↗️ Measuring human welfare in the Age of AI
Economic output isn't exactly the same as human flourishing. As artificial intelligence reshapes both, telling them apart is about to get as important as it has ever been during the past 250 years
My fellow pro-growth/progress/abundance Up Wingers in America and around the world:
Whatever way you want to slice the data, we can safely say George Washington was among the richest Americans of his era. (The wealthiest may have been this guy.) Yet even if he had sold Mount Vernon before his death in 1799, he couldn’t have purchased an air conditioner, antibiotics, a television, a smartphone, or comfortable dentures made from acrylic resin and metal alloys rather than hippo ivory and human teeth. These were not super expensive luxuries beyond even Washington’s financial grasp. They were simply unavailable two centuries ago at any price.
Let’s pull the camera back: On Independence Day, 1776, output per person in the thirteen American colonies was below $2,500, adjusted for inflation. Nearly a quarter millennium later, US real GDP per person has risen to about $70,000—nearly thirty times higher. Put another way, what the average American produces in a year today would have taken an American in the early days of the Republic era roughly a generation to produce.
Yet even that dramatic comparison understates the real transformation from technological progress. Americans in 1776 couldn’t have produced the same sort of output that their modern-day counterparts generate no matter how long they labored.
Living standards rise when people get more of the same goods, but only up to a point. The satisfaction or benefit from any single good eventually saturates. Buying another pair of sneakers or adding another flat screen to your home—or another horse or something, to harken back to Washington—eventually runs into a wall of diminishing returns. What smashes through that wall is beautiful, wonder-working innovation. New goods, higher-quality versions of old goods, and new services (a) expand how much people can consume and, more importantly, (b) substantively alter the nature of their lives—and to the better overall, the past half millennium would suggest.
GDP is a great but not perfect economic measure
So good reason to think that a period of extraordinarily rapid technological change might well be especially beneficial. (You know—faster, please!) We might be entering just such a period. And merely looking at gross domestic product as our economic scorecard might actually understate the extent of the rise in living standards. In “When GDP Misleads: Inferring Living Standards from the Value of a Statistical Life,” economists Philip Trammell and Charles I. Jones start with a focus on another superrich guy without access to modern conveniences and necessities:
“The richest person in the world in the 1830s was Nathan Rothschild, whose personal net worth was around 0.6% of British national income. Despite this vast fortune, Rothschild died at age 58 in 1836 of an infection that $10 of antibiotics could likely cure today.”
Rothschild couldn’t buy antibiotics at any price, nor could anyone for another century. But even if he could, he’d still face diminishing returns from other goods, from his 15th horse and carriage, and then his 16th. The same goes for Mark Zuckerberg and, say, his 15th home. But new kinds of consumption are different.
From the paper
“Similarly, the richest people in the world today, such as Bill Gates or Jeff Bezos, presumably have a marginal utility from additional consumption spending that is zero. Nevertheless, their utility still increases when new goods (smartphones or LLMs) are invented. These examples suggest that more consumption of a fixed set of goods eventually hits a marginal utility of zero while the invention of new goods or higher quality goods continues to increase wellbeing.”
So here’s your trouble: First, GDP is bad at measuring the value of new goods and services, the stuff most important for improving human flourishing.
Second, there’s a Baumol’s cost disease issue that can distort GDP:
“Imagine a world of two goods: food and string quartets. Food has rapid productivity growth at 5% per year while string quartets has slow productivity growth at 1% per year. Real GDP growth is the weighted average of these two rates, where the weights are the spending shares on the two goods. In the optimal allocation, because of Baumol’s cost disease, the spending share on entertainment rises and real GDP growth rates fall toward 1% per year. The incongruity is that if string quartets had never been invented, real GDP growth would be 5% per year forever — creating arbitrarily large discrepancies.”
Third, Trammell and Jones also note that many broader changes in human flourishing fall largely outside green-eyeshade GDP accounting, including “changes related to political and economic freedoms, crime, environmental quality, and climate.” All that stuff is important, too.
Measuring human welfare in the Age of AI
Trammell and Jones propose a complementary measure to GDP based on the “value of a statistical life,” or VSL. It’s a metric used by government policymakers in cost-benefit analysis to estimate the monetary value people place on small reductions in their risk of death—by buying a safer car or paying for better medical care.
VSL allows a clever bit of analytical shorthand that the economists ingeniously employ. By working off VSL, they don't need some crazy-long inventory of every good, service, freedom, risk, or environmental amenity that people value, nor do they need an entirely separate estimate of the benefit people get from each. Instead, Trammell and Jones use people's willingness to pay for small reductions in mortality risk—and how that willingness to pay changes over time—as a more organic analytical window into changes in lifetime well-being. That, while also accounting for the fact that an extra dollar means less to people as they get richer.
Using their VSL-based approach, the authors find that American lifetime well-being has risen roughly five- to sevenfold since 1940. That’s far more than the doubling you get from traditional consumption-based measures.1 That gap is the big finding here.2
Again the logic: If people's willingness to pay to stay alive has grown several times faster than their consumption, something other than merely the ability to buy more stuff has been making life more valuable to them. VSL would seemingly be a logical candidate to measure exactly what GDP struggles to count: “Our main contribution is a formula that is much simpler than any attempt to start from national income figures and correct for the massive problems posed by new goods, quality improvements, and changing social norms.”
So: In periods of especially rapid technological change—a time of entirely new goods, radical quality improvements, breakthrough medicines, much longer and safer lives—a VSL-based approach may better capture improvements in well-being than solely using an output measure like real per capita GDP as a proxy for flourishing. You can see how this could be helpful if you think AI (or AGI or even ASI) will prove wildy transformative.
As the authors conclude, the VSL-based approach “is an alternative and complementary way of measuring the gains from new goods and higher quality goods, not only in the past but also perhaps in the future. To the extent that some of the benefits of artificial intelligence are not fully captured by GDP, they may be reflected in increases in the VSL.” 3
All this may seem like a pretty wonky point, and it kind of is, but learning how to statistically measure the world around us will only become more important in the years ahead, especially when Down Wing opponents of AI attempt to argue for delay or a pause in progress.
The authors get at VSL partly by looking at the extra pay workers require to take more dangerous jobs. If workers need $1,000 more per year to accept a job with one extra death per 10,000 workers, that implies a statistical-life value of about $10 million. One data series the authors use applies this wage-risk method to census data from 1940 to 1980, then links it to a 2024 government estimate, yielding the headline figure: lifetime well-being up roughly five- to sevenfold since 1940. A second approach ties the VSL to income and lands in the same neighborhood.
The authors stress the calculation is "just suggestive," and on less favorable assumptions about interest rates or how fast the value of life rises with income, the measure can show welfare barely budging, or even falling, since 1940. The headline gap holds under their baseline assumptions.
One fascinating caveat from the authors worth including here: “Finally, note that the “lifetime utility” our approach tracks might more precisely be called the expected utility assigned to life over that assigned to death. It is possible that the VSL has grown rapidly over time not only because our lives have rapidly grown more enjoyable, relative to marginal consumption, but also because we have grown more averse to death for other reasons: a loss of belief in a heavenly afterlife, for instance, or an increase in the desire to stay alive for altruistic reasons.”
On sale everywhere by James Pethokoukis The Conservative Futurist: How To Create the Sci-Fi World We Were Promised
