Throughout history, people have predicted the end of work, but so far it has never materialized. Is this time different? In their book The Second Machine Age Erik Brynjolfsson and Andrew McAfee argue it is. Machines already beat people regularly in chess and in the near future they will drive our cars, educate our students or do Amazon’s logistics. The authors call this the 2nd machine age. The first machine age, the industrial revolution, fundamentally transformed the world of the 19th century. In a similar way, the infinite re-combinations of computers and robots will transform our society.
Lao Tzu allegedly said “Those who have knowledge, don’t predict. Those who predict, don’t have knowledge.” Indeed, the authors show that some predictions about the new age have proved remarkably wrong. Until recently, for instance, most experts believed that computers are bad at pattern recognition and good at routine tasks. Today it seems hard problems (computing a strategy for chess) are easy for computers, but easy tasks (found in the care or service sectors) are surprisingly difficult. In general however, the frontier of things machines can do is shifting rapidly. Automatically generated contents in newspapers or machines grading students’ essays are just two examples of what intelligent machines can do. The world seems at an inflection point. The authors invoke the famous Moore’s law that computer power doubles every year and find exponential growth confirmed in many dimensions of technological progress.
Skeptics wonder why this has not transformed into higher growth rates in the last years. Indeed, freely available content and cheap replication make it hard for many economists to see the profitability of it all. And yet, there is massive positive change even if its traces are more difficult to measure. First, the internet, Brynjolfsson and McAfee argue, has risen well-being by much more than growth in GDP; so much, in fact, that the 2nd machine age needs a different metric to think about social progress. The authors estimate that intangible assets would add another 2 trillion to existing capital assets in the U.S. alone. Second, it will need time to make people realize the full potential, just like in the 1st machine age: it took several decades from the invention of the steam engine to using it in transportation and construction. Third, modern information technology lets people increasingly access to the world’s stock of knowledge. Options multiply. This is good news.
The bad news is, the benefits are not evenly spread. Sure, consumers benefit enormously, but the majority of employees will lose. Skill-based technological change puts a huge premium on college degrees, and creates job loss especially in routine blue- and white-collar occupations. Technical change can be relentless. Photo company Kodak, which in its heyday employed nearly 150,000 people, filed for bankruptcy in the same year Instagram, merely employing 15, was sold to Facebook for $1 billion. Upswings in the economy create less and less new jobs, the share of labor in GDP is on decline in recent years, and even within this share, the spoils are distributed more and more unevenly. We experience the rise of what they (and others) call the superstar economy.
The dynamics of the superstar economy can be described in many ways: the Matthew effect (“For unto every one that hath shall be given, […] but from him that hath not shall be taken even that which he hath.”), a winner-takes-it-all society, or, more prosaically, the dominance of power laws. One J.K. Rowling sells a multiple of books of her closest competitors, those competitors sell a multiple of their closest rivals etc. These effects are hard to deny. Even if the authors refute simplistic fallacies, they are remarkably ambivalent about the possibility of what Keynes famously called technological unemployment. Granted, voluntary unemployment might be a society’s ultimate goal, freeing human potential from menial occupations, but this would require a very different social contract from the one we are seeing nowadays.
It is hard to predict in which domains the comparative advantage of humans will survive. Skills complementary to machines, e.g. engineering or data analysis, will probably be in higher demand. ‘Nerd is the new sexy’, as they say (though I suspect that it is, by and large, nerds who say that). To rather race with than against the machines the authors suggest a (laundry) list of short-term policy recommendations: teach children well, use technology, improve matching on the labor market, more infrastructure etc. In the long run, the authors recommend more controversial tools such as basic income schemes, or labeling products with high percentage of human content.
All in all, this is a very easy, sometimes gripping, even alarming read. The authors’ liquid writing style, in combination with a nice batch of anecdotal and systematic evidence attracts huge readership in and beyond academia. In some ways, the book is remarkable, especially given its provenance: M.I.T. It seems that the recent crisis has truly shaken mainstream economics and opened the cracks for heterodox thinking. Talking about technological unemployment seemed to be close to blasphemia only years ago. In this sense, the book is a welcome game changer that allows the public discourse to talk about really important issues of our time. The problem is, many economists tend to be ill equipped for these purposes.
In the same way economists for long have defined away the problem of inequality as none of their business, job scarcity has, by and large, been a temporary or government-related issue. In some sense, this has changed. The new rise of automatization has created a lot of scholarly interest, much of which is referenced in the book: David Autor, Daron Acemoglu, Joseph Stiglitz, Lawrence Summers to name but a few. The diagnosis coming from this research seems sound – as far as I can judge – even if somewhat partial. Most examples come from technology enclaves around M.I.T. and Stanford. It is a bit alarming that many of the aforementioned authors even use the same case studies, which questions the generalizability of these findings. A bit of long-term perspective would also sometimes be consoling: perhaps the largest transformation of labor markets so far was not the industrial, but the agricultural revolution, which essentially made everyone work for endless hours on the fields, only to barely survive on a highly unbalanced diet of mono-crops.
Whereas the diagnostic part of the book is fascinating, the sections about policy recommendations are disappointing. They basically don’t go beyond the litany of ideas found in any econ 101 textbook. Looking at some policy recommendations in more detail fully reveals a kind of helplessness usually only seen in a rabbit shortly before being gobbled up by a snake. When it comes to schooling the authors first criticize the tendency of educational institutions of being too lax and not making students work enough before they praise Montessori schools for letting pupils decide themselves what they are best at.
In more general, the book shows the kind of lop-sided thinking which cares much more about allocation than distribution. However, the main issue has and always will be a problem of distribution. Taking away unemployment benefit systems and job agencies, takes away (legally defined) unemployment, but it does not take away underemployment, inequality and poverty. In this sense, the recommendations found in the book are fairly limited. If social scientists like Karl Polanyi are right, the first industrial revolution generated a new kind of welfare model. Back then higher growth did also not turn automatically into benefits for everyone, and perhaps would have never, if not for better organization and more solidarity among workers. If contemporaries of the industrial revolution had recommended something akin to things Brynjolfsson and McAfee suggest, we would still have Speenhamland (British 18th century poor laws) instead of the modern welfare state. In this sense, the new revolution, assuming that there is one, will only benefit everyone, if accompanied by fundamental changes in the social contract.
Polidigwerkschdaeddle 