The robot is shaped like a human, but it sure doesn’t move like one. It starts supine on the floor, pancake-flat. Then, in a display of superhuman joint mobility, its legs curl upward from the knees, sort of like a scorpion tail, until its feet settle firmly on the floor beside its hips. From there, it stands up, a swiveling mass of silver limbs. The robot’s ring-light head turns a full 180 degrees to face the camera, as though possessed. Then it lurches forward at you.
The scene plays out like one of those moments in a sci-fi movie when the heroes think for sure the all-powerful villain must be done for, but somehow he comes back stronger than ever. Except it’s a real-life video released last month by the robotics company Boston Dynamics to introduce its new Atlas robot. The humanoid machine, according to the video’s caption, is intended to further the company’s “commitment to delivering the most capable, useful mobile robots solving the toughest challenges in industry today.” It has also freaked out many people, and the video has garnered millions of views. “Impressive? Yes. Terrifying? Absolutely,” wrote a reporter for The Verge. Terminator and I, Robot memes abounded. Elon Musk suggested that it looked like it was in the throes of an exorcism.
You might think that such reactions would concern Boston Dynamics, that it would seem bad for the public to associate your product with dystopian sci-fi. But the company is used to this. Over the past decade-plus, Boston Dynamics has become arguably America’s most famous robotics company by posting unnerving viral videos that elicit a predictable cascade of reactions: things like “Could you imagine this thing chasing you?” and “We’re doomed.” When the company posts a video like the one of the new Atlas, and viewers get worked up, it all appears to be part of the plan.
Even if you don’t know Boston Dynamics by name, there is a good chance you have seen one of its videos before. Clips of robots running faster than Usain Bolt and dancing in sync, among many others, have helped the company reach true influencer status. Its videos have now been viewed more than 800 million times, far more than those of much bigger tech companies, such as Tesla and OpenAI. The creator of Black Mirror even admitted that an episode in which killer robot dogs chase a band of survivors across an apocalyptic wasteland was directly inspired by Boston Dynamics’ videos.
The company got into the viral-video game by accident. Now owned by Hyundai, Boston Dynamics was founded in 1992 as a spin-off of an MIT robotics lab, and for years had operated in relative obscurity. In the 2000s, someone grabbed a video off the company’s website and uploaded it to YouTube. Before long, it had 3.5 million views. That first YouTube hit is when “the light went on—this matters,” Marc Raibert, the founder, has said. (Boston Dynamics did not provide an interview or comment for this story.) In July 2008, the company created a YouTube channel and began uploading its own videos. Almost every one topped 1 million views. Within a few years, they were regularly collecting tens of millions.
Many of Boston Dynamics’ videos seem engineered to fuel people’s most dystopian fantasies, such as the one in which it dressed its humanoid robot in camo and a gas mask. But the company is careful not to lean too far in this direction. Alongside videos of the robots looking creepy or performing incredible feats, it has offered ones in which the robots failed spectacularly, were bullied by their human makers, or did silly dances; in response, people professed to feeling “sorry for” or “emotionally attached to” these robots. The company’s recent farewell video for its old Atlas model, retired days before the new one was released, included clips of the robot toppling off a balance beam and tumbling down a hill. “What we’ve tried to do is make videos that you can just look at and understand what you’re seeing,” Raibert told Wired in 2018. “You don’t need words, you don’t need an explanation. We’re neither hiding anything nor faking anything.”
Boston Dynamics has not said much publicly about how it trains its robots. But when viewers watch videos of the recently retired hydraulic Atlas doing parkour, they might well assume that if it can execute such complex maneuvers, then it can do pretty much anything. In fact, it has likely been programmed to perform a handful of specific tricks, Chelsa Finn, an AI researcher at Stanford University, told me last year. As I wrote then, robots have lagged behind chatbots and other kinds of generative AI because “the physical world is extremely complicated, far more so than language.” The company posted its first video of Atlas doing a backflip in 2017; more than six years later, the robot still is not commercially available. “The athletic part of robotics is really doing well,” Raibert told Wired in January, “but we need the cognitive part.”
The actual business of Boston Dynamics is comparatively mundane. Currently, its humanoid robots are purely for research and development. Its commercial products—a large robotic arm and a small robotic dog—are used mainly for moving boxes and workplace safety and inspections. “The perception of how far along the field is that we get from these highly curated, essentially PR-campaign videos … from different companies is a bit distorted,” Raphaël Millière, a philosopher at Macquarie University, in Sydney, whose work focuses on artificial intelligence and cognitive science, told me. “You should always take these with a grain of salt, because they’re likely to be carefully choreographed routines.”
The company, for its part, has gestured at the limits of its robots in press releases and YouTube descriptions. But it still keeps posting dystopian videos that keep freaking people out. “They probably made a calculated decision that actually this is not bad press,” Millière said, “but rather, it makes the videos more viral.” The company recognizes that we love fantasizing about our own demise—to a point—and it supplies regular fodder. The strategy has paid off. Now pretty much all the top robotics companies post video demonstrations on YouTube, some of which are more advanced than Boston Dynamics’. Its video introducing the new Atlas robot garnered more than twice as many views as this frankly far more impressive video from the lesser-known robotics company Figure.
In recent years, AI companies seem to have taken a page out of the Boston Dynamics playbook. When OpenAI CEO Sam Altman talks about the existential threat of superhuman AI, he is in effect deploying the same strategy. So, too, are the other executives who have invoked the “risk of extinction” that AI poses to humanity. As my colleague Matteo Wong has written, AI doomerism functions as a fantastic PR strategy, in that it makes the product seem far more advanced than it actually is.
Boston Dynamics is poised to benefit from the revolution those companies have delivered. Hardly a week after the launch of ChatGPT in late November 2022, the company announced the creation of a new AI Institute. Last month, it posted a video about using simulations and machine learning to teach its robot dogs how to move through a range of real-world environments. And the press release for the new Atlas robot explicitly talked up the company’s progress in AI and machine learning over the past couple of years: “We have equipped our robots with new AI and machine learning tools, like reinforcement learning and computer vision, to ensure they can operate and adapt efficiently to complex real-world situations.” In normal English, Atlas might soon not just look but actually be, in a certain sense, possessed. Now that would really be scary.
