Relax—we’re a lot further away from an automaton takeover than you may think
The MIT spinoff company Boston Dynamics has become famous from its videos on YouTube showing off its nine robot creations: Atlas, Spot, Spot-Mini, BigDog, Wildcat, LS3 (a kind of robotic mule), Handle (an extendable, mobile upright manipulator), SandFlea, and RHex (a small all-terrain robot).
As the names suggest, most of the robots are patterned on ordinary animals, and one on human beings (Atlas). They perform astonishing tasks, such as turning doorknobs and walking through the opened doors; exploring the grounds of the company’s Waltham location; prevailing over attempts by lab personnel to thwart their efforts by kicking, pulling, pushing, and other acts of brute force; trotting, galloping, bounding, and jumping over obstacles; walking while maintaining their balance on rocky, hilly, or snow-covered ground; getting back on their feet when they trip or are knocked over; serving as artificial beasts of burden in simulated Marine combat; and, in the case of Atlas, jumping up on platforms, doing backflips, and stacking 10-pound boxes on shelves.
All of this appears to be irrefutable evidence that the scientists at Boston Dynamics have created artificial organisms that match and even surpass their real-world human and animal counterparts, reinforcing the widespread belief that robots are about to “take over.”
But not all is as it seems.
Legged motion is a much more complicated affair than we imagine. It requires more neurological resources in terms of sheer brain mass to walk on two legs than it does to play chess or prove mathematical theorems, both of which computers have been doing since the 1950s. Just standing in place is a virtuoso performance, requiring the ability to sense the force and direction of gravitational pull through the soles of the feet, while making thousands of instantaneous micromuscular adjustments in legs, arms, and torso. The adjustments must interact in a complex feedback loop that compensates when the body distributes too much weight in one direction or another. And yet standing still is a minor achievement compared to walking on two legs. Not only must the body continue its delicate and constantly recalibrated balancing act, but it also has to fall and simultaneously recover each time it moves a foot forward, while adjusting its frame to a perpetually varying center of gravity.
When we consider the range of possible postures, gaits, and speeds involved in walking, running, leaping, and so on, the number of computations the brain must perform increases dramatically. And while four-legged, six-legged, and eight-legged creatures are more stable than the bipedal variety, their movements present more difficult problems of coordination.
In short, designing and programming a robot to walk is an enormously challenging task.
In the 1980s, the Leg Lab at Carnegie Mellon University took on the challenge of programming a walkable robot. Directed by an associate professor then in his thirties, Marc Raibert, the lab developed a number of one-legged (hopping), two-legged, and four-legged robots based on biological organisms, following a research paradigm known as “biorobotics.” The basic idea is that it would be a waste of time to reinvent nature’s answer to the wheel when the evolutionary process has already solved the problem of legged motion. All we need to do is analyze the physiological structure, neurological connections, and repertoire of motions of legged animals and find ways to transfer our discoveries to the bodies of robots, equipped with sensors and motors as well as computers running the appropriate software.
None of this is easy. Projects often take years from conception to implementation, and involve teams of computer scientists, mechanical and electrical engineers, and biologists. But with sustained effort, and ever so slowly, nature begins to yield up its secrets as it did at the Carnegie Mellon Leg Lab. After six years at CMU, Raibert moved his operation to MIT, where he continued to direct it until leaving academia in 1995. The reason? He started a new company, named Boston Dynamics.
The pattern is familiar. Someone is hired or elevated at MIT to the status of a senior professor, assembles a team of promising graduate students, attracts major grant money, does research, and publishes papers for a few years, all the while making business and government connections. The academic finally leaves the university in order to start a company. By that time, it’s likely that he or she is already a multimillionaire from outside contract work.
The motive for creating a start-up company undoubtedly varies from person to person. For Raibert, it may have been a desire to enrich himself, or perhaps the belief that growing profits would free his research from the limitations of grant funding. Whatever the truth, Raibert began like many others in his position, by going to the well he knew best from his years at CMU and MIT, the Defense Advanced Research Projects Agency.
DARPA is the federal agency that funds high-tech military research. Its most celebrated achievement is the internet. It awarded Boston Dynamics several grants, the most important of which provided funds for turning the robot LS3 into a mechanical pack animal able to carry water and equipment for Marines in the field. There is a video of LS3 under simulated combat conditions, loaded down with supplies and following at a distance the digital QR (Quick Response) code printed on the back of the jacket of the company’s project director, who is walking side by side with the Marine squad leader. The director also operates a remote control console enabling him to command all of the motions of the robot more complicated than QR following. In other words, most of the mind in LS3 is that of a human being.
