An AI-driven efficient economy could also end up eliminating all demand

In many ways, a gig economy worker in a third-world country lives better than any 19th century monarch. The average 21st century individual lives decades longer, while enjoying better health and access to far superior medicare. Our infrastructure lets us travel easily across continents, and instantly communicate with people on the other side of the world. We can sit at home and watch an opera in Milan, while enjoying a Thai curry matched with a South African wine.

 

All this has been enabled by technological advance. Starting with the Industrial Revolution, there have been startling gains in productivity across three centuries. New technology goes mainstream and translates into huge improvements in human development indicators.

 

Thus, over three centuries it’s become a truism that the disruptions caused by technological advances eventually make things better: The jobs lost as an old technology is superseded are more than compensated for by new employment opportunities that arise. 

 

The arrival of the automobile and the disruptions it caused are often cited.  Many jobs and professions disappeared when the horse ceased to be the primary mode of transport.  But far more new opportunities arose across the new value chain. The positive externalities from the availability of easy personal transport created even more opportunities.

 

There can be issues during the transition period though. There’s a loss of consumption as old jobs disappear. It’s assumed that this is eventually offset by the new consumption generated by those who benefit from the new opportunities. But if the income lost by displaced workers isn’t quickly replaced by consumption from beneficiaries, there’s a problem.

 

What if productivity comes without commensurate growth in employment? We’ve seen “pilot” versions of this over the last 30 years, in scenarios that have played out across borders and industries. Manufacturing and information technology (IT) services have moved out of expensive high-wage locations.

 

Much of the turmoil around immigration has its roots in that transition. Low-skilled workers in the US have been left stranded as their employment opportunities vanished. This is a primary reason why non-college graduates in America are anti-immigration.

 

What if the productivity gains are effectively limitless, and the new employment opportunities are net negative? This may seem theoretical but it is a scenario economists have started to explore as artificial intelligence (AI) penetration increases across industries and nations.

 

 A new paper, “The AI Layoff Trap” by Brett Hemenway Falk and Gerry Tsoukalas, which looks at this, ends with a sentence that is stark and dystopian in its conclusion: “At the limit, firms automate their way to boundless productivity and zero demand.”

 

The paper, which was published in March, by two academics at the Wharton School, University of Pennsylvania and Boston University, is peer reviewed and mathematically modelled. It has a simple premise. Assume an economy that can produce everything quickly and efficiently. But there is no demand for anything because nobody is employed. The outcome would be catastrophic, not disruptive.

 

We could be soon seeing this sort of death spiral in some industries. Company A fires a large chunk of its workforce and replaces it with AI. Competitor company B does the same. Rinse and repeat. This is perfectly rational — we’re already seeing dark factories and mass replacement of low level IT workers and white-collar clerical workforce.

 

As unemployment rises, consumption drops. This causes a negative feedback loop where companies cut costs, with more layoffs and more AI. This leads to another round of falling demand, and another round of seeking more efficiency via automation. This could spread across the macro economy as AI improves.

 

What sort of policy interventions could break this death spiral? The paper considers concepts like universal basic income, capital income taxes, worker ownership of equity, and upskilling programmes. The writers believe none of these would work, though they may be wrong.

 

The only intervention that the researchers think would work is odd. Charge an “automation tax”. Every time a human is replaced by automation, impose a levy to compensate for the lost demand. The mechanism for translating this into demand would be complicated, and I cannot recall it being discussed seriously hitherto.

 

There’s nothing illogical about the extrapolation of visible trends to this extreme conclusion. It challenges the historical patterns of the last three centuries but that doesn’t mean it couldn’t happen. The implication is terrifying: AI could destroy the economy simply by being very good at what it does.