My fellow pro-growth/progress/abundance Up Wingers in the USA and around the world:

Fears that technology will render human labor obsolete are nearly as old as industrial machinery itself. Of course, artificial intelligence has given these anxieties loads of fresh momentum. Will human workers yet go the way of the horse?¹

Goldman Sachs takes that concern seriously — and then largely rejects it.

In an illuminating new analysis, “How Concerned Should We Be About a Job Apocalypse?”, the bank’s economists reiterate their estimate that AI could eventually automate tasks accounting for roughly a quarter of all work hours in the US economy. That exposure is large enough to justify unease. It isn’t, however, the same thing as total job destruction of the sort AI doomers often warn.

Translating exposure into employment outcomes, Goldman’s baseline forecast implies that about 6–7 percent of jobs are displaced over the full AI adoption period. Again, that’s meaningful disruption, sure, but hardly mass extinction.²

GS: “We remain skeptical that unemployment rate increases will be permanent as long as human labor maintains a competitive advantage in certain aspects of production.”³

Here’s how GS newly quantifies the labor-market impact:

• Under its central scenario — AI adoption spread over roughly a decade — the unemployment rate rises by a little over half a percentage point at the peak, equivalent to about one million additional unemployed workers.
• If adoption is faster or AI proves more labor-displacing than past technologies, the hit could double to roughly 1.2 percentage points.
• Only in a combined downside case does Goldman see the peak impact rising to around 2.4 percentage points, or roughly four million workers — and even then, the joblessness is transitional rather than permanent. The rise reflects the time it takes for displaced workers to retrain, relocate, and rematch — not a lasting surplus of idle labor. As productivity gains spread and new roles emerge, employment, in Goldman’s telling, eventually catches up with the technology rather than being left behind by it.

The key insight here is historically based.⁴ Periods of rapid, technology-driven productivity growth have repeatedly displaced workers in specific roles, while expanding employment elsewhere. Goldman’s model assumes a 15 percent AI-driven productivity uplift spread over roughly a decade. Based on past relationships between productivity shocks and labor markets, the result is that aforementioned temporary rise in unemployment as workers transition.⁵

Most importantly, technological change reshapes the economy by creating new occupations. Only about 40 percent of today’s workers are employed in jobs that existed 85 years ago. Goldman argues AI will repeat this process: eliminating some tasks, transforming many jobs, and spawning new roles that are fiendishly difficult to predict in advance.

To stress the point in GS’s words: “While increased AI-driven displacement may create significant real income losses and economic headwinds for affected workers, AI will also create new jobs.”

None of this minimizes the pain of transition or absolves policymakers from tweaking the safety net accordingly. Still, the Age of AI is likely to bring labor-market turbulence — not mass unemployment. That, as well as a more rapid rise in living standards, new and better health cures, and an overall science acceleration.

I’ll take the trade-off.

¹ America today has some 6-7 million horses—a mere third of the number at the turn of the 20th century, when they were indispensable members of the workforce.

2 It’s never a good idea to jump to quick conclusions from economic data when the news business and social media have a love for novelty. In the myth-busting new paper “AI-exposed jobs deteriorated before ChatGPT,“ a groups of economist show that jobs most exposed to AI were already weakening well before ChatGPT’s late-2022 debut. Unemployment risk in those occupations began rising in early 2022, and new college graduates started entering AI-exposed jobs at lower rates months before generative AI went mainstream — pointing to broader economic and tech-sector forces, not ChatGPT, as the main cause. At the same time, the authors find that graduates with AI-relevant skills actually did better after ChatGPT, landing jobs faster and earning more.

3 Keep in mind: Goldman is talking about narrow, task-level AI deployed inside firms to automate specific work processes — not artificial general intelligence that can fully replace humans across all domains. Unless AI becomes broadly cheaper and better than humans at judgment, interaction, adaptation, and physical-world tasks, labor markets should continue to adjust as they have in the past.

⁴ The GS analysis is a good example of the Acela Corridor Consensus (as opposed to San Francisco Consensus) I wrote about recently.

5 AI is likely to repeat this well-worn pattern through several channels, the bank explains. Data centers, power grids, and cooling systems must be built, wired, and maintained, creating work for construction crews, electricians, and engineers. Less visibly, productivity gains change the economics of demand. By cutting costs and raising capacity, AI allows firms to serve more customers and offer new services. Efficiency doesn’t destroy jobs so much as multiply activity — and with it, employment.