“AI Has Yet to Boost Productivity as Much as Expected”

[Financial News]  

An analysis found that AI has not raised productivity in the U.S. economy as much as expected. In other words, despite massive investment, the returns have yet to match it.
Axios reported this on the 28th (local time), citing a report by the Federal Reserve Bank of San Francisco.
On the surface, the effects of AI investment appear substantial, according to the report. Since early 2023, the annual productivity growth rate has risen by 1.3 percentage points, as investment increased and the use of labor and equipment intensified. This is remarkable growth, close to twice the increase seen over the previous 20 years.
A closer look, however, makes it unclear whether AI itself drove this productivity improvement. Excluding the effects of additional investment and more intensive use of resources, productivity has risen at an annual rate of 0.7% since early 2023. That is not much different from the annual growth rate of 0.5% in 2005–2022.
Some investments may themselves be related to AI, such as software that helps employees produce more in less time.
In the Beige Book, the Federal Reserve’s report on regional economic conditions, the Federal Reserve Bank of San Francisco said that traditional improvements such as training and process standardization contributed almost as much to productivity gains as AI did.
Productivity gains have, however, been particularly pronounced among AI developers.
In its preliminary prospectus for an initial public offering (IPO), or S-1, released on the 26th, Anthropic said that its AI model Claude handled more than 90% of its coding last month.
According to Anthropic, the average amount of coding per engineer per day has increased more than 45-fold compared with the company’s 2024 level.
Productivity gains do not appear to be nearly as pronounced outside AI developers. Although vast sums are being poured into AI, tangible results have yet to emerge across the broader economy.
Keith Waddell, CEO of staffing firm Robert Half, told investors last week, “There is considerable leakage in the process of turning individual productivity into organizational productivity.” Waddell added that it was true that workers’ productivity had improved to some extent.
It is also difficult to clearly distinguish productivity gains from the effects of AI.
S&P Global aims for productivity gains of about 2% a year, but has found that isolating the effects of AI alone is nearly impossible.
S&P Global CFO Eric Aboaf said on an earnings call with investors on the 27th, “Some of it is a direct effect of AI, but some productivity gains are unrelated to AI,” adding, “Most of it actually gets mixed together and bundled up.”
It appears that more time will be needed for AI’s gains to show up in economy-wide productivity statistics.

[email protected] Song Kyung-jae Reporter