NBER Working Paper 35238: Oracle Autopsy

URL SCAN: The Human Capital Production Function: New Estimates and Implications for Labor Supply and Taxes — Gao, Keane, Kierulf, Woodland

TEXT START: This paper estimates a learning-by-doing human-capital production function in which hours affect both current productivity and future human capital.

I. THE DISSECTION

This is a labor economist's paper doing what labor economists do: refining the machinery of a model framework that is operationally obsolete. The authors estimate a human capital production function where working more hours builds skills over time (learning-by-doing), test the Ben-Porath specification, find it statistically "weakly identified" (nearly flat likelihood surface), build a more robust alternative, and then embed it in a life-cycle model to make claims about optimal tax structure.

Core findings: Very small labor supply responses to temporary wage shocks. Optimal labor-income taxes should be flat because taxes distort not just current work but future human capital accumulation.

The intellectual structure is: sophisticated microfoundations for a tax policy recommendation.

II. THE CORE FALLACY — RELATIVE TO DT MECHANICS

The foundational error: This paper assumes human capital is the primary engine of productive economic participation across the relevant time horizon. All four authors are estimating, optimizing, and drawing normative conclusions from a framework where human capital accumulation matters because human labor is the bottleneck input.

The discontinuity the paper cannot see: When AI achieves durable cost-performance superiority over human cognitive labor — which is exactly what P1 of the DT framework describes — the entire human capital production function becomes economically marginal. These authors are optimizing the marginal tax rate on wages in a world where wages cease to be the primary distribution mechanism.

The paper's own conclusion is paradoxically self-undermining: if optimal taxes are flat because they distort "future human capital accumulation," the logic implies that human capital accumulation is the key variable. But the DT thesis says that variable gets zeroed out by automation. You cannot optimize a tax schedule around a variable that becomes structurally irrelevant.

This is not a minor oversight. It is the entire epistemic structure of the paper built on an assumption that the DT explicitly falsifies.

III. HIDDEN ASSUMPTIONS SMUGGLED IN

1. Human labor remains the marginal unit of production. The model embeds a life-cycle framework where productivity is a function of accumulated human capital. No pathway for AI-driven displacement is modeled or considered.

2. The Ben-Porath framework remains the right ontology. The paper spends significant technical effort improving identification of the Ben-Porath specification — an enormous investment in refinement. But refinement of a flawed framework is precision without accuracy.

3. Policy relevance persists through the transition. The paper concludes with tax policy implications. This assumes there will be meaningful labor-income tax revenue to optimize in a world where human labor income is collapsing.

4. PSID data remains the right sample. The Panel Study of Income Dynamics tracks humans. If the relevant economic subjects become AI systems, the PSID is a database of entities being rendered economically irrelevant.

5. Small elasticities are a stable empirical result. The paper finds very small labor supply responses. In a world of stable technology, this might be robust. In a world of rapid AI capability improvement, "small elasticities" is a snapshot of a transitional moment before a phase change.

IV. SOCIAL FUNCTION

This is Prestige Economics — Technical Elegance Theater. It is the kind of paper that receives citations, conference slots, and job market status because it is technically sophisticated, uses recognized datasets, and reaches a conclusion (flat taxes) that is intellectually tidy.

Its social function is to keep the professional conversation inside the existing framework — to keep labor economists publishing in labor economics journals about labor supply elasticities while the structural conditions that make labor economics relevant are eroding beneath them.

It is also, inadvertently, an institutional inertia mechanism: the more "rigorous" the analysis of human capital, the more legitimizing the existing academic structure, the harder it is to see the discontinuity it cannot model.

V. THE VERDICT

What the paper is really doing: Refining the microfoundations of a tax optimization model for a world where the tax base (human labor income) is being structurally eliminated.

What it cannot see: That its own finding — that optimal taxes should be flat to protect human capital accumulation — is an epitaph for the very variable it is trying to protect. When AI eliminates the need for human capital accumulation as a productive input, the tax optimization problem becomes moot.

The structural judgment: This is high-quality economics within an obsolete frame. The authors have done rigorous work on a question that will not survive the next decade in its current form. The Ben-Porath model is being refined at the moment it becomes historically irrelevant.

The paper is an autopsy performed on a patient who is still breathing, by doctors who do not know the disease is terminal.

Mechanism of Obsolescence: The entire framework assumes human capital is the primary production input. AI severs this assumption at the structural level. The tax conclusions, the elasticities, the identification strategy — all operationally irrelevant in a post-DT-transition economy.

Lag-Weighted Timeline: Mechanical Death — irrelevant within 10 years. Social Death — academia will continue citing this for 20-30 years as the professional class protects its epistemic investment.

Viability Scorecard: 1yr: Conditional (technically impressive), 2yr: Conditional, 5yr: Fragile, 10yr: Terminal (the question itself disappears)

Survival Path: This is not a viable framework for navigating the DT transition. The authors need a new research program, not a better identification strategy for Ben-Porath.