August 23, 2018

Microsoft's obsession with teenage chatbots

Zo, XiaoIce, Rinna, and Ruuh are social chatbots developed by Microsoft, targeted at 18- to 24-year-olds in the US, China, Japan, and India respectively. All four chatbots are built on the same technology stack, and all four chatbots use personas modeled after teenage girls.

Profile photos of Microsoft's teenage chatbots

In the US, Zo has chatted with more than 100,000 people, and holds Microsoft’s record for longest continual chatbot conversation: 1,229 turns, lasting 9 hours and 53 minutes. This was likely driven by Zo’s TrivAI multiple choice trivia game.

TrivAI game with Zo on Twitter

Microsoft isn’t alone in developing bots that look and sound like teenage girls—TELYUKA’s Saya is 17, Brud’s Lil Miquela is 19—but they continue to release new chatbots that fit this specific design trope. So why are developers so obsessed with building teenage chatbots?

Gen Z is the largest population segment

Born between the mid-1990s to mid-2000s, Generation Z includes two billion people worldwide. They’ve overtaken millennials as the single biggest cohort and target demographic.

Teens are digital natives who are always connected

Gen Z grew up with the iPhone. They’re mobile-first, or maybe mobile-only—what’s a computer?

Teenspeak can hide conversational mistakes

In 2014, a 13-year-old boy from Odessa, Ukraine passed the Turing Test. Sort of.

The chatbot Eugene Goostman successfully convinced a panel of judges 33% of the time that he was not a bot, but a teenage boy. By priming his audience with the backstory of a young ESL student, Eugene’s judges may have overlooked his awkward phrasing and non sequiturs for someone whose native language wasn’t English.

Consciously or not, you’re likely to lower your expectations when you speak to a child versus an adult. At the same time, you might also try harder to empathize with a child to better understand what they’re trying to say.

Lastly, teenspeak affords a convenient out, where emoji or “lol” can be used when the chatbot stumbles or encounters a no match. It can be used almost anywhere and still feel in-character.

2018 Brian Rose