If AI steals jobs, who will be left to buy all the stuff?

Chris Gay is a contributing columnist for The Globe and Mail. He is a former Wall Street Journal staffer and writes the newsletter Figure at Center.

Going by the direst predictions, artificial intelligence will make human workers irrelevant over the next few years, slashing the ranks of knowledge workers, in particular, by huge percentages. That raises an alarming prospect: a world in which few are employed and therefore few consume. Put somewhat facetiously, if no one is employed, who is going to buy stuff?

Since personal consumption is a core driver of modern economies, this could be a problem. When we talk about “economic activity” in advanced economies, we are talking largely about consumer spending, often described as the engine of the economy. That’s because in industrialized societies, consumer spending can account for half or more of gross domestic product. (It’s 55 per cent in Canada, 67 per cent in the United States.) This implies a significant economic threat should the worst predictions for AI-related job losses prove accurate.

The economists I’ve consulted know of no study examining precisely the question of AI-impaired consumer spending, but they have studied the potential impact on employment and spending generally. There’s not much doubt that a significant fall in employment would reduce personal consumption and by extension overall growth, at least as a first-order effect.

As banks set the tone for AI adoption, senior executives rethink hiring

We can get a rough idea of how much by looking at the COVID-19 crisis. U.S. consumer spending fell 2.7 per cent in 2020, while real GDP shrank 3.5 per cent, a postwar record. Canada’s personal consumption contracted by 3.2 per cent (5.2 per cent without government transfer payments) and real GDP by 5.4 per cent, the latter Canada’s worst GDP performance since comparable data was first recorded in 1961. Worldwide, collective consumer spending shrank 4.5 per cent and world GDP by 2.9 per cent.

To be sure, the COVID-19 crisis involved a supply shock, not just the demand shock that AI job losses could produce. But what would be the impact on GDP if AI reduced employment by, say, 10 per cent, roughly what happened in 2020?

That depends partly on two things: what economists call the “marginal propensity to consume” – how much of each discretionary dollar a consumer spends, as opposed to saving – and “multiplier effects” that amplify the total spending generated by an initial purchase as a dollar spent by one consumer moves on to the next consumer, and so forth. Multiplier effects magnify spending, but also spending contractions.

A simple textbook example suggests that, assuming consumers on average spend 60 cents of each dollar of discretionary income, a 10-per-cent reduction in employment would mean about a 25-per-cent decline in GDP in the short term, all else equal. (Variables include such factors as household consumption’s share of GDP and compensating government responses, such as unemployment insurance.)

A lot could depend on “what happens to all the money saved by AI from those job losses,” says Stanford University economist Nicholas Bloom. “Consumption could drop if the savings to companies don’t end up getting spent.”

Self-fulfilling psychology could also inhibit consumption even if AI does not carve a swath through labour forces. Knowledge workers in fields most vulnerable to AI-caused job displacement could ramp up precautionary saving in anticipation of hard times, reducing overall spending.

Will AI be the end of jobs? The reality looks different in Canada

That’s partly because of the “wealth effect,” which cuts in two directions: When people feel secure in their jobs and their stock portfolios are up, they’re likely to spend more. A shrinking portfolio and job insecurity, naturally, means they spend less. Multiplying this “reverse wealth effect” by millions of people is not good for GDP.

On the bright side, not everyone foresees an employment apocalypse in an AI-driven future. AI could serve merely to supplement existing work, or even increase employment. “I think there will be some job losses and some job gains, probably netting each other out,” says New York University economist Robert Seamans, who studies the economic consequences of advanced technologies.

Some argue that AI will prove no more disruptive than previous watershed technologies, and that today’s handwringing is just the latest in a long history of doomsday hysterias, from Malthusian impoverishment to the “population bomb.”

The AI job apocalypse is coming. Or is it?

But it may be prudent to expect the worst outcome, not an ideal one, if only because AI is a species for which history may offer little guidance. The loom and the printing press displaced workers, but they weren’t capable of recursive self-improvement or general intelligence, and they couldn’t easily infiltrate the entire industrial landscape.

“The adoption of these technologies is the fastest we’ve ever seen because the adoption cost is so much lower,” says ​​Ekkehard Ernst, chief macroeconomist for the International Labour Organization. “Anybody can open a ChatGPT account without much cost.”