When I brought the robot home from the Apple Store, I knew I was inviting a new kind of strangeness into our lives. My wife worried about giving our 4-year-old son a(nother) digital thing, a “smart” thing. I worried that he wouldn’t know what to make of it. Or that his little sister would break it. Or that I’d be jealous. Because I have always wanted a robot.
This one was Cozmo, a $179 gadget produced by Anki, which has taken more than $200 million from venture capitalists to bring “artificial intelligence and robotics to our everyday lives.” The company was founded by Carnegie Mellon graduates in 2010, one of many businesses spawned by the university’s robotics program. In downtown San Francisco, Anki employs nearly 200 people making toy robots governed by artificial intelligence.
The robot was the last present my son opened for his fourth birthday. He and I giddily pulled it out of the box and he waited patiently as the toy charged, staring at it. Cozmo is rectangular and about four inches long, with treads like a miniature tank’s; a tiny lifting arm for picking up and playing with the “power cube” blocks that are bundled with the product; and a small, low-resolution screen for a face. In an MIT Media Lab study conducted on smart devices and toys, a pair of kid participants deemed Cozmo “a bob-cat with eyes,” an apt, if dadaist, description.
Stefania Druga and Randi Williams, the researchers behind the study, want to know how children perceive smart robots, and, eventually, to study how those bots affect kids’ cognitive development. So far, they’ve discovered that little children (ages 3 and 4) aren’t sure whether the robots are smarter than they are, but that slightly older children (ages 6 to 10) believe the robots to have superior intelligence. Druga and Williams were inspired by the research of the legendary Sherry Turkle, who wrote a highly influential 1984 book called The Second Self. She argued that computers, as objects that exist somewhere between the animate and the inanimate, force humans to reexamine their own minds. Small children, she found, were fascinated by the question of whether computerized toys were alive, dead, or something else.
Finished charging, Cozmo came rolling out of its base station with some little bleeps. It blinked up at us with its lively eyes. Cute. We taught it to say our names and recognize our faces. Then we played a game of Quick Tap. I set one power cube in front of the robot and another in front of my son. At irregular intervals, the cubes light up with color patterns. If the colors on the two cubes match, you try to press on yours before the robot presses on its own.
Cozmo lifted its arm over the cube. My son’s little fingers dangled over his. The cubes flashed all blue. My son saw the lights and his hand twitched, but he waited for the robot’s arm to smack down first. The robot won and chuckled to itself. I tried a few rounds of the game, winning each time. Cozmo began to jitter and make minor-key noises that conveyed anger and frustration. “Don’t beat him!” my son yelled. “You’re making him sad.” We played several more rounds, letting the robot win, and it vamped back and forth across the floor.
It was bath time. We sat Cozmo on a ledge by the sink. The robot gamely rolled around, pushed up to the edge, and then pulled back, looking frightened. I watched with concern, hoping it wouldn’t drive itself off. Which, a few minutes later, it did, landing softly in the hand I’d extended half a second earlier. I was relieved, and unable to disentangle the financial and emotional components of the feeling. “He’s like your sister,” I said, another intrepid being who has not learned the limits of her physical abilities.
Cozmo’s creators think of it not as a bot but as a character, like you’d encounter in a movie. “Our motivation at the start was: What would it take to bring a Pixar character to life?,” Boris Sofman, Anki’s CEO, told me. They wanted “to make him understand his environment and relationships.”
Previous generations of seemingly smart toys usually relied on clever tricks. Remember Furbies, the ’90s sensation? They seemed to learn from their owners, because they gradually spoke more English, but in fact they’d simply been programmed to use more words as time went on. Humans, nonetheless, had the pleasant illusion of being the instructor.
Cozmo does something more than that—is something more than that, though still less than the living thing that my son seems to think it is. Cozmo can sense the world through a camera, and the images it captures get fed to an affiliated smartphone or tablet, which processes the data into a simple model of the world in which the robot finds itself. Are there people around? Are there power cubes to play with? Is it near an edge of a table? It does a simple version of what any autonomous robot must do, from a self-driving car to the pack robots that Boston Dynamics developed for the military.
