Why Tech Gets Cheap but People Don’t

Productivity surges in one sector (like AI) trigger Jevons-style demand explosions and cheaper outputs there, while pushing up wages and prices in less-productive, human-intensive services via the Baumol effect. This can even occur within a single job as AI automates most tasks, leaving a scarce ‘human-in-the-loop’ bottleneck that earns a premium. The result is a wealthy but odd-looking economy where goods get cheap, services get costly, and the right strategy is to keep driving productivity.

Key Points

• Jevons paradox: large productivity gains lower costs and unlock vast new demand, expanding jobs and uses (e.g., computing, AI tokens).
• Baumol effect: sectors with limited productivity growth still see rising wages and prices because they compete in the same labor market as booming sectors.
• AI will amplify both forces: some services will get radically cheaper and more abundant, while many human-intensive services will get pricier yet remain in demand.
• Within single jobs, the human ‘last 1%’ (often mandated by regulation) can become a high-wage bottleneck until fully automated.
• Prosperity looks weird: goods get cheap while human services get costly, but overall wealth increases make those costs bearable.

Sentiment

The overall sentiment of the Hacker News discussion is largely critical and skeptical. While some commenters found elements of the article's core arguments agreeable or offered reinforcing examples, a significant portion challenged the article's definitions, specific examples, underlying economic assumptions, and the author's perceived credibility, often citing oversimplification or misunderstanding.

In Agreement

• Childcare serves as a compelling modern example of Baumol's effect, being a highly labor-intensive service with minimal opportunities for productivity enhancements, which explains its escalating relative cost.
• The Baumol effect is accurately demonstrated by professions like concert violinists, whose wages have risen despite zero productivity growth in their field, driven by overall productivity gains elsewhere in the economy that prevent them from accepting arbitrarily large pay cuts.
• The 'Turbo-Baumol' scenario for human-in-the-loop automation, such as with radiologists, could be managed by AI 'boxing' questionable cases for human review, which may help radiologists stay sharp and provide valuable training data over time.

Opposed

• The article's definitions of Jevons Paradox and the Baumol Effect are inaccurate, particularly its assertion that 'we'll consume more of everything'; instead, new production bottlenecks will command higher prices, but not all industries become bottlenecks.
• Specific examples used in the article, like comparing the cost of fixing a wall to buying a flatscreen TV, are deemed facile, meaningless, and reflective of a privileged viewpoint rather than insightful observations about labor costs.
• The author's admission of speculating from a place of ignorance (e.g., about radiologists' workflow) undermines the credibility and seriousness of the article's arguments.
• The article misinterprets Jevons paradox in the context of AI tokens by confusing the cost of 'creating/extracting the resource' (token generation economics) with the 'cost of using the resource' (calling an API).
• The article's premise that productivity growth in one sector necessarily increases wages in other sectors is challenged by the observed and growing disconnect between productivity and wage growth.
• The claim that people won't accept less pay for enjoyable work, which underpins the wage-pull effect, is debatable, as many individuals do prioritize job satisfaction over maximum earnings.
• Skepticism exists regarding the direct link between AI and the described effects, especially given current downturns in tech hiring and wages, with 'Dutch disease' suggested as a more relevant economic concept for some of the phenomena.
• Concerns are raised about the effectiveness of human-in-the-loop safeguards in automated systems (e.g., radiologists), predicting that humans may suffer from attention fatigue and simply 'spam LGTM' when 99% of assessments are correct, potentially leading to skill atrophy and reduced human oversight.