Why do economists generally seem more cautious than many folks in Silicon Valley about the potential economic impacts of artificial intelligence, especially when forecasting extreme scenarios? One big reason: Their baseline case is informed by economic history, a hardly unreasonable starting point when dealing with an important new innovation.
And what does history show? The economic stories of transformative technologies often display two important plot beats. First, expect a rough patch at some point. Canal mania collapsed in 1837. Railroad overbuilding fueled the panics of 1873 and 1893. Turn-of-the-century electrification saw speculative excess in utility holding companies. The 1990s dot-com boom went bust.
Yet in each instance, business productivity eventually surged, living standards climbed, and the prophets of permanent technological unemployment turned out to be wrong. Workers got reallocated, not eliminated. And while what those workers did changed shape, work itself didn’t disappear.
In a new analysis, “Lessons from the Five Innovation Waves That Preceded AI.” Morgan Stanley’s economics team highlights this encouraging pattern across 250 years and five broad innovation waves. From the paper: “While technological revolutions invariably bring short-term disruptions and uncertainty, they also deliver significant long-run benefits.”
Certainly, from a pro-progress perspective, the most intriguing finding is the denouement of these five waves rather than the mid-course disruptions:
- Canals and early factories accelerated America beyond its agrarian roots, cutting farm employment from 75 percent to just over half by 1850.
- Railroad investment averaged 2.5 percent of GDP annually, connecting regional markets into a continental economy.
- Electrification and motorization helped double output per hour in a generation and enabled mass production and national consumer markets.
- Postwar electronics and aviation coincided with GDP growth above four percent a year, alongside a major expansion of federal R&D.
- The internet era roughly doubled productivity growth in the late 1990s, shifting investment from factories and structures toward software and intellectual property.
The big analytical question: Might AI, a possible sixth innovation wave of equal or greater importance to those other five, be different? Morgan Stanley doesn’t rule it out. The bank acknowledges that AI could mostly substitute for labor rather than augment it—and trigger a falling labor share and more concentrated gains.
From the report: “We do not rule out that AI could defy historical precedents and create more extreme outcomes.”
That’s their robopocalypse caveat, basically.
The historical record, however, suggests the burden of proof belongs to the discontinuity camp. So far, their case rests more on what AI could theoretically do than on what labor markets are actually showing.
And what are they showing? According to a second Morgan Stanley report—one very much reflecting the current state of play—the bank finds not so much. Its new AI disruption tracker finds aggregate labor market impact remains tiny, at most about 10 basis points on the unemployment rate, and even that may reflect post-COVID overhiring rather than AI.
Yes, there are some early signs of stress among younger workers in AI-exposed roles. But at the industry level, the sectors most exposed to AI have actually been hiring faster than others, although the report cautions this may reflect underlying trends rather than AI itself. Moreover, workers in those roles increasingly report changes in their day-to-day tasks, a pattern more consistent with AI reshaping work than replacing workers. And even these signals remain statistically weak.
By the way, Goldman Sachs is out with a similar finding, one that also considers some underappreciated upside impact of AI on the job market:
Our estimates using our new measure suggest that AI substitution and augmentation have been a modest net drag on the labor market over the past year, reducing monthly payroll growth by roughly 16k and raising the unemployment rate by 0.1pp, with much of the cost falling on less experienced workers. The true aggregate impact of AI is likely smaller than this, as these estimates do not fully capture the offsetting effects of hiring for data center construction or incremental labor demand generated by AI-related productivity and income gains.
So far, at least, the 250-year track record is holding.
The post The AI Jobs Scare Meets 250 Years of Data appeared first on American Enterprise Institute – AEI.
