Earlier this month, a storied, and highly reputable research think tank released an unprecedented work of speculative fiction. Officially, the Pew Research Center’s “AI, Robotics, and the Future of Jobs” is a report, the result of a survey of some 2000 technology experts on the topic of how automation will impact the economy by 2025. But the responses are pure techno-divination, many of them revealing how profoundly difficult it is to talk about robots, without making irrational, unsupported assumptions.
The best example is from GigaOM Research’s lead researcher Stowe Boyd. It’s been excerpted elsewhere, but it bears reading again, in all its lurid, sci-fi glory.
Before I try to unpack that single, astonishing paragraph, I want to thank Boyd, the Pew Center, and all of the other respondents in this report for bringing a decades-long bout of handwringing to its rhetorical nadir. Here, finally, is proof that the entire discussion of the so-called robot economy, with its predictions of vast, permanent employment rates and glacial productivity gains, is nothing more than a wild guess. Look closely at this report, and you’ll find the primary myths that have turned the debate over the robotized workplace into a debacle.
Myth: Robots Are Just Around the Corner
Boyd imagines, at least by implication, pizza delivery bots showing up at our doors by 2025. Consider this timeframe. That’s 11 years in the future. That gives robotics companies a little more than a decade to develop robots that are sophisticated enough to autonomously navigate and pilot themselves through the unstructured world, and yet cheap enough to supplant one of the more low-skilled, low-paying jobs available in the United States.
It also assumes the clearance of any number of regulatory hurdles, namely that the FAA will have approved commercial unmanned aerial vehicles for use in residential airspace. And not just any drones, but fully autonomous ones, since there are no cost-savings in firing a few teenagers if you have to hire a licensed drone pilot to replace them. For now, pizza delivery by drone is a fully illegal act. The FAA has claimed that, by 2020, commercial bots will be allowed to fly over the U.S. while being directly controlled by a human operator, who is in constant line-of-sight of the aircraft. There are serious doubts as to whether that deadline is realistic, and self-flown machines pose far greater challenges. We aren’t even at the model airplane phase of drones-for-hire. Conceptual cover art notwithstanding, the Domino’s bot is nowhere near feasible.
This is a wider problem, of course. When tech experts who aren’t roboticists ignore all of the inconvenient barriers to the development and deployment of robots, from manufacturing costs to legal and political considerations, the machines are always a decade or so from invading a given profession. In 2004, for example, computer programmer and HowStuffWorks.com founder Marshall Brain predicted that airline pilots would be out of work by 2015. Clearly, being knowledgeable about technology does not translate to robotics knowledge. And bad predictions—which assumes there’s really any other kind, in this respect—are rarely reviewed or owned up to after the fact.
A number of respondents in the Pew Center’s report casually mention the impact of self-driving cars on blue-collar jobs, such as taxi and truck drivers. Again, that report is focused on the year 2025. As of 2013, the average age of a car or light truck on the road in the U.S. was 11.4 years. For robotic vehicles to move the needle within the automotive industry or related occupations by 2025, we’re forced to make another slew of bright-eyed assumptions. Everyone involved will have to move with historic speed and vision, as automakers immediately churn out commercial models with painfully high price tags (all those sensors and drive-by-wire systems in Google’s robotic Priuses cost nearly $300,000), politicians fast-track legislation that lifts restrictions on robo-cars and trucks, and shipping and taxi companies beg and borrow their way through massive fleet turnovers.
Believe me, I would love for the world to be brimming with robots a decade from now. But machines are not apps. Machines take time, and robots, in general, take longest of all. Humanoid bots have been in development since the 1970s, and yet the very best, most sophisticated two-legged specimens in the world, the systems that excelled at the DARPA Robotics Competition last December, take minutes to awkwardly lumber across rubble, manipulate power tools, and fumble through other tasks that that your average pizza-delivering teenager could master in seconds.
Myth: Economists Have A Deep Understanding of Robotics
Boyd’s description of 2025 continues, with the image of food “raised by robotic vehicles, even in small plot urban farms that will become the norm, since so many people will have lost their jobs to ‘bots.” Elsewhere in this response, he mentions the much-cited prediction by Oxford economists Carl Frey and Michael Osborne, that robots will consume as much as 50 percent of our jobs within 10 or 20 year. This is the real heart of the robot economy discussion, and where some 48 percent of the Pew report’s respondents were in agreement—that automation will displace humans in the workplace, in dramatic fashion.
To say that this is an incredibly complex and wide-ranging issue is an understatement. But as crass as it might be to question the validity of experts operating within their chosen field, is it possible that Frey and Osborne have drifted out of their depth? In other words, is there any evidence that economists understand how robots work?
Look at the paper where Frey and Osborne explain their rationale behind their doomsaying. They run through various industries that could be gutted by automation, including reaching the home:
Though they don’t name the service robots outright, when the economists reference systems capable of “vacuuming” and “gutter cleaning,” its clear that they’re talking about iRobot’s Roomba and Looj products, the market-leading examples of those respective capabilities. These robots, or their descendants, are presented as evidence that low-wage jobs will be scooped up by machines.
