What Will Be Scarce?

  • Author/Source: Alex Imas (University of Chicago Booth), Substack
  • Original: https://aleximas.substack.com/p/what-will-be-scarce

  • Companion technical note: http://www.aleximas.com/s/Technical-Note-Structural-Change.pdf

  • Key Ideas

  • Economics is the study of decision-making under scarcity. If AI delivers material abundance, economics doesn't become irrelevant --- the kind of scarcity that matters changes. The question is what becomes scarce.
  • Imas's claim: as AI makes commodity production cheap, expenditure and employment shift toward a "relational sector" --- human-intensive, provenance-rich goods/services where the human element is part of the value. This is structural change driven by income effects, not just by Baumol's cost disease.
  • Empirical anchor for the structural-change story: Comin, Lashkari, Mestieri (Econometrica 2021) estimate that income effects account for over 75% of observed sectoral reallocation; price effects ~25%. As people get richer, they want fundamentally different things, not just more of everything.
  • Behavioral microfoundation: mimetic desire (Girard) makes some categories income-elastic in a way that doesn't satiate. Imas & Madarasz (Restud) show willingness-to-pay roughly doubles when subjects learn a random subset will be excluded from a product --- not status signaling, not scarcity heuristic, just preference for having what others don't. Imas & Mandel show AI involvement undermines exclusivity: human-made art gains 44% from exclusivity, AI-generated only 21%.
  • The Starbucks reversal as canary-in-the-coal-mine evidence: after years of automation, the company concluded reducing baristas had been a mistake. CEO Niccol cited handwritten notes, ceramic cups, "great seats." Hospitality and small details drove satisfaction; automation is being rolled back.
  • Implication for jobs: durable post-AGI jobs aren't "monitor AI" or "prompt engineering" (transitional roles in the automated sector). They're nurses, therapists, teachers, boutique fitness instructors, personal chefs, bespoke tailors, craft brewers, live performers, hospitality, care, and emerging roles like experience designers, human-AI collaboration artists, provenance certifiers, community curators.
  • Caveat 1: narrower claim than the strongest labor-share story. Aggregate labor share can fall while the relational sector still grows substantially. The argument is about sectoral reallocation in rich economies, not about labor share staying high.
  • Caveat 2: the framework works for the developed world. For the developing world, whose economies depend on producing commodities for rich countries, the picture is more complicated and potentially worse.

  • Demand-collapse counter: mimetic desire is a release valve against the negative-growth scenario --- comparative preferences don't satiate, so expenditure can keep shifting toward relational margins as incomes rise.

  • Summary

Imas applies the economics of structural change to the AGI question. The classical pattern is well-documented: as a sector gets dramatically more productive, its share of GDP shrinks (agriculture from ~40% of US employment in 1900 to <2% today) while spending and labor reallocate to higher-income-elasticity sectors. Comin, Lashkari, Mestieri formalize this with non-homothetic preferences and find income effects dominate price effects 3:1.

The novelty of the post is the behavioral micro-foundation. Imas argues that mimetic preferences --- our valuation of goods is partly a function of how much others want them and can't have them --- give a large class of goods especially high income elasticity. This dimension does not satiate, because it is comparative. Crucially, AI involvement destroys the exclusivity premium (the Imas-Mandel art experiment), so AI commodifies its own outputs in a way that reinforces the human-made/AI-made wedge. That wedge is what fuels the relational sector.

The framework predicts: automated sectors shrink as a share of GDP; relational sectors grow; the kind of work humans do shifts toward jobs where presence, judgment, attention, memory, warmth, or provenance is the product. Imas is careful that this is not a Marxian decommodification --- a tailor selling relational labor to capital is still selling relational labor to capital --- but it is a real shift in the composition of demand.

The companion technical note works out the formal model with non-homothetic CES, sectoral productivity differentials, and the mimetic dimension of preferences. The post itself is the prose case.

  • Relevance to Economics Research

This is direct economic theory of the AGI transition, by an active researcher (Imas is at Chicago Booth, with Restud, AER, etc. publications). Several reasons it matters for the wiki's audience:

  1. Theory anchor. Most of the AI-and-econ conversation is empirical, journalistic, or extrapolative. Imas grounds the discussion in standard structural-change economics (Comin et al., Hubmer's "Race Between Preferences and Technology") and a behavioral micro-foundation he's published on. That's the right shape for an economics-paper-quality argument.

  2. Specific testable claims. Top-quintile vs. bottom-quintile expenditure ratios on relational categories from the BLS Consumer Expenditure Survey are concrete, replicable starting points. The Imas-Madarasz exclusion experiment and the Imas-Mandel AI-art-exclusivity experiment are published primary evidence.

  3. Useful counterweight to two extremes. The "AI eliminates all jobs" story (Trammell-asymptotic) and the "AI changes nothing" story (some Acemoglu-skeptical reads) both miss the structural-reallocation channel that Imas is highlighting. For someone building a course or research agenda on AI-and-labor, this is the framework to assign.

  4. Pedagogically useful examples. The Starbucks reversal, the Armani suit, the agriculture-to-services transition, and the BLS top-quintile spending pattern are clean demonstrations of the mechanism.

  5. Related Concepts

  6. concepts/ai-as-normal-technology
  7. concepts/ai-adoption-academia

  8. concepts/agentic-ai

  9. Related Summaries

  10. summaries/thread-alexolegimas
  11. summaries/ai-normal-technology
  12. summaries/cc-series-17-zero-profit
  13. summaries/cc-series-21-attention-congestion
  14. summaries/can-ai-replace-researchers