How Big Will the AI Shock Be?
How Big Will the AI Shock Be?
“Stop Hiring Humans.”
Last October a San Francisco-based start-up company named Artisan used this slogan for a break-out public relations campaign. The company sells AI tools that work with humans to support sales and marketing activities. The campaign put this slogan on Bay Area billboards, kiosks, social media feeds, and its booths at business conventions such as TechCrunch Disrupt.
The campaign reinforced its main message with additional taglines: “Artisans Won’t Complain About Work-Life Balance;” “Humans are So 2023; “Artisans Won’t Come in to Work Hungover;” and “The era of AI employees is here.”
The PR campaign was a wild success. The slogan hit a nerve by saying out loud what hundreds of thousands of tech-savvy businesspeople had been reluctant to voice. Commercially available, domain-specific AI tools are sufficiently sophisticated today that most companies need to begin adding “AI employees” to their routine hiring plans if they want to remain competitive.
Previous posts have discussed how AI tools are beginning to cause labor market disruptions. The PR campaign by Artisan created a dramatic way to visualize the issue.
But just how big will the AI shock be over the next few years? And how will the AI shock be distributed across different groups of people and different geographies?
The answer, of course, is unknown. Most analyses of the potential scale and scope of the AI shock focus on estimates of lost jobs and income. The scale of change that is projected depends on how quickly today’s domain-specific AI tools will be disseminated throughout the U.S. economy.
Heather Long of the New York Times recently asked several prominent economists for their opinions.[i] According to MIT’s David Autor and Harvard’s Gordon Hanson, two of five authors of a widely circulated recent analysis,[ii] one benchmark for what we can expect is the so-called “China shock,” which refers to the period from 1999 to 2011 when fast-growing trade with China led to the loss of more than two million manufacturing jobs in the U.S.
Autor and Hanson argue that the China shock was large, partly because it happened in only ten years, partly because its effects were focused on one sector of the economy, partly because the losses were concentrated in a relatively small number of midwestern metropolitan regions, and partly because there was no targeted Federal government effort to help retrain displaced manufacturing workers so they could shift their careers into other opportunities. Few displaced workers ever made up for the losses they experienced. They have since become an important part of the political coalition that re-elected Donald Trump.
As Long summarizes, Autor and Henson argue that the AI shock will be less severe than the China shock for several reasons. First, they expect AI to take several decades to diffuse throughout the economy. AI tools help people make decisions, and it takes years for established companies to evolve their decision-making cultures.
Second, the diffusion of AI won’t be concentrated in one economic sector, like manufacturing. Instead, AI will diffuse throughout all sectors. AI targets white-collar office jobs, not the blue-collar manufacturing jobs. Third, the effects of AI will not be focused within specific metropolitan regions. Office jobs are scattered around Downtowns and suburban office parks throughout the nation. So, the geographic effects will be highly diffused.
The better analogy, according to Hanson, is the long-term effect of the most recent wave of office automation: PCs and word processing. Starting about the same time as the China shock in manufacturing, almost two million jobs in administrative support occupations, including secretaries, have been lost in the U.S. economy. But the loss was spread over two decades, was spread across almost every sector, and was equally spread across every state and locality. Labor market data suggest that companies largely coped with this evolutionary change through attrition rather than lay-offs.
Molly Kinder of the Brookings Institution adds an additional factor into her assessment of the AI shock.[iii] Her data lead her to conclude that the AI shock will focus more heavily on entry-level white-collar jobs, the kind that are most often held by recent college graduates. Marketing analysts, for example, are more likely to be replaced by AI tools than marketing managers. Likewise, Kinder argues that AI tools will replace sales reps before they replace sales managers. Few companies will be eager to trust AI employees in supervisory roles.
Applying this logic to other sectors could yield similar analogies. Universities will replace teaching assistants before replacing faculty. Law firms will replace interns and associates before replacing partners. Accounting firms, architecture firms, engineering consultants, newspapers, logistics firms, banks, investment firms, etc. may all jointly create a series of new challenges for recent college graduates who will struggle to gain new footholds in terms of career paths because of the impact of AI.
