Torsten Slok is right: There is no “AI” in the TFP data. But he's wrong to stop there. "No sign of AI in the productivity data" is a headline that confuses the scoreboard with the game.
He presents an SFFed chart:
and comments:
Torsten Slok: No Signs of AI in the Productivity Data <https://www.apollo.com/wealth/insights-news/insights/daily-spark/No-signs-of-AI-in-the-productivity-data>: ‘Hundreds of billions of dollars are flowing into data centers, chips and model training, and that capital deepening should mechanically lift output per hour. The harder question is whether AI is also raising TFP, meaning whether it is making the economy fundamentally more efficient. So far, the data says no. The chart below shows utilization-adjusted TFP from the San Francisco Fed, and it is currently sitting slightly below zero with no sign of acceleration since the AI capex cycle began. Output per hour, by contrast, is running near 2.5%, comfortably above its post-2005 average, and that strength is exactly what gets cited as evidence that AI is already working. But strong output per hour alongside flat TFP is the signature of capital deepening, not of a technology shock. The AI boom is clearly visible in investment data and in equity valuations, but it is not yet visible in the productivity statistics, which means the productivity payoff from AI remains a forecast rather than an observation…
A technology shock can show up in three places:
User surplus — enabling better-quality goods and services that statistical agencies never fully capture in measured output (without truly heroic hedonic adjustments).
Capital deepening — enabling more investment that clears the hurdle rate but not by much, thus lifting output per hour.
TFP — enabling more output from the same inputs, the thing economists actually mean by “total factor productivity.”
“AI” is certainly doing (1) (and will keep doing so unless it goes horribly wrong the way so much social media did, and our brains are hacked to our detriment by AI-slop). And AI is certainly doing (2): that is growing wedge over the past three years between the “labor productivity growth” and the “TFP growth” line. What "AI” is not (or not yet) doing is (3)—delivering a supernormal increase in measured labor productivity over and above what we would have expected to see given the hurdle rate capital-deepening projects need to earn.
Slok's chart doesn't show that AI isn't working. It shows that AI isn't yet working in the one way TFP measures. That's worth knowing. It is not quite the same as "no signs of AI in the productivity data".
Having spent a lot of time looking at TFP, I concluded it is not a robust measure of productivity. Much goes back to the reliance on the CD production function. And if you read the original article on that, your eyes will roll in your head -- statistical sloppiness. The CD production function is just not compelling, nor empirically supported, once you think seriously about the role of energy in production and technology.
What about technology shocks that show up as medium-term net losses, not gain? Look at longevity and repairability of major consumer items like cars, home HVAC, and home appliances. Digital controls in these items make some contribution to energy savings & improving the environment, but the products' relatively limited life span and non-repairability create a huge amount of premature junk that serves no good purpose. Junking a 15 year old home appliance solely because it has a bad board that was $50 or $100 of the item's original cost is a net cost, surely -- where is that accounted for? I chose these items because they are no better for a houshold's selfish purposes than the.less technologically advanced products which they replaced