Robots have long existed in human imagination and only recently in the real world. The baggage of past imagination often intrudes into our understanding of how real robots will affect our economic lives. This explainer rigorously defines ‘robots’ and artificial intelligence; presents empirical evidence and analysis on the impacts of robotisation on production processes and aggregate economic outcomes; and puts forth policy recommendations for ensuring social benefits from robotisation at minimal cost.
Artificial intelligence (AI) is a rapidly evolving field with robots impacting our lives with increasing intensity. The time is therefore ripe to speculate about how robotisation will impact economic development in the world in the next 20 years. This period is so chosen because it is impossible to foresee invention and innovation in AI in the longer run. We ‘speculate’ instead of ‘predict’ because even within these 20 years the timing and extent of innovation will be largely random.
Some excellent models about the economic impact of robotisation exist in the literature (for example, Sachs, Benzell and LaGarda 2015 Mookherjee and Ray, 2017),1 which make specific assumptions to gain deep insights. In our research we refrain from model-building and instead try to uncover all major mechanisms of impact. We hope that the very general and broad framework developed by us will serve as an input into further rigorous and precise empirical and theoretical model-building.
Robots and AI: Concepts
AI endows machines with the power to replicate both human faculties, such as calculation and cognition, as well as non-human ones, say satellite imaging (Charniak and McDermott 2009, Simon 1995). AI can vary from ‘very narrow’ to “fairly broad” and “very broad”, exemplified respectively by a calculator, a computer capable of running several software programmes simultaneously, and a network of computers replacing a team of consultants devoted to project implementation. Over time, there has been a tendency for AI to become broader.
Robots can be considered as receptacles in which AI is stored. These, according to the International Federation of Robotics (IFR), are machines that are “automatically controlled, reprogrammable and multipurpose” (Acemoglu and Restrepo 2017). Unlike humans, robots are capable of working in 24-hour shifts and can be reprogrammed easily to implement changes in manufacturing processes. Contrary to popular belief, robot programming need not be done by humans: by just watching humans, software residing in a robot can note the sequenced steps needed to perform jobs and then replicate the sequence. This is known as ‘machine learning’. But most importantly it seems that the cost of training robots is a fraction of the cost of training humans. For example, consider the Baxter robot manufactured by Boston-based ‘Rethink Robotics’. A manufacturer can train a single Baxter robot by conditioning its limb movements to do a variety of jobs. This knowledge is then propagated to other robots, numbering hundreds, using a USB device2 (Ford 2015).
Machine learning gets better as more and more data on humans handling jobs emerges, thus providing clear directions for handling contingencies. However, machine-learning algorithms are clearly difficult to construct for replacing humans whose jobs involve dealing with a large number of contingencies, especially ones which are difficult to anticipate in advance and therefore require soft or fuzzy skills. This is a major reason for recently developed humanoids not being able to emote, collaborate, and interact socially as well as humans (Stephane et al. 2016), a weakness which helps us understand the empirical predictions of Deming (2017) and Hartley (2017): jobs requiring soft skills will probably continue to grow at a rapid rate whereas those based on technical skills will contract.
The impact of AI on production processes
Mechanisation of agriculture in developed countries is being further enhanced through use of robots, imbued with judgement about colour and texture through a combination of machine vision technology and algorithms, to harvest ripe fruits and flowers and prune vines (Vision Robotics Corporate Website, Ford 2015).
In regard to currently labour-intensive agriculture in developing countries, policy changes seem to indicate that job-displacing automation and even robotisation might be on the anvil: for example, in India, the Model Agricultural Land leasing Act, 2016 encourages the consolidation of hitherto small agricultural holdings, and official legislation such as the Agricultural Produce Market Committee Act, 2003 has given a push to contract farming (Mani 2016, Singh 2015).
In manufacturing, industrial robots – some with three-dimensional vision and others with precisely timed movements – have come to the fore. Projections of past trends suggest an increase in efficiency over time such that these will surpass humans in shifting, packing, and assembling objects. Hitherto un-robotised manufacturing activities, such as production of cars, associated with the constant interaction of men with machines, might soon be robotised. This is because of the development of the Robot Operating System which helps in generation of software for powering robots to do different tasks and is therefore analogous to Microsoft Windows/Google’s Android in regard to computers/mobile phones (Ford 2015). Second, Moore’s law (Ford 2015, Schaller 1997), which refers to statistically observed and mainly software-powered annual doubling of computing power, and economies of scale in production of industrial robots, might imply that conventional production will be fast replaced by robotic production in hitherto untouched sectors.
