The good, bad and ugly of AI’s economic impact
LONDON, July 9 (Reuters Breakingviews) - A year ago Ford Motor (F.N)
CEO Jim Farley warned
“artificial intelligence is going to replace literally half of all white-collar workers in the U.S.”. In February Mustafa Suleyman, head of Microsoft’s (MSFT.O)
AI unit, upped the ante by predicting
that most white-collar work would be fully automated within 12 to 18 months.
Sceptics may dismiss these prophets of doom as riddled with self-interest. Yet signs that AI is disrupting human labour markets are everywhere. In March, self-driving robotaxi company Waymo announced
that it had hit more than 500,000 paid rides per week. Last month the United Kingdom unveiled a new Defence Investment Plan
“a seismic shift to smaller, cheaper, uncrewed and autonomous systems” driven by computers’ comprehensive besting of humans in the Ukraine war
. Last week, the U.S. Center for AI Safety reported
that the ability of the leading large language models (LLMs) to substitute for human workers, as measured by its Remote Labor Index
, has quadrupled in just 8 months.
Yet for all these ructions, evidence that AI is displacing jobs in the aggregate remains elusive. Even in the United States, the epicentre of the AI explosion, a recent comprehensive review
found that “most datasets find little evidence of economywide job loss or wage decline,” concluding instead that the impact of AI to date has shown up in “task reallocation and within-firm productivity gains, rather than mass displacement”.
There are three reasons for this mismatch between the micro and the macro lens. The first is that AI’s main achievements to date have come in automating tasks for which success can be clearly defined. Predicting the most likely next word in a sentence – the bread and butter of LLMs – is one such case. Winning at games like chess or go is another. For most real-world tasks, however, it is much harder to specify a precise “loss function
” to guide self-teaching machines. This verification bottleneck
frustrates comprehensive automation and requires that a human stay in the loop.
The second reason is that the majority of economically useful human occupations do not consist of discrete tasks but are what the economists Luis Garicano, Jin Li, and Yanhui Wu call “messy jobs
”. They comprise complex combinations of tasks, only some of which can be automated, and rely on tacit and embodied knowledge
that cannot be digitally codified.
The third reason is that the doomsayers focus only on the supply side of the economy when the demand side matters too. Specifically, if the price elasticity of demand is positive, the so-called Jevons paradox
obtains: demand rises as production efficiency improves sufficiently to boost, rather than shrink, employment. That explains why call centres in the Philippines continue to hire
at a rapid clip, the number of radiologists in the United States has risen
by nearly 20% since Nobel Prize-winner Geoffrey Hinton predicted in 2016 they would become obsolete; and even employment of American coders is still creeping up, hitting a new all-time-high of 1,875,000 in 2025, according to
Bureau of Labor Statistics data.
The importance of the demand side is even more basic than that. It has been 30 years since computers check-mated
humans at chess. Yet today human tournaments draw bigger crowds than ever and pay out record prizes
. Educational resources of every conceivable kind are freely available on the web. Even so, students still flock to human tutors and patronise yoga studios. The lesson is one of the most basic in economics: consumer preferences matter just as much for market outcomes as production technology.
Even if AI won’t take our jobs, however, it does pose other, more insidious, macroeconomic threats. One is deskilling – the risk that delegating work to AI depletes humans’ own mental capacities. One alarming study
, for example, documents a strong negative correlation between the ability to think critically in adults and the use of AI tools. Another
found that students who were AI-tutored, but later had access withdrawn, performed nearly 20% worse on maths tests than peers who had never used the technology.
A second left-field risk is that AI will crush the creative diversity on which modern service economies depend. An infamous 2020 article
by marketing guru Alex Murrell argued that individual imitation of past successes has produced an “age of average” in which distinctiveness has died and everything looks the same, leaving everyone worse off. Recent research
from Tilburg University suggests that generative AI, with its reliance on probabilistic extrapolation from vast datasets of historical text and images, is making such homogenisation worse.
Then there is the fact that AI has dropped the cost of many bureaucratic processes close to zero. This is why recruitment campaigns regularly attract unmanageable quantities
of AI-authored applications, and public consultations on planning or new regulation are swamped by AI-generated responses
. AI risks ensuring that Brandolini’s Law
– which states that “the amount of energy needed to refute bullshit is an order of magnitude bigger than that needed to produce it” – increasingly applies on an industrial scale.
These are serious drawbacks. Yet a final potential risk stands to put them in the shade. In a brilliant recent column
the Oxford University economist Carl-Benedikt Frey
points out that if a technology allows consumers to supply the service they demand themselves, rather than making existing producers more efficient, the benign Jevons paradox does not hold. Just as washing machines abolished washerwomen, or online booking decimated real-life travel agents, AI’s empowerment of people to be their own lawyer, accountant, or even plumber will drive activity out of the market sector altogether and into the home. Jobs across the service economy would then vanish.
This version of the AI jobs apocalypse would be even uglier than the doomsayers predict. In the first chapter of “The Wealth of Nations” Adam Smith explained
that specialisation and the division of labour are the ultimate secrets of capitalism’s success. It would be a bitter irony if AI were to unravel the very system which has produced it by returning us to the preindustrial age.
Comments
Post a Comment