A Pareto-improving UBI For Global Automation

We design a pareto-improving universal basic income policy harnessing the first large-scale computable general equilibrium model of endogenous global technological change.

By Victor Ye and Seth Benzell

What is your proposal? We build a generational CGE model featuring 2.6 million agents, 17 regions, and three skill groups. Automation is simulated by allowing firms to adopt new production technologies - calibrated by projecting forward historical trends of technological change in the U.S. - as they are invented. We utilize this model to propose a UBI to equitably redistribute gains from automation. Specifically, a transfer of up to 5.1 percent of U.S. GDP, debt-financed for 20 years and paid for with progressive income taxation thereafter, provides a Pareto welfare improvement to all American workers regardless of age or skill.

Versus no automation at all, global automation benefits the U.S. economy, producing growth and capital deepening at the cost of greater inequality. Our proposed UBI guarantees that low-skill workers of all generations are weakly better off, while still allowing for 3-15% of welfare gain for high-skilled workers compared to a scenario without automation.

What problem does your proposal address? Our UBI policy is designed to address inequality induced by skill-biased technological change. Our proposal differs from other UBI proposals in that it is precisely constructed based on state-of-the-art estimates of the impact of automation on firms, intergenerational redistribution mechanisms, progressivity of existing redistribution mechanisms, and the effects of global capital flows. We also analyze the welfare consequences of UBIs in a general equilibrium framework with different choices of taxes and incidence and with calibrations of deadweight loss. This allows us to quantitatively estimate the scale of transfers required to make each method of financing the UBI Pareto-improving.

How does this policy proposal relate to artificial intelligence? AI plays a critical role in projected trends of automation - as we model it, a shift of production input shares to capital and high-skilled labor - by replacing costly menial labor with a relatively small amount of developer time and cheap processor time. We consider several different automation scenarios, with different assumptions about how quickly which types of AI technologies will advance (e.g. machine vision vs. robotics). By combining the `suitability for machine learning’ scores of Brynjolfsson, Mitchell and Rock (2018) with international statistics on occupation, we can simulate how different scenarios will differentially affect regions, skill-groups and age cohorts.