What remains scarce after AGI? – Alex Imas and Phil Trammell
TL;DR
Alex Imas and Phil Trammell analyze what remains scarce after AGI, arguing that while a 'relational sector' where humans provide intrinsic value may persist, increasing variety in capital goods could cause labor share to collapse to zero unless we collect critical data on consumer preferences for human involvement.
🤝 The Relational Sector and Human Value 3 insights
Human involvement carries intrinsic value
The 'relational sector' comprises goods and services where human participation itself constitutes value—such as healthcare, performances, or artisanal goods—potentially remaining scarce even after full automation of other sectors.
Empirical evidence reveals preference premiums
Experimental auctions demonstrate consumers pay significantly more for human-created art than AI-created equivalents, but this premium vanishes for mass-produced human goods, indicating value derives from perceived connection rather than mere scarcity.
Humans differ from horses
Unlike horses which were valued solely for outputs, humans provide intrinsic value through interaction itself, though this only sustains wages if consumers consistently pay premiums for human-in-the-loop tasks across sufficient economic sectors.
📉 Labor Share Dynamics and Automation 3 insights
Historical stability masks potential fragility
Labor share has remained approximately 60% for centuries despite automation because network-adjusted supply chains still embed human labor and because labor and capital function as complements.
Zero labor share is theoretically possible
If network-adjusted capital share reaches 100% for automated goods while capital variety increases faster than relational goods expand, labor share could collapse to zero despite ongoing demand for human services.
Satiation creates ambiguous outcomes
Fully automating non-relational goods could drive their marginal utility toward zero faster than production rises, potentially reallocating wealth to human services—or eliminating labor income entirely if capital variety prevents satiation.
🔮 Forecasting Limits and Data Requirements 3 insights
Economists have poor predictive track records
David Ricardo correctly predicted the automation of 1820s jobs but failed to anticipate new service-sector employment, illustrating the 'lump-of-labor fallacy' and the difficulty of forecasting structural economic change.
Scenario modeling outperforms point predictions
Rather than making specific forecasts, economists should map distinct scenarios—such as labor share dropping to zero versus full employment—to identify which observable data would distinguish between them.
Critical data gaps hinder analysis
Current frameworks lack measures of consumer demand elasticities for human involvement, necessitating a 'Manhattan Project for data' including conjoint analysis of willingness to pay for automated versus human-in-the-loop services.
⚙️ Capital Variety and Satiation 3 insights
The Mongolian economist fallacy
Analogous to a 1400 Mongolian economist predicting all wealth would flow to singers once transportation was automated, modern forecasters risk underestimating how capital variety expands demand faster than satiation sets in.
Moore's Law and value decline
The economic value of transistors has effectively halved every 18 months as supply outpaced demand, but AI may reverse this by generating infinite new uses for compute, potentially maintaining or increasing capital's share indefinitely.
Compute demand defies historical trends
Unlike declining transistor value, H100 rental prices have risen despite technological progress, indicating AI is creating sufficient new demand varieties to prevent capital satiation.
Bottom Line
Policymakers should prepare for multiple economic scenarios—including potential collapse of labor share—by immediately funding large-scale data collection on consumer willingness to pay for human involvement, rather than relying on historical patterns that may not survive full supply chain automation.
More from Dwarkesh Patel
View all
Chip design from the bottom up – Reiner Pope
Reiner Pope explains how AI chips work from fundamental logic gates up, revealing that the physical cost of moving data between memory and compute units (via multiplexers) often exceeds the cost of the actual mathematical operations, and that circuit area scales quadratically with precision, making low-precision arithmetic exponentially more efficient than commonly assumed.
Building AlphaGo from scratch – Eric Jang
Eric Jang demonstrates how modern LLM coding tools and algorithmic improvements have democratized AI research, enabling a single researcher to rebuild AlphaGo for thousands of dollars rather than millions, while explaining how Monte Carlo Tree Search combined with neural networks solved a game previously considered computationally intractable.
David Reich – Why the Bronze Age was an inflection point in human evolution
Geneticist David Reich reveals that contrary to decades of evolutionary theory, natural selection has been rampant in human populations over the last 10,000 years, with the Bronze Age triggering an unprecedented acceleration in genetic adaptation to immune and metabolic challenges.
The math behind how LLMs are trained and served – Reiner Pope
Reiner Pope explains the mathematical mechanics behind LLM inference costs, demonstrating how 'Fast Mode' APIs charge premiums for smaller batch sizes that reduce latency, and why physical memory bandwidth constraints create hard limits on how fast or cheap inference can get regardless of budget.