Recently, Boston University scholar James Bessen reviewed Tyler Cowen's new book, "Average Is Over," for the Switch. Cowen argues that the rise of ever more sophisticated software will destroy middle-class jobs, leaving us with a world of widening inequality. Bessen argues, to the contrary, that as information technology matures, it is likely to create a growing number of decent-paying jobs for moderately skilled workers.
Kevin Drum, a blogger for Mother Jones, didn't like Bessen's argument, calling it "literally the worst possible case you can make for the continued relevance of the middle class." Ouch. In a tweet, I described Drum's post as "sneering and thoughtless." He pleads guilty to sneering but denies his post was thoughtless. So let me substantiate that charge and then get to the meat of his argument.
Drum started his post with "I haven't read Tyler Cowen's Average Is Over." And that's the problem. "Average Is Over" is not a book about a science-fiction future in which computers can do absolutely everything human workers can. Obviously, that world will be radically different from our own, and it might very well be a bad environment for middle-class workers. Rather, "Average Is Over" is about a medium-term future in which computers are getting steadily better at tasks like driving and operating factories, but there's still significant room for human labor to complement the efforts of the machines. Drum faults Bessen for not answering the question, "What happens to human labor when machines are smart enough that they need virtually no human guidance at all?" But that's not what "Average Is Over" is about, so it's bizarre to fault Bessen for not addressing it in his review of the book.
But should middle-class workers be worried about machines becoming so smart that they take everyone's jobs? The argument that Drum and other futurists make for this proposition dramatically overestimates the importance of raw intelligence in the labor market.
Consider computer chess. Chess-playing computers are obviously smart in some sense, and they're getting smarter every year. But as Cowen puts it, "no one fears that the programs will stage a coup and and take over the United States Chess Federation." Computer scientists have become skilled at building computers to solve individual "smart" tasks like playing chess, playing Jeopardy, recognizing speech and driving cars. But computers perform these tasks in very different ways than humans do, and there's no reason to think that progress on these individual problems will eventually lead to the creation of machines that think and act like human beings.
And being able to act like a human being is an essential qualification for a large fraction of the labor market. We might be able to design a robotic nurse capable of performing the medical functions currently performed by nurses, but patients are likely to prefer to interact with human beings in the hospital. Computers are good at math, but children are unlikely to ever respond to a virtual math teacher as well as they will to a flesh-and-blood one. A computer might be better than any human being at enumerating the virtues of a product and answering technical questions, but no computer is going to develop the kind of customer rapport that a good human salesman can.
This isn't because computers aren't "smart enough." It's because raw intelligence isn't the only qualification for these jobs. Human beings are social creatures. We care about our interactions with other people in ways that we'll never care about our interactions with machines, no matter how intelligent they might be. Already, jobs with a social component account for a large fraction of the workforce. As machines become ever better at performing purely mechanical tasks like driving a truck or assembling an iPhone, the fraction of jobs with a significant social component will approach 100 percent.
And as both Bessen and Cowen have pointed out, jobs with a social component often complement the purely mechanical kind. So as the latter get cheaper, demand for the former will grow. As ATMs made it cheaper to provide basic teller services, they stimulated demand for human beings to explain high-margin banking services to customers. Drum interprets this as a sign that computers aren't yet "smart" enough to perform these functions, but that's not true; there are plenty of Web sites that will let you sign up for a mortgage or a retirement account. The point of having humans sell these services is that customers more easily connect with a human bank employee than a bank Web site.
More generally, as increasingly sophisticated automation pushes down the cost of manufactured goods, society will allocate a larger fraction of its income to math tutors, psychologists, yoga instructors, bartenders and others whose jobs aren't amenable to automation.