This is a clip from The Breakdown newsletter. To read the full editions, Subscribe.
“Productivity isn’t everything, but in the long run it is almost everything.”
– Paul Krugman
“Total factor productivity” (TFP) is how economists measure the contribution of technological innovation to economic growth – the sustainable ability of an economy to produce more output with the same amount of input.
As such, this is arguably the most important measure for economists, because the constant process of producing more with less is how life improves.
Paul Krugman explains that “a country’s ability to improve its standard of living over time depends almost entirely on its ability to increase its output per worker.”
Technology is what makes it happen, and TFP is how we measure it.
To get a more concrete understanding of how important technology-driven productivity is, consider the following: a Recent paper The National Bureau of Economic Research estimates that an additional 0.5% increase in annual factor productivity growth would stabilize US government finances at today’s debt-to-GDP level.
0.5%!
That doesn’t sound like a big number, but if it holds over the next 10 years, the National Bureau of Economic Research estimates that it would reduce the baseline forecast for US government debt by $2 trillion.
More than 30 For many years, sustainably boosting factor productivity by 0.5% would leave the US government’s debt-to-GDP ratio about 42 percentage points lower than the National Bureau of Basic Economic Research’s forecast (and about 80 percentage points lower than its pessimistic forecast).
Given the desperate state of government financial resources, maintaining today’s level of indebtedness constitutes a dream scenario that seems too good to be true.
But Anthropic researchers believe we can do better.
Anthropic conducted a study of 100,000 Claude.ai conversations to “estimate how long tasks in these conversations will take with or without the help of AI, and examine the implications for productivity across the broader economy.”
Its conclusion? LLMs can raise total factor productivity by 1.1 percentage points.
1.1%!
If 0.5% can stabilize the US government’s finances for decades, what might 1.1% do? It will probably fix almost everything.
There are, of course, reasons to question these optimistic forecasts.
The study found, for example, that Claude saves teachers four hours of work by creating curriculum in just 11 minutes. But estimating how this time saving might increase economic output requires the kind of economic modeling full of best-guess assumptions and false precision.
So, even if Anthropic is right about saving time, it might be wrong about productivity: We might be using all the time AI saves us to do something economically unproductive, like watching more TikTok videos or reading more newsletters.
In this case, AI will increase our well-being (more leisure time) but not our wealth (more economic output) — which is still great news for people, but not helpful for governments hoping for a magic solution to their debt problem.
Conversely, there are reasons to believe that the anthropic model is also like this pessimist“We don’t take into account the rate of adoption,” he explains, “or the greater productivity impacts that might come from more capable AI systems.”
In other words, her study assumes that we continue to use AI just as we do now, and that we are still using existing language models, without improvement, for another. 10 years.
Language models are improving significantly every few months, and we’re only just starting to learn how to use them — so Anthropic is right to say that their estimates may represent “a rough lower bound on AI productivity impacts.”
If so — if 1.1% is minimum As for AI-induced productivity – we might be able to pay off government debt and You have more time for TikTok.
And this is just taking into account the impact of AI on non-physical work – just wait until we get robots!
To dismiss this optimism entirely is to believe that the trillions of dollars that companies plan to spend on capital expenditures in AI and R&D will be wasted.
Perhaps it is the case that technological revolutions do not always arrive on time.
But the biggest reason for optimism is that Anthropic’s 1.1% estimate relies only on AI to “make existing tasks faster to complete” — its model doesn’t fully take into account AI’s potential to change the way we complete those tasks.
“Historically, transformative productivity improvements—from electricity, computing, or the Internet—have come not from speeding up old tasks, but from fundamentally reorganizing production,” Anthropic notes.
There is no way to model these new ways of doing things, but it seems likely that their impact will be greater than that which Anthropics has attempted to measure.
The Anthropic researchers are careful to caution their hopeful findings by enumerating the limitations of their methodology and documenting the many assumptions they make.
Even if all of these assumptions work out and AI productivity solves the US government debt problem, lawmakers will likely do their best to return to this problem.
But given the 100% probability that everyone seems to expect a looming financial catastrophe, even a small chance that Anthropic’s estimates will prove correct is reason to update our forecasts: the US government’s finances are not as insurmountable as we think, and the US dollar is not as doomed as we think.
In the long run, productivity is almost everything, and AI may be about to make us even more productive.
Get news in your inbox. Explore Blockworks newsletters:
