The Job Terminator? Automation, Artificial Intelligence and the Future of Work

The upcoming G-20 summit in Argentina prioritizes the future of work, searching for ways to meet the challenges of a digitized economy. Francisco Rivera-Batiz analyzes the economic and policy implications of increasing automation on the labor market.

Francisco L. Rivera-Batiz
November 30, 2018

One of the priorities of the upcoming G-20 meeting in Buenos Aires is to discuss how automation and artificial intelligence (AI) will affect the future of work. Fears of job destruction are running rampant. Bestsellers like Martin Ford’s The Rise of the Robots have warned about it. Academics have provided some dismal forecasts. Carl Frey and Michael Osborne estimate in a 2013 paper that close to half of all US jobs are highly susceptible to computerization. A McKinsey Global Institute 2017 report concluded that by 2030, 400 million workers worldwide could potentially be displaced, with up to 30% of the global workforce automated.

But cataclysmic fears of the replacement of humans by machines have often been exaggerated. From the automation of silk looms and windmills in the 18th century to the factories and assembly plants of the 20th century, the specter of machines taking over human employment have persisted. Kurt Vonnegut’s 1952 novel, Piano Player, tells the story of a divided future society where mechanization has taken over human employment. In 1958, The Nation warned about “automation depression” caused by “worker-displacing innovations.” More recently in 2011, Clayton Christensen, a professor at Harvard Business School, predicted that as many as half of American universities would close or go bankrupt within 10 to 15 years due to disruptions of online learning on higher education business models. This has not happened yet.    

Although the AI revolution will require significant adjustments in the world economy that cannot be ignored, warnings about the human workforce's demise are vastly exaggerated. First, existing studies make forecasts that have huge standard errors. For instance, the study by Frey and Christensen noted earlier has been recently reanalyzed in a 2018 OECD report using a more disaggregated and detailed analysis of how susceptible various occupational classifications are to automation. They find that about 14 percent of jobs, instead of 50 percent, risk replacement by new technologies. Second, there is a large variety of technological changes’ possible effects, some of which can destroy jobs but can also create them, and their relative weight is difficult to assess.

The main negative effect of greater automation and AI-induced technical change is the substitution of capital for tasks previously performed by humans. This is already happening. Some of the tasks taken over by automation are routine tasks – tasks that follow a set of rules that can be codified mechanically or through software, and that machines or robots can perform faster and more efficiently than humans. These include manual tasks that involve repetitive or standardized production, like putting together auto parts in an assembly plant or vacuuming the floor. But recently, the AI revolution has allowed automation of activities that use significant deductive skills, such as bookkeeping, managing cash registers, automobile driving, and even cancer diagnosis. 

On the other hand, there are a wide range of tasks that are much more difficult to automate. These are non-routine tasks that require abstract, problem-solving, and/or creative thinking skills. These activities are typical of professional, technical, and managerial occupations – including software engineers who program robots – and are usually filled by highly educated workers. A second set of tasks that are non-routine and therefore difficult to computerize and automate are those that require human understanding, adaptability, judgement, and interpersonal interactions. These can include quite a variety of jobs in the service sector, including maintenance/janitorial jobs and food preparation. While many of these jobs tend to be manual, they do require a myriad of human understanding and nuances that present substantial challenges for automation. Although machine vision and translation, speech and facial recognition, and machine learning in general can over time chip away at some of these jobs, it is unlikely they will do so in the near future because they would have to be standardized to an extent that consumers would deem unacceptable. Since humans have a comparative advantage in fulfilling these tasks, it is unlikely that it would be relatively cost-effective to replace them.

Even when innovation substitutes some tasks that humans currently perform, it creates tasks that did not exist before and are complementary to the new technologies.  Computers and the information technology revolution eliminated the need for secretarial typists, but generated increased demand for administrative assistants who utilize computers and the internet to manage the increasing variety of tasks carried out by global firms.

Automation and AI also create jobs by making the economy more productive as well as making goods and services less costly. This generates a substantial rise of real income that spills over into increased demand for goods and services that spurs employment growth. Economists David Autor and Anna Salomons have examined the effects of automation using data for the last four decades and find that displaced jobs have been compensated by job creation due to greater production in the automated industries, higher labor demand generated by income growth, and the induced demand for inputs and raw materials used by the automated industries. Of course, the problem remains: as automated sectors of the economy shed workers while other sectors grow, what happens to the displaced employees? Policies that assist workers in the transition when jobs are lost due to automation, whether occupationally or geographically, are rarely implemented in any significant way. They may need to in the future.

In the past, major technological revolutions have been accompanied by booming economies. Economist Robert Gordon has documented the increases in productivity growth and real income that have been associated with innovation breakthroughs, including electric lighting, home appliances, motor vehicles, air travel, and IT revolutions in his 2016 book The Rise and Fall of American Growth. Many expect the automation and AI breakthroughs to do the same. But Professor Gordon is skeptical about the past benefits continuing in the future.

Despite the likely long-run positive effects of a new innovation revolution, there are significant short and medium-term challenges in the decades ahead. These challenges are magnified by recent trends that may derail the encompassing benefits past innovation waves have produced unless significant policy actions are implemented.

First of all, the countries that can innovate, produce, and export the new technologies first will have not only first mover advantages but also potential protections provided by patents and intellectual property. The programming, design, and production activities required to build a new generation of AI-connected products will provide a boom for economies that have a comparative advantage in them. That raises serious problems for countries that do not have the levels and quality of education or the expertise required to design or produce new, highly-profitable goods and services associated with the automation revolution. As has been the case in recent decades, global inequality is likely to continue rising, to greater extremes.

Secondly, the tasks that will be more difficult to substitute for automation and AI tend to be in occupations that require either greater cognitive skills, such as creative thinking skills, or otherwise significant non-cognitive skills, such as interpersonal and communication skills. Both skills are nurtured by schools and educational institutions. As labor economists have documented, the technological changes in the last 30 years have produced an increase in the demand for highly-skilled jobs, but have reduced demand for the less-skilled, particularly for those with middle-level skills (as noted before, some manual routine jobs are difficult to automate and, therefore, may not have been affected as much by job losses). But employers surveyed in high-income countries by ManpowerGroup and in developing countries by the World Bank report severe shortages in sectors requiring high-level cognitive and non-cognitive skills. These shortages are likely to grow over time. As a result, a skills mismatch that could severely limit the gains from the automation and AI revolution may emerge, as economists Daron Acemoglu and Laura Restrepo have pointed out.

For workers to remain competitive in the new labor market, the level and quality of schooling would need to rise in tandem with the new innovations. But rising inequalities in income and in education both within-countries and between-countries stand in the way of this happening unless major policy changes are implemented worldwide that ensure the access of the poor and the working class to high-quality schooling (see the 2018 World Development Report for a discussion). 

Ultimately, automation and AI can generate great improvements in the global standard of living. But for this to be realized, serious challenges, especially in the education area and in dealing with the rise of inequality, need to be resolved in the short and medium-term. Without major policy actions now, potential benefits of new technologies may be swamped by negative socioeconomic effects.

Francisco L. Rivera-Batiz is a Professor of Economics and Education Emeritus at Teachers College, Columbia University and a Professor of International and Public Affairs at the School of International and Public Affairs at Columbia. He holds a Ph.D. in Economics from MIT. He has written on international and development economics, as well as on education and its impact on labor markets. He has served as a consultant or technical advisor to the OECD, the United Nations, the World Bank, and a variety of governmental and non-governmental organizations in the US and in developing countries.