The LS3 project was a failure. The robot’s huge engine was too noisy for the reconnaissance missions it was supposed to support. DARPA discontinued the grant. What was in the minds of DARPA officials when they funded the project in the first place? Why didn’t they suggest that the Marines buy a few mules or maybe a camel instead? These biological animals have been successfully hauling supplies for centuries, and at far less cost than any conceivable robot. The answer is that mules and camels are useless if your goal is to build a robotic army.
LS3 funding was suspended in early 2015. Two years earlier, Boston Dynamics had been purchased by Alphabet, the parent company of Google, and absorbed into the latter’s “Replicant” division. The end of DARPA funding may have been one of the reasons the company’s buyer became nervous about its acquisition. In spite of the image Google cultivates of encouraging no-strings-attached futurist speculation, it remains as focused on profits as any other corporation. When DARPA funding fell through, and Boston Dynamics also failed to create a marketable consumer product, the digital media giant decided to cut its losses. In 2016, Google (technically Alphabet) found a buyer who took Boston Dynamics off its hands: the multinational Japanese conglomerate SoftBank.
The unsettled financial situation of their company must have caused considerable anxiety in Raibert and his co-workers. Its video productions ought to be seen in that light. Those available on YouTube are not scientific documents, but promotional pieces. They are designed to create an ongoing buzz about the company in order to increase its market value at a time when its future is uncertain, as is still the case. While speaking in 2016 in San Francisco, at an annual tech conference sponsored by the online magazine TechCrunch, Raibert joked that while he used to write academic papers, he now counts YouTube “likes.”
Like many jokes, his was only a slightly distorted version of the truth.
Let’s take a look at some of the more recent videos that have caused a stir online. Three involve the quadrupeds, Spot and Spot-Mini, and one the humanoid, Atlas. As their names indicate, Spot and Spot-Mini are patterned on the model of an ordinary domesticated canine. Spot-Mini is a smaller, lighter, and somewhat enhanced version of its predecessor. The names, however, do more than signal the intention of the robots’ creators. They also shape the expectations of the video audience.
“Introducing Spot-Mini” was made in 2016. It begins inside the Boston Dynamics building in Waltham. Spot and Spot-Mini are in a lineup of robots that also includes BigDog, Wildcat, and Atlas. Spot-Mini is the first to depart from the line, while Spot follows its lead after a short delay. They both walk at a moderate pace with the familiar gait of dogs, although Spot-Mini’s smaller frame is closer to our experience of canines. The two exit an open door, turn to their left, and walk along the outer wall of the building, apparently to explore the corporate grounds.
In the scene just described, Spot-Mini has no head. In those that follow, it has acquired one that looks like a reptilian skull. The head is at the end of an extendable neck folded like an accordion on the back of the robot. We watch the robot using its head as a mechanical gripper, grasping dirty dishes and loading them into a dishwasher, and putting an empty beer can in the garbage. This is nothing like any dog we have seen. In subsequent scenes, Spot-Mini crawls under a table, gets back on its feet after slipping on a banana peel, and brings a guy on a couch a can of beer, but wrestles over the item like an ordinary dog when the man tries to take it. In each scene, the robot appears to be completely autonomous. As of the time I am writing, the video has received more than 11 million views.
The second video, “Hey Buddy, Can You Give Me a Hand?” was released this year. It begins with a headless Spot-Mini approaching the closed laboratory door, looking it up and down, and walking off after realizing there is no way through. A second Spot-Mini, this one with a head, emerges from the shadows. It uses its head-gripper to get hold of and turn the door handle, and its extendable neck to open the door and hold it open. The first, headless Spot-Mini walks through, followed by its polite, better-equipped counterpart. This time, no humans appear in the scene. Although the video has received only 817,699 views, the viewer reaction was more agitated than in “Introducing Spot-Mini,” with comments ranging from “I want one,” to “We’re screwed,” to “The new season of Black Mirror looks promising,” to the serious suggestion that Spot-Mini, which initially received DARPA funding, is part of a globalist conspiracy.
The responses to “Hey Buddy” are exceeded in emotional intensity by the reaction to a third video, “Testing Robustness.” A Spot-Mini equipped with a head attempts once again to open the closed lab door. This time a researcher uses a hockey stick to prevent the robot from getting out. He knocks the robot away from the door handle, slams the door shut after Spot-Mini manages to open it halfway, and finally grabs a strap tethered to the rear of the robot, yanking it away from the door. Despite all obstructionist efforts, Spot-Mini never gives up, finally succeeding in opening the door and exiting. The will of the robot seems relentless. It just keeps on coming, no matter what the man does. This video got more than 3 million views. Comments include “Oh Jesus,” “Awesomeness,” and “Goodbye human race.”
