The American writer Kurt Vonnegut began his career in the public relations division of General Electric. One day, he saw a new milling machine operated by a punch-card computer outperform the company’s best machinists. This experience inspired his novel Player Piano. It describes a world where children take a test that determines their fate. Those who pass become engineers and design robots used in production while those who fail have no jobs and are supported by the government. Are we converging to this dystopian world? How should public policy respond to the impact of automation on the demand for labour? These questions have been debated ever since 19th century textile workers in the UK smashed the machines that replaced them. As the pace of automation quickens and affects a
Guerreiro, Rebelo, Teles considers the following as important:
This could be interesting, too:
Haidar writes Late-movers outperform first-movers in export markets
Adams Greenup, Koijen, Lichtenfeld, Van Nieuwerburgh writes A win-win for health and life in times of COVID
Buch writes Evaluation of too-big-to-fail reforms
Djankov, Glaeser, Shleifer writes Measuring property rights institutions
The American writer Kurt Vonnegut began his career in the public relations division of General Electric. One day, he saw a new milling machine operated by a punch-card computer outperform the company’s best machinists. This experience inspired his novel Player Piano. It describes a world where children take a test that determines their fate. Those who pass become engineers and design robots used in production while those who fail have no jobs and are supported by the government. Are we converging to this dystopian world? How should public policy respond to the impact of automation on the demand for labour?
These questions have been debated ever since 19th century textile workers in the UK smashed the machines that replaced them. As the pace of automation quickens and affects a wider range of economic activities, Bill Gates reignited this debate by proposing the introduction of a robot tax (Delaney 2017).1 Policies that address the impact of automation on the labour market have been widely discussed – for example, by the European Parliament – and have been implemented, to some extent, in countries like South Korea.
In Guerreiro et al. (2020), we use a model of automation to study whether it is optimal to tax robots. The model has two types of occupations: routine and non-routine. Robots are complements to non-routine workers and substitutes for routine workers. In our baseline model, overlapping generations of workers with heterogeneous costs of skill acquisition choose their skills before entering the labour force. At a given point in time, when optimal policy is being designed, there is an initial generation of workers who can no longer retrain to perform a new occupation.
The cost of producing robots falls over time as a result of technical progress. We choose parameters so that the status quo of the dynamic model is consistent with the time series for the non-routine wage premium and the fraction of the population with routine occupations in the US economy. We show that, under the current tax system, a sustained fall in the cost of automation generates a large rise in income inequality and a substantial fall in the welfare of those who work in routine occupations.
When the government can levy different lump-sum taxes for each type of worker, technical progress is always welfare improving because its gains can be redistributed. But, in practice, discriminatory lump-sum taxes cannot be levied either because of lack of information on worker types or political and cultural constraints on the tax system.
For this reason, we solve for the optimal tax system, while imposing, as in Mirrlees (1971), the constraint that the government does not observe the worker type or the worker’s labour effort. The government observes the workers’ income and taxes it with a nonlinear schedule. In addition, robot purchases are also observed and taxed with a proportional tax.
We consider first a setting in which occupational choices are exogenous. Some workers are born to do non-routine jobs and others to do routine jobs. In this setting, it is optimal to tax robots if the planner wants to redistribute income toward routine workers. Because the tax system is the same for all workers, non-routine workers can also choose the income-consumption bundle of routine workers. This possibility restricts the generosity of the redistribution system. The bundle of the routine worker can be particularly attractive to the non-routine, high-productivity workers because they can earn the same income as routine workers in just a few hours. Taxing robots reduces the non-routine wage premium making the routine-worker bundle less attractive to non-routine workers. As a result, the planner can redistribute more towards routine workers. The optimal robot tax balances these benefits of wage compression with the inefficiency losses from distorting production.
Costinot and Werning (2018) and Thuemmel (2018) show that these arguments hold in more general environments. Costinot and Werning consider a model with a continuum of worker types. They derive optimal tax formulas that depend on a small set of sufficient statistics that require relatively few structural assumptions. Thuemmel considers an economy with three occupations (non-routine cognitive, non-routine manual, and routine workers), heterogeneous productivity within occupations. He also considers occupational choice which plays an important role in our model.