In the formal service sector, robots are already a significant phenomenon: these prepare and serve meals and manage cash counters (Tabuchi 2010); vend products ranging from DVDs to cars (Semuels 2011, Ford 2015); distribute medicines in pharmacies and hospital rooms (King 2010); and replace middle-level secretarial and managerial skills (see Autor 2010, Jaimovich and Siu 2012) as also those of highly paid consultants. In addition, high-quality mass online courses offered by reputed academics are fast becoming popular (Ford 2015, Salmon 2012, Selingo, 2013, Chafkin 2013) because of access provided to large numbers in distant locations. Logistical difficulties in regard to testing, facilitation of interactive learning, and reluctance to subscribe to business models involving large student-teacher ratios (Ford 2015) are cramping the expansion of mass online education; with time, these problems might be overcome.
Robotic activity is picking up in the following sectors though humans still have a massive lead over robots which seems unlikely to be erased in the next 20 years: translation, writing of novels, news reporting, art, and composition and rendition of music (see Hicks 2018, Schaub 2016, Carr 2009, Moses 2017, Ford 2015, Shubber 2013, Smith 2013). Given the mentioned limitations of humanoids, humans would still be required in the hospitality sector, and as versatile caregivers for the elderly.
Empirical studies on the macroeconomic impact of robotisation
A labour market study of the US and Western Europe by Acemoglu and Restrepo (2017) that uses IFR data, points to a 10.4% annual growth of industrial robots in 1993-2007. This study establishes that the introduction of a new robot per 1,000 workers in a commuting zone reduces the local employment-to-population ratio by 0.16 percentage points and local wages by 0.25%. The estimated negative impact on wages would have been higher if the analysis had accounted for purely cerebral human skills displaced by AI. Moreover, job loss will be magnified by future increase in efficiency of robotic software and hardware.
In regard to developing countries, the ILO (International Labour Organization) has extensively researched the impact of automation and AI on employment in the ASEAN region. This project fed into a noteworthy paper (Chang and Huynh 2016), which applies research methodology developed by Frey and Osborne (2013). Labour-intensive sectors in the ASEAN (Association of Southeast Asian Nations) might become a victim to automation and the replacement of outsourced work by robotic operations in developed countries such as the US, where it is now possible for a single robot to turn fabric into garment (see the case study of the ‘Sewbot’ as discussed in Device Plus (2018); Zhou and Yuan 2017). The study concludes that 56% of all employment in the ASEAN-53 is at high risk of displacement over the next decade or two. Across ASEAN-5 countries, industries that have been identified as having high potential for automation are hotels and restaurants; wholesale and retail trade; and construction and manufacturing.
Much more alarming are the developments in China. The motive of ramping up production of various commodities in the face of a rise in human wages caused by increasing consumption demand has led to substitution of human with robotic labour. By 2014, Chinese factories accounted for about 25% of the world’s industrial robots, buoyed by a 54% increase in the number of industrial robots over 2013-14. Displacement of blue-collar workers is not being adequately neutralised by the creation of white-collar opportunities. In mid-2013, Chinese government statistics revealed that half of the country’s current crop of college graduates remained unemployed.
Trends in employment growth in India do not present a rosy picture either, as shown by Abraham (2017): NSS (National Sample Survey) data reveal that India’s average annual employment growth rates have shown a steady decline from 2 to 0.4% over three consecutive five-year periods starting 1999-2000. Three different but reliable surveys reveal negative growth in the period 2013-14 to 2015-16. With no comparable slowing down of growth of GDP (gross domestic product) over this period, a marked substitution of labour by capital is the probable cause of such negative growth of employment. There is also enough evidence to show that computerised automation is not new to India and evidence of its use is available in research done 20 years back (Narain and Yadav 1997). Large-scale robotisation is logically the next step.
Finally Rodrik (2016) and World Bank (2016) document that Africa has evidently embarked on a path of expansion of the non-agricultural sector through robotisation. Clearly it seems that barring major changes in the demand scenario, robotisation would pose a potent threat to human jobs.
Robotisation’s imprint on the future of economic development: Analysis and policy recommendations
As robots become more efficient capital would move from the traditional sector to the robotised sector because of higher returns. This will result in labour in the traditional sector becoming less productive. However, employers will most probably not reduce money wages to retain all labour hitherto employed by them. Instead, downward money wage rigidity, as assumed by Keynes (1936), and explained by others as resulting from the need to keep the morale of employed workers high would be associated with laying off of workers (see explanations provided by Slichter (1920, 1929), Solow (1979) and Akerlof (1982); and field evidence by Kube, Maréchal and Puppe (2010).
The bidirectional causality between recession in consumption demand and unemployment would however be alleviated by an AI-facilitated increase in time available for consumption by entrepreneurs, managers, and workers. There is a very strong possibility that the recent reduction in length (in hours) of the work week (Konnikova 2014) would continue. A five-day, 25-hour work week for workers in Europe by 2030 is a distinct possibility as robotisation provides greater room to worker welfare-oriented legislators, norm setting institutions, and efficiency-oriented corporate management.
