Three years ago, I talked to a team of designers at Mattel who were so excited they could barely contain themselves. They had developed what they believed to be the next big hit for kids: An $300 AI speaker called Aristotle. It was essentially an Amazon Echo reimagined as a nanny to read to your children and keep them entertained, and it could connect to Barbies, Hot Wheels, and more Mattel toys. Yet within a year of its announcement, Aristotle got canned because parents freaked out about the privacy implications of a smart assistant in a child’s room.
Aristotle was not a terribly thought-out product. But it demonstrated that making devices for kids, especially AI-infused, cloud-connected tutors, is extraordinarily hard, even for a company that knows children and parents pretty well.
And now, the latest gamble in this space is a $1,500 robot called Moxie designed by Fuseproject for the startup Embodied. It’s a 7-pound, adorable-as-heck robot that’s built for children ages six to nine. An emotionally aware buddy and teacher, it’s designed to share new lesson plans that download from the cloud weekly.
Embodied was founded by Paolo Pirjanian, who sold a robotic vacuum company to iRobot, maker of the Roomba, before taking over as CTO. Embodied has raised $41 million in funding from an all star group of investors in robotics and AI, including Amazon, Intel, Sony, and Toyota. Chief Creative Officer Craig Allen previously worked at Jim Henson and Disney.
Slated for release this fall, Moxie was originally imagined for children on the autism spectrum, but it’s now being marketed to all young children through a subscription model that promises new educational content each week.
Given the poor sales numbers of social robots (Sony sold just 11,000 Aibo pet robots at its relaunch, which was considered a success but pales in comparison to a hit smartphone), the odds might not be in Moxie’s favor. Companies have introduced many promising social robots over the past few years, but none have managed to stay afloat. There was Anki, which raised over $200 million to take AI-infused robots mainstream, and Jibo, a desktop family robot that raised millions on Indiegogo only to go bust a year after launch. But even if you’re skeptical of the business plan, Moxie is a fascinating case study on robotic UX and building computers specifically to speak to children—a topic that’s more relevant than ever, given how much time kids are spending during the pandemic learning on screens from home.
Building the perfect buddy
Pirjanian’s passion for robots has an unexpected source: animation. He was in his teens when he first saw the famous Pixar short Luxo Jr., which turned two desk lamps into convincing characters. “I was fascinated that you can have pixels on a screen evoke such deep emotions,” he says. Instead of becoming an animator, Pirjanian skipped a step and got his PhD in robotics to literally bring objects to life.
Pirjanian first began working with Fuseproject founder Yves Béhar during Pirjanian’s time at his company Evolution Robotics. Béhar later helped develop the updated design language behind the inconspicuous vacuum cleaner Roomba; you see it in the Roomba s9 models on the market today.
Béhar’s previous work on robots, at iRobot and elsewhere, was aggressively abstract. “I had a theory about taking away human traits and focusing on the experience of a robot rather than duplicating human personality and characteristics,” Béhar says. Béhar didn’t add a smile, eyes, or a face to the Roomba because he thought robots and their AI brains should be defined as machines, not anthropomorphic buddies. Even Béhar’s robot for the elderly, ElliQ, displays the head and neck posture of a small person, but has a single glowing eye as opposed to a face.
But as Béhar met with Paolo’s team and learned about the mission—to connect with children on the autism spectrum—he knew that he would have to change his philosophy. For kids to learn to recognize human emotion, they would need the robot to have human elements, including facial expressions and gestures.
I briefly demoed Moxie through Zoom, and the robot’s charm is irresistible. Early versions were loaded with articulating arms and hands, which are motor-driven pivot points that added both expense and points of failure to the design. Refined to what I see today, Moxie’s body resembles that of a pudgy penguin, with two flapping arms that pop out with excitement to emphasize points, and a single finger on each hand. One bending articulation around Moxie’s waistline affords the robot a surprising amount of motion. The design team tried to eliminate it, but realized it would sacrifice too many movements that bring Moxie to life.
Some of the industrial design details that give Moxie personality disguise important functions. Moxie’s head is shaped like that of a Smurf or Hershey’s Kiss, and it hides four microphones that allow Moxie to turn to anyone in the room who calls for its attention. An earhole on each side lets children whisper secrets to the robot, but they also serve as vents for all the heat that builds up in Moxie’s noggin.
And then there is that face. Glowing blue on a custom-made curved screen, it allows Moxie to express everything from sadness to laughter, with oversized green eyes that change shape from squares to tear drops in milliseconds. Moxie’s eyes are like if Wall-E‘s Eve got an anime makeover. Moxie actually tracks your gaze with a camera in its forehead so that it can make eye contact. The illusion is convincing through my webcam. Sometimes, as Pirjanian speaks, I’m distracted by Moxie, which appears to be giggling at some secret conspiracy, like a student acting quietly up in class.
But what does Moxie actually do? It’s designed to carry on a conversation with children without pesky wake words like “OK Google.” Look past the charming industrial design, and conversation is the the core of Moxie’s UX, which means how well it can actually converse is the ultimate value proposition. What’s still unclear to me, despite asking the company directly, is exactly what Moxie’s lesson plans will include. What is the actual content, and how is it presented within the flow of conversation? Is Moxie a friend who likes to talk about feelings, or a serious instructor who might teach your child math?
The deeper design consideration
I like Moxie, and of course I do: Moxie has been designed to lure me in, like a G-rated WestWorld host. Could Moxie be too powerful of a draw for children? Like many parents, I worry about exposing my child to too much screen time, and I’m even more worried about their privacy. Amazon’s Alexa is a boring, black cylinder, and yet you still hear stories about how kids become obsessed with it. Imagine if Alexa had eyes.
Pirjanian offers me all the security promises you’d expect: Moxie handles tasks like its vision AI locally, rather than in the cloud. A child’s data that is uploaded will be encrypted and protected by a single key that only the parents have. (And if that key is ever lost, well, so is your child’s history with Moxie.)
But it was encouraging to hear some of the deeper, social considerations that Pirjanian’s team is making, the largest of which is this: Moxie cannot be used for over an hour a day, and most lessons end at the 35-minute mark.
“One thing we thought about a lot is that this is a technology to help children with emotional social communication skills…as opposed to other technology which is trying to get addictive loops going, and allow children to binge,” says Pirjanian. “Moxie is not a destination but a springboard to the real world. We decided a child could not binge and go through two years of content in a month.”
So even though Moxie’s battery is good for four hours, when a child’s hour with Moxie is up, the robot will explain that it is getting tired and take a nap. And the child will have to wait until tomorrow to hang out with Moxie again.
There are a lot of unanswered questions about Moxie: Has the robot truly cracked the code on conversing with children—a challenge that’s hard, not just for AI, but for human adults? Will it be an effective emotional tutor for children on the autism spectrum (Pirjanian himself is making no claims until he says he has the clinical data to back them up)? And how many people, even if they want Moxie, can afford the $1,500 price and the $60 a month subscription fee that kicks in after the first year?
For all of these reasons, Moxie, like other social robots, will face an uphill battle in the market. But it looks a whole lot more appealing than the alternatives my generation got: MyBuddy and Teddy Ruxpin.