Is it still optimal to tax robots in a model where different generations overlap and skill choices are endogenous? In such a setting, taxing robots reduces incentives for acquisition of non-routine skills. Designing an optimal tax system requires balancing two objectives. First, the planner wants to give the young generations incentives to invest in skills and become non-routine workers. Second, the planner wants to redistribute income toward routine workers, since their wages fall as robots become cheaper. Taxing robots reduces the non-routine wage premium and helps redistribute income toward routine workers.
It is optimal to tax robots to redistribute income from the initial old non-routine workers to routine workers. Old workers chose their skills in the past, so their choice of occupation is not affected by the planner’s generosity. In contrast, the occupations chosen by future generations are affected by the redistribution system. The planner does less redistribution towards future routine workers to give them incentives to acquire non-routine skills. As a result, it is optimal to tax robots in the short run but not in the long run.
In our quantitative exercise, we find that it is optimal to tax robots for three decades. During this period, the labour force includes many older workers that chose their occupations in the past. The optimal robot tax rates implied by the model are modest: 7% in the first decade, 3% in the second decade, and 1% in the third decade. Once the initial generations retire, the optimal robot tax is zero.
The fact that it is optimal to tax robots is a failure of the classical Diamond and Mirrlees (1971) result on the optimality of production efficiency. According to the production efficiency theorem, taxing intermediate goods is not optimal even when the planner has to use distortionary taxes. According to this logic, since robots are an intermediate good, they should not be taxed.
Why does the Diamond and Mirrlees (1971) theorem fail in our setting? The first reason is that it requires the ability to tax net trades of different goods at different rates. In our model, this requirement means that the labour income of different types of workers can be taxed at different rates, even when those workers earn the same income. We do not allow for this kind of tax discrimination. Instead, we require that all worker types face the same nonlinear tax schedule.
The second reason for the failure of the production efficiency theorem is the presence of general equilibrium effects. Robots are substitutes for routine labour and complements to non-routine labour. By taxing robots, the planner can raise the pre-tax relative wage of routine workers through a general equilibrium effect. As a result, it is worthwhile to distort production decisions to improve the planner’s redistribution choices.
The world economy has undergone many structural changes that destroyed some jobs while creating other jobs. Isn’t the advent of robotisation just another one of these changes? Why should public policy intervene this time? What makes this time different is the speed with which automation can occur. Many of the prior structural changes occurred slowly. The older generations kept their jobs and it was their children who had to adapt to the brave new world. However, given the rapid pace of automation today, it can destroy many of the jobs held by the older generations and lead to a dramatic rise in income inequality. Public policy can avoid turning modern economies into the bleak world described in Player Piano.
Acemoglu, D and D Autor (2011), “Skills, tasks and technologies: Implications for employment and earnings”, Handbook of Labor Economics 4:1043–1171.
Cortes, G M, N Jaimovich and H E Siu (2017), “Disappearing routine jobs: Who, how, and why?”, Journal of Monetary Economics 91: 69–87.
Costinot, A and I Werning (2018), “Robots, trade, and luddism: A sufficient statistic approach to optimal technology regulation”, NBER Working Paper 25103.
Delaney, K J (2017), “The robot that takes your job should pay taxes, says Bill Gates,” Quartz, 17 February.
Diamond, P A and J A Mirrlees, (1971), “Optimal taxation and public production: Production efficiency”, The American Economic Review 61(1): 8–27.
Guerreiro, J, S Rebelo and P Teles (2020), “Should Robots Be Taxed?” NBER Working Paper 23806, Sep 2017, revised January 2019.
Mirrlees J A (1971), “An exploration in the theory of optimum income taxation”, The Review of Economic Studies 38(2): 175–208.
Thuemmel U (2018), “Optimal taxation of robots”, Cesifo working paper.
1 See Acemoglu and Autor (2011) and Cortes et al. (2017) for discussions of the impact of automation on the labour market for routine workers.