Democratic regulation of AI in the workplace

When artificial intelligence (AI) displaces lower-skilled workers with higher intensity, electoral democracies may slow down automation in fear of unemployment and voter resentment. Using a Downsian model of elections where parties promise to limit automation and redistribute automation surplus, we show that when automation is highly productive democracies implement maximum automation, making all workers vulnerable to redundancy and distribute the entire surplus among the working population. Majority of the workers are gainers in the sense that their expected earnings exceed their (pre-automation) wage. When the automation surplus is low, democracies restrict automation and protect the high-skilled workers (including the median-skilled worker) but redistribute nothing to the vulnerable workers. Here, because of no compensation for redundancy all vulnerable workers become losers as their expected earnings fall below their basic wage. For highly productive automation, democracies achieve the first best worker welfare but otherwise may over- or under-provide automation. • When automation is electorally determined, its adoption depends on automation surplus and its distribution rule. • If the workers beyond the reach of automation do not get any surplus, fiscal populism accompanies any scale of automation. • If the surplus is shared by all, automation can be high with fiscal populism or modest with regressive redistribution. • In reference to worker welfare, democracy automates too little or too much except when the automation surplus is very large.