But I must say that I’m having a tough time accepting the relative complacency that Long finds among economists about the scale and scope of the upcoming AI shock. Today’s domain-specific AI tools already have the potential to unlock trillions of dollars of productivity gains by substantially reducing the demand for paid human workers. The incentive to adopt AI already seems greater for eliminating highly paid workers than for replacing lower paid people. The more sophisticated these tools become, the easier they are for companies to use. Lower paid workers who can be trained how to use sophisticated AI tools may yield more profits for companies. This logic may become even more difficult to resist as today’s AI tools evolve into the next generation of breakthrough General Artificial Intelligence (GAI) tools.
Company cultures can be slow to change. But not when trillions of dollars are at stake. Consequently, it may be more appropriate to think big about the scale and scope of the coming AI shock. Very big.
The AI shock could be a fundamental body blow to the entire system we currently use to generate earned income for most households throughout the economy. The coming AI shock to the economy will not be limited to lower and moderate-income households. It may be an even bigger threat to the viability of households who live in highly-mortgaged, tax-abated, high-rise Downtown apartments, as well as those who live in so-called McMansions in the nearby suburbs and exurbs.
According to the most recent data from the U.S Census Bureau’s 2022-2023 Survey of Program Participation and Income (SPPI) and the Federal Reserve Economic Data (FRED), Americans reported about $16.8 trillion of income from all sources in 2022. The largest proportion, about sixty-three percent, or $10.5 trillion, came from earned income such as wages & salaries.
It may seem impossible, and it may seem unnecessarily dystopian, but policy makers need to begin considering the possibility that within current planning cycles of 3-5 years, the explosive diffusion of increasingly advanced AI agents could undermine the capacity of a substantial proportion of currently employed people to sustain their current living standards with their current levels of earned wages and salaries from private sector jobs.
If the security of only one job in every five was threatened in the next five years, that would raise doubts about $2.5 trillion of earned income. That is the amount of income that all Americans currently receive from social insurance programs, the largest of which is Social Security.
The number of households that could experience sudden loss of earning potential would make the China shock seem small in comparison. In addition, these millions of downwardly mobile households will not be concentrated in a handful of cities and rural areas that could be ignored by national political currents. The scope of income loss would permeate every major urban region, and the scope of loss would disrupt governmental tax income and service provision for cities, counties, states, and the Federal government.
Now is the time for policy makers to confront previously unthinkable ideas about how the AI shock may require wholesale changes to the systems of social relations we rely on for households to receive earned income from private-sector employment. Perhaps we are facing a fundamental intellectual challenge to rethink our understanding of the era in which we live.
An analogy I’ve been thinking about recently is embodied in the series of three histories written by Arthur Schlessinger, Jr. that are collectively known as his “Age of Roosevelt” series. Published in 1957, the first volume was titled The Crisis of the Old Order: 1919-1933. It was followed by The Coming of the New Deal (1958) and The Politics of Upheaval (1960). These three histories provide a masterful account of the complex social, political, and economic dynamics that created the social relations and social institutions we still rely on today to ensure widespread prosperity and stability for most American households.
The crises to our system of social relations we are facing today are very different than the crises that came together to produce the era of New Deal reforms. But the scale of reforms we may need to cope with the crises we face today, and in our near future, may be best understood by refreshing our collective memory of how flexible American culture can be. Flexible social systems evolve and adapt. Inflexible ones break and collapse. Today we clearly face a crisis to our “old order.”
Bob Gleeson
IMPORTANT PROGRAMING NOTE
Substack now provides easy tools that encourage writers to add podcasts to their sites. Ever since Bill Bowen and I started this weekly post, readers have encouraged us to experiment with podcasting in addition to writing.
Consequently, we are happy (and nervous!) to announce our first podcast!
We will livestream this experiment from 11am to Noon Eastern Daylight Time on Tuesday, May 6th.
We picked this time to maximize the ability of our current readers to watch it live if they wish. That will be early morning for our readers on the U.S. west coast and late afternoon/early evening for our readers in Western and Eastern Europe. The podcast will be recorded and posted on the site, so it will not be necessary to join the livestream.
We hope this experiment will prove interesting! Our topic will be a discussion of the potential consequences of the AI shock to cities.
[i] Heather Long, “Another ‘shock’ is coming for American jobs.’ NYTs, April 1, 2025.
[ii] David Autor, David Dorn, Gordon H. Hanson, Maggie R. Jones, and Bradley Setzler, “Places versus People: The Ins and Outs of Labor Market Adjustment”. NBER Working Paper Series, January 2025.
[iii] Molly Kinder, “How AI Could Break the Career Ladder,” Bloomberg Weekend, November 15, 2024.