Fast-embraced AI will be slow to lift productivity

It’s more likely the numbers are roughly correct, and the AI era will be nothing like its internet counterpart, when PCs already sat on every desk. To use LLMs, businesses are retraining employees and hiring consultants, Federal Reserve surveys show. Such costs probably help explain why output and employment gains for AI users have been hard to spot. The investment splurge is nevertheless a direct boon for developers, even as concerns about potential threats prompt a movement to throttle chatbot advancements. Such fears may be overblown, but either way the broad societal payoffs will take time to show up. Follow Jon Sindreu on X and LinkedIn. CONTEXT NEWS Nonfarm U.S. labor productivity increased 2.2% in the second quarter from a year earlier, the Bureau of Labor Statistics said on September 3. Unit labor costs grew 1.2%, reflecting a 2.6% rise in hourly compensation and a 1.4% increase in productivity. Dario Amodei, CEO of AI developer Anthropic, on September 12 argued for a slowdown in the development of artificial intelligence, joining a growing number of executives and politicians raising concerns about potential threats from the technology. (Editing by Jeffrey Goldfarb; Production by Pranav Kiran) ((For previous columns by the author, Reuters customers can click on SINDREU/[Jon.Sindreu@thomsonreuters.com])) The rate surpasses the 1.5% of the 1980s. But during the latter part of the 1990s, huge capital expenditure contributed to a 2.7% productivity leap, with quarterly rates that sometimes exceeded 4%. Then, slowing inflation underpinned consumption. Today, wars and supply-chain snags are eroding real incomes. It's possible statisticians are understating the value U.S. AI giants capture from the huge spending on data centers when products designed in America are manufactured and sold overseas, as with Nvidia NVDA.O chips made in Taiwan and shipped to Europe. Annual GDP measurements could fall short by as much as two percentage points by 2028, research institute Epoch AI estimated last month. More generally, data often struggles to reflect upgrades in quality from technology. The baffling lack of a productivity windfall from early investments in the 1980s, famously noted by Nobel laureate Robert Solow, prompted "hedonic adjustments." These mechanically boosted 1990s output and reduced inflation as dollars went further in real terms. Most of the spending surge then reflected computers getting 40% faster annually, rather than companies buying more hardware. Perhaps services are equally improved by today's thinking machines and the statistics can’t distinguish it, thus overstating inflation. Even so, beyond the legal services consumer price index, which the U.S. Bureau of Labor Statistics stopped publishing because of its unreliability, there are few suspicious jumps in price growth. Even in wholesale trade and finance, where AI adoption exceeds 60%, inflation is consistent with worldwide trends. The author is a Reuters Breakingviews columnist. The opinions expressed are his own. *By Jon Sindreu* LONDON, Sept. 16 (Reuters Breakingviews) - A single contract clause that would have taken a lawyer hours to research three years ago can now be done by a specialized large-language model in minutes. Although such AI-powered time savings repeat millions of times daily in offices and factories, they cannot be found in productivity statistics. It suggests the latest technological revolution will echo the 1980s more than the 1990s. Corporate America is adopting artificial intelligence at three times the pace it did personal computers some four decades ago. About 44% of workplaces were using AI in May, according to a tracker supported by Harvard Business School. It isn't translating into observable labor efficiency, however, and industry executives are now calling for a slowdown in developing the technology. Hourly U.S. output expanded at an annual 2.2% pace in the second quarter.

Fast-embraced AI will be slow to lift productivity

The broader payoff may take time to appear

Why the data may understate or overstate AI's effect

AI adoption is fast, but productivity gains are still hard to see

Related News

Pure Cycle CEO Harding to present at Lytham Partners Fall 2026 investor conference

McGraw Hill acquires AI instructional coaching platform TeachFX

PetMeds names Jeff Willard CEO, president effective Sept. 28

Applied Digital jumps after Wells Fargo initiates with 'overweight'

Pagaya adds Marqeta founder Jason Gardner to board of directors