The core idea of the paper is simple but also overlooked – workers are not merely employees; they are also consumers. In simple words, when companies replace workers with AI, the workers lose income. If they are unable to find new jobs immediately or generate alternative means of income, their spending will drop. As consumer spending drives demand for goods and services, widespread layoffs could likely reduce overall demand in the economy.

According to the researchers, this creates a feedback loop. As demand weakens, companies will see lower revenues. In extreme cases, they could automate so aggressively that they undermine the very market they depend on. This is what the authors describe as “automating their way to boundless productivity and zero demand”.

The surprising aspect of the study is that even though companies understand this risk, they are still likely to over-automate. This phenomenon is described as ‘externality’ by economists. Put simply, when companies replace workers with AI, they get full benefits of cost savings. However, the negative impact, which is reduced consumer spending, is shared across all firms in the market.

Even amid intense competition, each company only bears a mere fraction of the demand loss it creates. As a result, it continues to automate even if, collectively, all firms would be better off slowing down. According to the study, this dynamic leads to what is effectively an automation arms race where companies replace more workers than is economically viable.

The researchers use a model where companies decide how much of their workforce to replace with AI. They find that firms tend to automate more than what would be best for everyone overall. In simple terms, even though each company is making logical decisions for itself, together they end up hurting the economy and their own profits. Most importantly, these losses are not limited to workers. While employees suffer from lost wages, companies also experience lower profits owing to declining demand. This could likely lead to a situation where its value is diminished rather than redistributed between groups.

The paper also assesses several commonly proposed policy responses. Measures such as universal basic income (UBI), capital income taxes, and worker profit-sharing can help cushion the effects of job losses; however, they do not address the core problem. Even though these policies may improve income levels or redistribute wealth, however, they do not change the underlying incentive for firms to automate.

On the other hand, upskilling and retraining programmes may reduce the impact if the displaced workers manage to find employment immediately. Regardless, according to the study, such measures are unlikely to fully balance the demand loss even in the most realistic situations.

According to Falk and Tsoukalas, the only policy that directly addresses the issue is a Pigouvian tax on automation. A Pigouvian tax is a tax that is imposed to correct an externality that arises when a person engages in behaviour that harms others without their consent. In this case, it would involve taxing firms for each unit of labour they replace with AI to reflect the broader economic damage caused by reduced consumer demand. The researchers feel that such a tax would align private incentives with social outcomes, discouraging excessive automation.

With AI adoption accelerating across industries, the paper underscores the need for policies that go beyond managing the consequences of automation and instead address the incentives driving it.