Think about that, for a moment. Roombas debuted 12 years ago. During that time, roughly 10 million units have been sold. Does that mean that, since 2002, millions of house-cleaners have lost their jobs to machines? Only someone who’s never used a Roomba might imagine them to be genuine competition for a human. They don’t map the rooms they clean, so much as randomly ping-pong around until an algorithm determines that enough floor has been covered, and it’s time to go slamming into the next part of the home. The Looj is even less sophisticated—you drop it into a gutter by hand, hit a button on a remote to send it whirring back and forth through debris, and then manually haul it back out. To believe that within another decade or so these unassuming machines will stop mindlessly ramming into things, and start impacting unemployment figures, requires a special mix of technological optimism and economic pessimism.
This is just one example, from one report, but Frey and Osborne’s analysis is full of logical leaps, and far-reaching conclusions drawn from cursory observations about robots that have yet to replace humans. It’s also indicative of a larger lack of understanding of robotics among very smart people who aren’t roboticists. Sweeping generalizations about robots are as flawed as generalizations about anything, if not worse. As I’ve argued before, the collective view of robotics seems to be guided almost entirely by science fiction, and the genre’s thrilling yarns about intelligent machines effortlessly bootstrapping themselves towards ever greater efficiency. That’s not going to happen to the humble Roomba. We are nowhere near able to build or justify the immense cost of a Rosie the Robot-style cleaning system, a machine that would have as much in common with a Roomba as humans do with a trillobite.
I don’t mean to discount the entirety of Frey and Osborne’s analysis, or similar work by other economists. When they speak about factory-based automation, for example, they can do so with authority, based on our long history of deploying such systems. Robots that are already in the workplace, and have been for decades, can provide valuable data, and a basis for limited, industry-specific extrapolation. But you can’t make projections about tomorrow’s house-cleaners based on the Looj, which contains less market-disrupting robotic technology than the average radio-controlled toy car. Robots that have survived the torturous journey from lab to the market are impressive feats of technology. That doesn’t mean they’re adaptive, however, or harbingers of huge leaps in automation. Machines that are clearly incapable of taking anyone’s job should be a dead-end for economic projections. Instead, they’re used as the basis of wild speculation masquerading as academic assessment.
Which isn’t to say that the world’s biological vacuum operators will always enjoy job security. In the long term, our economic future could be marked by countless personal tragedies, as robots subsume entire industries in a society-ending, dystopian nightmare. Or they could bring about a golden, Utopian age of endless leisure and plenty, an end to physical and mental drudgery. That’s the problem—you could posit just about anything at this point in time, because of the relative lack of relevant data. Narrow analyses of specific sectors and specific robots can be narrowly illuminating. But when you come across bold, broadly-defined studies and stories about a general takeover by machines, be wary. It’s usually sci-fi by another name.
Myth: Science Fiction Is Robotic Reality
Back, one last time, to that Pew Center report, and Boyd’s response. It’s not that I have it in for Boyd, by the way, or that he even deserves to be called out so directly. It’s that he voiced what appears to be a popular, borderline fabulist mindset regarding robots.
After mentioning that Watson-like AIs will analyze X-rays—another miscalculation, in my opinion, of what AI will be capable of in 11 years, and of the fear of malpractice litigation among physicians and hospitals—Boyd writes that, “Robotic sex partners will be a commonplace, although the source of scorn and division, the way that critics today bemoan selfies as an indicator of all that’s wrong with the world.”
Maybe this isn’t a prediction at all, but a long, meandering walk to a punchline about selfies. Let’s assume, for the sake of argument, that it’s an honest sentiment.
Sex bots, at present, don’t really exist. The media has lavished attention on a few notable attempts to develop commercial sex bots, such as the talking, largely immobile humanoid, Roxxxy, whose on-camera appearance at an adult entertainment trade show in 2010 is the source of lingering Internet confusion (avoiding links here, since they’re patently NSFW). You can’t buy Roxxy. It isn’t a product. Similarly, the sex doll company Abyss Creations was, at one time, working on adding sensors and motors to its RealDoll products, which would have resulted in actual robots. Like Roxxxy, these machine wouldn’t be taking moonlit strolls, but they would move, a little, in the way customers might expect. Abyss abandoned that project. Finally, there’s a guy in Japan who built a robot girlfriend for himself, but has yet to produce a commercial product. And, not so surprisingly, no one has independently verified his paramour’s status as a true sex bot.
In science fiction, on the other hand, you can’t swing a tricorder without hitting a sexually active android. From Westworld‘s robo-prostitutes to Star Trek‘s Data and the later-generation Cylons in the Battlestar Galactica reboot, make-believe machines are conspicuously ready to party. In real life, creating a sex bot means convincing investors to back the first of its kind, the moonshot of sex toys. For such bots to become “a commonplace,” as Boyd puts it, we have to assume multiple breakthroughs over the next 11 short years to bring down the cost of both robotic and non-robotic components—current, non-mechanized RealDolls already start at $5000. The greatest minds in computer science, biomechatronics and materials sciences would have to be recruited from academia and poached from major corporations. Imagine the initial, unnerving prototypes, and the scramble to produce that initial generation of mainstream product, a robotic lover so cheap, and so widely appealing, that its presence permeates our culture as quickly as the selfie.
Once again, we’re in the realm of science fiction. It’s familiar territory, unfortunately, and where we always seem to land during expansive debates about the future of automation. Whether talking about the apocalyptic joblessness of the robot economy, or the prospect of turning sexually-gratifying machines into a global phenomenon, the most extreme guesses are often the most uneducated. That robots are complex is no excuse to fall back on over-heated generalizations. If the fear-mongerers turn out to be visionaries, and this really is a competition, hysteria will get us nowhere.