Hitherto under-consumed goods and services, a few examples being tourism, paid companionship, wellness services, and myriad forms of entertainment, should see a spurt in consumption. Many of these are associated with time-intensive consumption and human capital-intensive production. The employment thus generated will further increase aggregate demand, thus counteracting the initial tendency for contraction of employment and national income. An expansion of employment and economic activity in the next 20 years is thus not only possible but very probable.
At the same time, significant pain will surely be caused to human workers moving from robotising sectors to other sectors through associated search unemployment; it is also possible that effective loss of human capital will be associated with a loss in individual incomes. Moreover, the highly variable and random pace of innovation points to the possibility of a bulge in search unemployment of those seeking new jobs in a scenario of job destruction accompanying creation. This might give rise to significant and dangerous recessionary tendencies requiring prompt government action.
An obvious antidote is a national tax on robotised production to fund basic incomes for the unemployed. However, such a tax can lead to capital flight and would do nothing to solve the problem of cheaper goods produced by foreign robots destroying markets for domestically produced goods and thus, domestic jobs, or of foreign insourcing replacing outsourcing to developing countries.
A global fund financed by a uniform global robot tax for providing basic incomes and developing the mentioned soft and fuzzy skills would be better. It is very likely that governments of affluent countries, the hubs of robotic activity, would insist on basic incomes that are indexed to national per capita incomes. Countries such as India, bound to be affected by robotisation and with a track record of savvy negotiating at international forums, might have to take the lead in parleys. A basic income scheme would also stimulate consumption demand and help to revive both incomes and employment, assuming that those taxed would have a lower propensity to consume than those receiving the proceeds of taxes.
A dilemma may confront policymakers as to whether basic incomes should be provided to all global citizens or just those rendered unemployed by robotisation. The latter option might result in the unemployed developing inertia and not looking out for jobs. There could also be problems in correctly identifying the unemployed or of the self-employed feigning unemployment. Provision of a low level of basic income which is enough for subsistence but not for very comfortable living would provide motivation as well as security to those unemployed to look for jobs. Often such security is associated positively with initiative, mental health, and entrepreneurship. Moreover, ‘identification’ will cease to be a problem.
This explainer is based on an academic paper titled ‘ Thorny Roses: A Peep into the Robotised Economic Future’ by the same authors.
- The first paper uses a set up in which each good can be potentially produced by ‘robots only’ or through the complementarity of machines and humans. Technological progress can result in the robotic mode of production becoming more popular, thereby reducing wages under conditions of full employment. The second paper discusses the possible achievement of singularity, a stage of economic evolution, in the long run in which everything – including the production of robots themselves – will be robotised, thus rendering human capital redundant.
- USB (Universal Serial Bus) is the most popular connection used to connect a computer to devices such as digital cameras, printers, scanners, and external hard drives.
- Indonesia, Malaysia, Philippines, Thailand, and Vietnam.
- Abraham, Vinoj (2017), “Stagnant employment growth: Last 3 years have been the worst”, Economic and Political Weekly, Vol. LII., Issue. 38, pp. 13-17.
- Acemoglu, D and P Restrepo (2017a), ‘Robots and Jobs: Evidence from US Labor Markets’, NBER (National Bureau of Economic Research) Working Paper No. 23285.
- Akerlof, George A (1982), “Labor contracts as partial gift exchange”, The Quarterly Journal of Economics, Vol. 97, Issue 4, pp. 543-569.
- Autor, D (2010), ‘The polarization of job opportunities in the U.S. labor market: implications for employment and earnings’, a paper jointly released by The Center for American Progress and The Hamilton Project.
- Carr, D (2009), ‘The robots are coming! Oh, they’re here’, The New York Times, 19 October 2009.
- Chafkin, M (December 2013/January 2014), ‘Udacity’s Sebastian Thrun, Godfather Of free online education, changes course’, Fast Company.
- Chang, Jae-Hee and P Huynh (2016), ‘ASEAN in transformation: The future of jobs at risk of automation’, Bureau for Employers’ Activities, Working Paper No 9, International Labour Office.
- Charniak, E and D McDermott (2009), Introduction to artificial intelligence, Fourth Impression, Pearson Education.
- Deming, David J (2017), “The growing importance of social skills in the
labormarket”, The Quarterly Journal of Economics, Vol. 132, Issue 4, pp. 1593-1640.
- Device Plus (2018), ‘SewBot is revolutionizing the clothing manufacturing industry’, Device plus, 19 February 2018.
- Ford, M (2015), Rise of the robots: Technology and the threat of a jobless future, Basic Books, New York.
- Ford, M (2015), ‘China's troubling robot revolution’, The New York Times, 10 June 2015.
- Frey, C and M Osborne (2013), ‘The future of employment: How susceptible are jobs to computerisation?’, University of Oxford.