In the three clips, the videographer manages to evoke feelings we have had in the presence of ordinary dogs, and then replace them with shudders when we realize we are dealing with something no one on earth has seen before. The only framework we have for understanding the experience is the apocalyptic science fiction movie—Terminator, AI, Bladerunner, Ex Machina, and so forth.
Why would Boston Dynamics want to provoke apocalyptic anxiety in YouTube viewers?
After more than a year of the Trump administration, we ought to know the answer. The intentional intimation of disaster is good for ratings. It keeps the audience on the edge of its seats, whether a president or corporation is on display.
This is even more evident in a video introducing Atlas. In one scene the humanoid bends to pick up 10-pound boxes and stacks them on shelves. The message is unmistakable: Atlas is coming for your job.
Boston Dynamics videos are science fiction movies. In each, the robot’s actions are caused by a human being operating a remote control console who never appears on camera. We see the console and operator only in early videos of LS3 made by the Marines as well as one of Spot-Mini made by organizers of a robotics conference in Barcelona. The teleoperator is missing in the videos Boston Dynamics produces. The robots in these videos are similar to those in pre-CGI science fiction movies, which were operated by people hidden inside them.
To be fair, “Testing Robustness” is followed by a written comment indicating that a “person (not shown) drives the robot up to the door, points the hand at the door handle, then gives the ‘GO’ command, both at the beginning of the video and again at 42 seconds.” The rest of the time the robot presumably executes its routine without human intervention. But this kind of task-and-time-limited autonomy has been around for decades, while few people know what “driving” the robot to the lab door means or that a teleoperator is necessary for nearly all of its movements. Also, like the fine print at the bottom of a contract, the disclaimer is unlikely to register in the mind of the average viewer. For those who actually read it, the additional disclaimer that “testing does not irritate or harm the robot” adds to the impression that Spot-Mini can be irritated, implying that it is sentient.
In the meantime, the buzz about the company grows louder: “Boston Dynamics Robot Dog ‘Spot’ is Terrifyingly Lifelike” (NBC News); “Boston Dynamics New One-Armed Robot Is Adorable and Terrifying” (Popular Mechanics); “Boston Dynamics Robot Fights Back” (Independent); “Boston Dynamics’ Robot Dog That Opens Doors Is Freaking Out the Internet” (Science, CNBC), “The Latest Generation Atlas Humanoid Robot Is Absolutely Incredible” (TechCrunch).
Nowhere is there mention of remote control consoles or teleoperators.
Reporters aren’t the only ones who overlook this little detail. Many with scientific training seem equally vision-impaired. A software engineer is wildly enthusiastic in his comment on “Testing Robustness” regarding its implications for a project he’s planning to develop robots that care for the elderly without human involvement.
But Boston Dynamics has made little progress in robot autonomy, whatever we might think of abandoning the old folks to nonhuman machines. It is simply not its forte. In a different context, even Stephen Hawking ignored the actual state of robotics and gave in to fevered dreams of an autonomous robot takeover. But we have only the most rudimentary understanding of the workings of our own brains, and we still cannot transfer more than disconnected fragments of their intelligence to machines. We may make some progress integrating AI modules like speech recognition, image analysis, and probabilistic reasoning into more global structures. But considering the small quantity of grey matter inside our own skulls, we may never arrive at the holy grail of a conscious machine.
Scientists have a habit of turning into fools when they profess expertise on matters beyond their own laboratories. It took the uneducated proletarian magician, Harry Houdini, to expose the fraud and trickery of seances after dozens of scientific observers certified the ghostly phenomena as genuine.
In high-tech industries, when money intersects with this seemingly congenital flaw, the result is dissimulating hype for a technologically literate but credulous public, all too willing to disappear down the rabbit hole. Robotics is interesting enough without demanding that we follow the White Rabbit to Wonderland.
The achievements of Boston Dynamics in robot movement on two and four legs are undeniable and sometimes breathtaking. That’s where its real contribution lies. But its videos play on widespread, potentially dangerous fantasies of autonomous robotic armies without human casualties (on our side, that is), old people left in the care of machines, and a near future when robots do all the work. It’s high time we freed ourselves from the techno-marketing hype and began to see AI and robotics for the fascinating, although limited, developments they really are.
Gary Zabel is a senior lecturer in philosophy at UMass Boston, and has a longstanding interest in robotics and artificial intelligence.