The Case For a Robot Tax

Did we miss the elephant in the room at the recent AI summit in Delhi? 

If indeed, as the AI gurus who spoke at the summit suggested, Large Language Models and advanced robotics are poised to snatch away jobs from every conceivable sphere of the economy, should developing economies, especially those that run labour surpluses, use policy to prevent this to some degree?

If the objective of the Delhi meet was to produce a list of concrete imperatives for the Global South, perhaps the deliberations should have brought, front and centre, a discussion on both the possibility and desirability of using economic incentives and deterrents to allow a certain kind of AI and block others. A simplistic example would be the viability of the so-called “Robot Tax” that deters excessive automation when labour is abundantly available, and productivity gains from automating are limited.

In designing an appropriate policy for India and other developing economies, the first step is to move beyond viewing AI policy largely as a question of data privacy, security, and the democratisation of AI resources. It has to be seen as a broader economic challenge that involves key macroeconomic variables like employment and the composition of growth.

Productivity growth, which AI promises in plenty, is welcome. That said, economists and policymakers must begin to fret over the risk of “shadow GDP growth” driven by spectacular productivity gains but with a sharp fall in the workforce. This is particularly compelling for low-middle-income economies like India, where resources for strengthening social safety or retraining displaced workers are limited.

Economists have historically been wary of machines despite their ability to improve human productivity. John Keynes, writing in the 1930s, groaned, “We are being afflicted with a new disease of which some readers may not have heard the name, but of which they will hear a great deal in the years to come—namely, technological unemployment” His sentiment was echoed through the decades by economics heavyweights such as Wassily Leontief and Robert Heilbroner.

Their pessimism was misplaced as economic expansion driven by technological progress ensured that jobs that became obsolete were countervailed by new jobs, and the emergence of new sectors led to a large expansion in net job creation over the last ninety-odd years.

However, Nobel Laureate Acemoglu and his co-author Restrepo’s work on robots and employment in the US between 1990 and 2007 (Robots and Jobs: Evidence From US Labor Markets, NBER 2017) saw a significant adverse impact of automation on both employment and wages after controlling for factors such as the offshoring of jobs. They also found that the creation of new tasks and occupations was negligible. The pace of this net labour displacement is likely to have picked up sharply after this period as the pace of automation picked up.

So-So Innovation
Acemoglu and Restrepo describe this phenomenon as “so-so innovation”, a net gobbler of jobs without much second-round effects of creating new roles for workers. This is often automation for its own sake; the use of new technology is simple because it’s there. Economists might argue that low wages in a labour surplus economy should tilt the balance against unnecessary automation. However, it is possible that AI companies might initially price their products below marginal cost to gain scale in large markets like India. The wage advantage logic might not quite work.

Ideally, economic policy should aim to prevent the proliferation of “so-so innovation” and promote adoption where AI effectively becomes a platform that spawns a wide range of new functions. Public administration is an example of a platform where AI can release human resources from drudge work and enable them to be used for better implementation and monitoring. This could create a whole new portfolio of tasks and functions that not only absorb the displaced workers but also create a demand for new work. In contrast, the larger economic and social benefit of automating a production line in a textile factory is dubious.

This calls for a new kind of economics. Economic theory, as it exists today, sees technological change as “capital augmenting”, improving the efficiency of machines but not changing the fundamental relationship between labour and capital. Acemoglu points out that this misses out the unique feature of technologies such as AI – that it changes who performs which tasks. Simply put, automation increases the number of tasks that capital can produce and diminishes the number for labour.

To understand the impact of AI and related technologies, economics needs to focus on tasks generated and lost through the adoption of technology in its analytical apparatus. It cannot confine itself (as it does now) to aggregate and firm-level output. The desirability of the adoption of a particular AI application hinges on the following question: Does it merely displace labour, or does it have the potential to create a countervailing slew of new tasks?

India’s economic history is marred by the adoption of technology that goes against its comparative advantage, which is its stock of labour. Instead, wrong policy choices influenced by factors such as the Nehruvian fetish for big capital-heavy industries such as steel and the pressure from trade unions to coddle the relatively small pool of formal workers at the cost of the informal, perversely gave capital intensity the upper hand. The emergence of IT and later electronics did its bit to correct that. AI threatens to reverse that and destroy the demographic dividend

Those who believe that technological diffusion is inevitable and the thought of leaning against AI’s momentum is an antiquated Luddite fantasy need to be reminded of the early 1990s, when free trade and globalisation seemed like the only game in town. It took a couple of decades and a global crisis to see its dark side. The push back against it has been severe. That leaves us with a simple question: should we wait for the potential social and economic costs of AI to reach a critical point to consider limiting the damage? Or should India and the Global South be more prescient and pre-emptive?