- Hartley, S (2017), The Fuzzy and the Techie: Why the Liberal Arts Will Rule the Digital World, Houghton Miffin Harcourt Publishing.
- Hicks, M (2018), ‘Microsoft’s new AI translates Chinese-to-English as well as a human translator’, 14 March 2018.
- Jaimovich, N and HE Siu (2012), ‘The trend is the cycle: Job polarization and jobless recoveries’, NBER (National Bureau of Economic Research) Working Paper No. 18334.
- Keynes, JM, (1936), The general theory of employment, interest and money, Chapter 2, Macmillan, London.
- King, R (2010), ‘Soon, that nearby worker might be a robot’, Bloomberg Business Week, 2 June 2010.
- Konnikova, M (2014), ‘Why Not a Three-Day Week?’, The New Yorker, 5 August 2014.
- Kube, S, MM André and C Puppe (2010), ‘Do wage cuts damage work morale? Evidence from a natural field experiment’, Working Paper No. 471, Working Paper Series ISSN 1424-0459, Institute for Empirical Research in Economics, University of Zurich.
- Mani, Gyanendra (2016), “Model Agricultural Land Leasing Act, 2016: Some Observations”, Economic and Political Weekly, Vol. 51, Issue No. 42.
- Mitra, S and M Das (2018), ‘Thorny Roses: A Peep into the Robotised Economic Future’, 28 May 2018. Available at SSRN.
- Mookherjee, D and D Ray, (2017), ‘Capital and Inequality in the Long Run: Automation without Technical Progress’, Mimeo. Available here.
- Moses, L (2017), ‘The Washington Post’s robot reporter has published 850 articles in the past year’, Digiday, 14 September 2017.
- Narain, Rakesh and C. Yadav (1997), “Impact of computerized automation on Indian manufacturing industries”, Technological Forecasting and Social Change, Vol. 55, Issue 1. pp. 83-98.
- Raffard, Stephane, Catherine Bortolonab, Mahdi Khoramshahic, Robin N Salessed, Marianna Burca, Ludovic Marin, Benoit G Bardy, Aude Billard, Valérie Macioce and Delphine Capdevielle (2016), “Humanoid robots versus humans: How is emotional valence of facial expressions recognized by individuals with schizophrenia?”, Schizophrenia Research, 176(2), pp. 506-513.
- Rodrik, Dani (2016), “Premature deindustrialization”, Journal of Economic Growth, Vol. 21, Issue 1, pp. 1–33.
- Sachs, JD, SG Benzell and G LaGarda (2015), ‘Robots: curse or blessing? a basic framework’, NBER Working Paper 21091.
- Sachs, JD, SG Benzell and G LaGarda (2015), ‘Robots: curse or blessing? a basic framework’, NBER Working Paper 21091.
- Salmon, F (2012), ‘Udacity and the future of online universities’, Reuters blog, 23 January 2012.
- Schaller, Robert R (1997), “Moore’s law: past, present and future”, IEEE Spectrum, Vol. 34, Issue 6, pp. 52-59.
- Schaub, M (2016), ‘Is the future award-winning novelist a writing robot?’, Los Angeles Times, 22 March 2016.
- Selingo, JJ (2013), College Un-bound: The Future of Higher Education and what it means for Students, New Harvest, New York.
- Semuels, A (2011), ‘Retail jobs are disappearing as shoppers adjust to self-service’, Los Angeles Times, 4 March 2011.
- Shubber, K (2013), ‘Artificial artists: when computers become creative’, Wired, 7 August 2013.
- Simon, Herbert A (1995), “Artificial intelligence: An empirical science”, Artificial Intelligence, Vol. 77, Issue 1, pp. 95-127.
- Singh, S (2015), ‘APMCs: the other side of the story’, Business Line, 8 February 2015.
- Slichter, Sumner H (1920), “Industrial morale”, The Quarterly Journal of Economics, Vol. 35, Issue 1, pp. 36-60.
- Slichter, Sumner H (1929), “The current
laborpolicies of American industries”, The Quarterly Journal of Economics, Vol. 43, pp. 393-435.
- Smith, S (2013), ‘Iamus: Is this the 21st century's answer to Mozart?’, BBC News, 3 January 2013.
- Solow, Robert M (1956), “A contribution to the theory of economic growth”, The Quarterly Journal of Economics, Vol. 70, No. 1, pp. 65-94. Available here.
- Solow, Robert M (1979), “Another possible source of wage stickiness”, Journal of Macroeconomics, Vol. 1, pp. 79-82.
- Tabuchi, H (2010), ‘For sushi chain, conveyor belts carry profit’, New York Times, 30 December 2010.
- World Bank (2016), ‘World Development Report 2016: Digital Dividends’, Washington DC.
- Zhou, M and Z Yuan (2017), ‘Textile companies go high tech in Arkansas’, China Daily, 25 July 2017.