robots.
In a world beset by Covid-19, we might prefer more robots, including home deliveries by drones. The thing is, most people don’t realize how much manual work still goes on inside tech companies—even at Google. Somewhere in the background, there are always humans. Figuring out the balance is the key: what to do manually and what to do automatically—and then balancing humanity with computation to make your customers content with the result. That’s what lets entrepreneurs build fast, break less, and make something we hope people want.