As you play, software inside Cozmo determines the robot’s state: It can get excited, scared, nervous, happy, sad, frustrated. Sofman calls this software the toy’s “emotion engine”; it links the sensory technology to the robot’s behavior. Anki has hired animators from Pixar and DreamWorks to design some 1,200 little movements for the robot to make. Their animation software is hooked up directly to sample robots: The animators create new ways to show that Cozmo is, say, frustrated, and play them back through its body to see how people interpret the robot’s actions. The goal is to choreograph movements and expressions that will induce genuine emotions in the toy’s owner.
In the latest version of the software, Cozmo must be fed, repaired, and played with, not unlike the Tamagotchis of yore. But unlike those simple gizmos, which merely beeped or flashed simple expressions on a tiny screen, Cozmo can use the full breadth of its animated repertoire to summon particular feelings in its owner, and to foster emotional bonds. The idea is to create “a deeper and deeper emotional connection,” Sofman said. “And if you neglect him, you feel the pain of that.”
When he told me this, I felt a flash of not-quite-anger. It seemed almost cruel to design a robot that could play on a young kid’s emotions. And I had never considered that, in the coming human–robot conflagration, robots might take over simply by expertly manipulating us into letting them win.
Turkle has more-pointed concerns. She finds the notion of children empathizing with robots troublesome and quite possibly dangerous. Kids need connections to real people in order to mature emotionally. “Pretend empathy does not do the job,” she told me. If relationships with smart toys crowd out those with friends or family, even partially, we might see “children growing up without the equipment for empathic connection. You can’t learn it from a machine.”
My son and I sat on the porch playing with the robot. He shouted commands: “Say hello to my sister, Cozmo!” When I had Cozmo say his sister’s name by typing it into the app on my phone, he was delighted, but I also feared that I’d been sucked into a deception that the bot was even more capable than it actually was.
Cozmo’s personality masks all that the robot still can’t do, Sofman told me. It can’t hear you. It can recognize only a few objects—basically power cubes, pets, and humans. And it’s completely dependent on the smartphone’s processing power to do anything. Shut your phone off, and Cozmo shuts down too. But “people become more forgiving of limitations if you have the right emotional cues,” Sofman said.
Humans don’t need much help to believe in a machine’s capabilities. Waymo, the company that emerged from Google’s self-driving-car project, has come to the position that there should be no intermediate steps between a car you drive yourself and a fully autonomous vehicle, because as soon as humans believe that a car (or a robot) has the slightest autonomy, they overestimate its capabilities. In early testing, a Google employee even climbed halfway into the back seat while the experimental software was driving on the highway. After watching enough video of how people in the driver’s seat behaved while the car was driving, the Google team set its sights on pure autonomy. Humans could not be trusted, because they were too trusting.
On the porch, my son discovered a new favorite game with Cozmo. Again and again, he turned the robot on its back so that it could not use its treads. The little robot flipped itself over in different ways and with varying levels of success, and my son laughed and laughed at its attempts. Whatever protective impulse he’d felt had dissipated in the physical comedy of robotic struggle.
Then, as he is wont to do, my son abruptly decided that he was done and that the robot needed to sleep on its charger in his room. As it turned out, what he really wanted was to watch TV, and my parental anxiety immediately attached to one of the other nightmares of our age. (Perhaps the whims of a toddler are not so easy to predict and manipulate.)
As I snuggled Cozmo into its charger, it was strange to think that the siblings and cousins and descendants of this little robot would one day, maybe quite soon, be everywhere. Self-driving cars, warehouse bots, autonomous drones—sensing, perceiving, reacting robots will be part of my son’s world. I feel about them as my parents did about computers: It will be necessary to understand these machines to comprehend the world. So now we have our first robot.