AI Policy Exchange
Let's decide the future of AI, together.
The Luddite fallacy: Is it time to reconsider it?
by Vihang Jumle, Project Officer, TRAFFIC India, World Wildlife Fund
20 December, 1922 marked an important transition in the history of employment. As the winter mist settled on the dawn-lit streets of Brooklyn Heights in New York, the city and its many residents stood wide awake to witness the dramatic and historic end of natural fire engines. As the clamour of hooves and neighs of Engine Company 205’s horses broke the morning silence and marched to the ultimate ceremony at Brooklyn Borough Hall, the world embraced the motorised car and all its possibilities. With a long stint of over 50 years, horses, which until that day were used to pull the fire engines, lost their jobs to technology. Unfortunately, so did numerous horsemen, stable workers, and trainers.
Similarly, the 1990s witnessed bank tellers lose their jobs to ATMs. During different periods of history, factory workers, farmworkers, and craftsmen too, met with the same fate. Technology, it was feared during all these periods, was going to automate all the jobs. Possibilities of such havoc exist today too. Oxford University researchers, Carl Benedikt Frey and Michael Osborne, in 2013 concluded that 47 percent of 702 different American jobs could be automated in the next two decades. A similar study was conducted by the OECD, taking a different approach than the Oxford researchers, which also spoke about high rates of job automation.
However, on the contrary, the world has only added more jobs over time with most displaced jobs circling back into the system. Do people still have reasons to be worried about the adverse effects of rising automation?
Technological unemployment is not unprecedented. It is perhaps as old as the invention of the wheel. Aristotle in 350 B.C.E spoke about how machines would one day eliminate the need for human labour. The eighteenth-century population, amidst mass unemployment, was rather pessimistic about machines taking over jobs. Keynes in the 1930s coined the term “technological unemployment” to denote job losses caused due to automation and argued that the short phase of unemployment was merely a “temporary phase of adjustment” before the jobs circle back into the economy.
It is accepted wisdom that any technological innovation causes only short-term unemployment, with negative implications being minimal in the long-term. This wisdom exposes the “Luddite fallacy” — a term used to point out that technological innovation causes no long-term negative effect on employment. Economists like Jacob Mincer and Stephan Danninger, have relied on the “compensation theory” to justify the minimal to none negative effects of technological innovation on employment in the long-term.
Here’s how it works: automation leads to an increase in productivity since machinery produces goods at a lower price and at a faster pace. Since a product’s demand generally tends to rise only little over the short term, workers are laid off as machines take up their share of work. Decreased product prices allow consumers to spend their surplus money in different markets, creating newer demands elsewhere. Similarly, the costs saved by business owners by laying off workers allows them to reinvest money into new ventures, creating its own new demand for labour. This in the long run compensates for the temporary loss in jobs since labourers tend to get re-employed to meet these newly created demands.
For instance, after robotic arms replace labourers in assembly lines, newer demand for labourers prop up in industries that create robotic arms, hence re-generating the lost jobs. This reasoning became the foundation of “compensation theory” and other frameworks designed by classical economists who argued that construction, maintenance, and development of machines required to do the automated work creates its own demand for labour. Traditionally this did hold true.
However, recent spikes in the pace of automation have raised red flags over the validity of compensatory mechanisms. It is less known if the compensation effect can make up for the piling structural unemployment caused due to rapid and disruptive automation.
Coal mine workers, especially those engaged in drilling, digging, etcetera, were amongst the first to lose their jobs to automation brought in by heavy machines. The compensation effect enabled laid-off workers to find new jobs in manufacturing industries and factories after a short phase of unemployment. It is however unclear if all lost jobs will manage to circle back into the economy due to rapid automation going forward.
Workers may be laid off more often since jobs are likely to be automated much faster than before. This implies that unemployed workers will need to develop skills for new jobs frequently and quickly. And assuming not all workers will be able to do so, the pool of unemployed workers will continue to grow larger. This situation appears more pessimistic if we account for rising levels of job complexity. Coal mine workers could easily adjust to assembly line jobs due to the similarity in functions: a combination of human strength and following a set instructions.
However, such repetitive jobs in the future will remain scarce because they may have already been automated leaving only high-skill specialized jobs as available options for low-skilled unemployed workers. This may compel them to upgrade their skills in short spans. Some high-skill level jobs that demand cognitive power, such as technicians, nurses, etcetera also require work certifications which may add more hurdles. Not to forget that available high-skill jobs created as compensation too stand the risk of being automated before low-skilled unemployed workers manage to upgrade their skills.
Much like other public policy issues, this problem too tends to have a larger impact on less privileged workers. It is so because repetitive jobs — that follow a set of instruction — stand a high risk of automation. These generally include middle-class blue-collar jobs that employ assembly line workers, machine operators, etcetera. Categories on either ends of the skill spectrum (low skill jobs like house cleaning and security and high skill jobs like in engineering and healthcare) have in fact gained more jobs whereas the middle-class blue collar jobs have shrunk due to rapid automation over the last couple decades.
From the 1980s to 2010s, the global share of workers in middle-income class has shrunk from 65.9 percent to 58.4 percent, whereas the share of workers in the lower-income class has risen from 26.1 percent to 33.5 percent in the same period. This trend is likely to continue due to the nature of blue-collar jobs, and perhaps, governments could play their role by slowing down automation in this sector and focus more on skill development.
Similar policy level moves are a separate discussion altogether; the objective here is to highlight that compensation theories likely need to be reworked to account for the recent high pace of automation. Perhaps we need to acknowledge that people do have valid reasons to worry about their jobs going permanently if conversations to regulate automation do not gain momentum.
Views expressed above belong to the author(s).
© 2019-23 AI Policy Exchange
AI Policy Exchange is a non-profit registered under section 8 of Indian Companies Act, 2013.
We are headquartered in New Delhi, India.
Founded in 2019 by Raj Shekhar and Antara Vats