A technical error in Afonso (2024) reverses its long‑run predictions: when the balanced growth path is defined correctly, skilled labor comes to dominate and the skill premium explodes, while per‑capita growth falls back to the semi‑endogenous rate derived by Jones (1995).

Main Finding

The paper is a theoretical critique of Afonso (2024). It shows that Afonso used an incorrect definition of a balanced growth path (BGP). When the BGP is defined correctly, the long-run implications of automation differ sharply from Afonso’s conclusions: the ratio of skilled labor to total labor converges to unity, the skill premium diverges to infinity, and the long-run per‑capita growth rate equals the growth rate implied by Jones (1995) semi‑endogenous growth (i.e., automation does not drive growth to zero as Afonso claimed).

Key Points

• Diagnosis of error:
  • Afonso’s BGP specification leads to an internal inconsistency: under his characterization, automation progress implies both the skill ratio and the skill premium go to zero and eventually the economy’s growth rate becomes zero — a contradiction with his own stated conclusions.
  • The paper shows this contradiction arises from a mis-specified balanced growth condition in Afonso’s model.
• Corrected dynamics:
  • With the corrected BGP, skilled labor becomes dominant: lim_{t→∞} (L_s / L) = 1.
  • The skill premium (w_s / w_u) diverges to infinity as automation progresses.
  • Long-run per-capita growth does not collapse; instead it matches the semi‑endogenous growth rate familiar from Jones (1995).
• Interpretation:
  • Automation capital, as modeled, is complementary to skilled labor and substitutes for unskilled labor. Over time, this drives up demand and returns to skilled labor relative to unskilled labor.
  • The model’s corrected asymptotic behavior implies increasing inequality between skilled and unskilled workers and sustained growth driven by the same mechanisms that produce semi‑endogenous growth in the literature.

Data & Methods

• Nature of the work: theoretical / analytical critique (no new empirical data).
• Methods used:
  • Re-examination of Afonso’s model structure: economy with automation capital, traditional capital, skilled and unskilled labor.
  • Re-derivation of balanced growth path (BGP) conditions and steady-state/asymptotic limits.
  • Asymptotic analysis of factor shares, wage ratios, and growth rates; comparison with Jones (1995) semi‑endogenous growth results.
  • Logical consistency checks to expose the contradiction in Afonso’s original conclusions.
• Robustness / scope:
  • Focused on the steady-state/BGP implications of the specified production structure and factor accumulation equations. Results hinge on how automation capital enters production and how factor returns adjust over time.

Implications for AI Economics

• For predictions about automation and labor:
  • Automation modeled as capital that complements skilled labor implies long-run labor composition heavily favors skilled workers, not the elimination of skilled labor.
  • Expect rising skill premia and widening wage inequality unless counteracted by policy or endogenous skill acquisition.
• For growth forecasts:
  • Automation (as in the model) does not cause eventual collapse of per-capita growth; long-run growth aligns with semi‑endogenous growth predictions (growth tied to R&D/population dynamics).
  • Policy levers that affect research productivity and population/skills accumulation remain central to long-run growth outcomes.
• For policy design:
  • Emphasizes the importance of education/training and policies that facilitate skill acquisition to mitigate inequality.
  • Suggests scrutiny of taxation and redistribution systems given diverging returns between skilled and unskilled labor.
  • Points to the relevance of R&D and population policies for sustaining per‑capita growth in the presence of automation.
• For modeling AI/automation:
  • Highlights the sensitivity of long-run predictions to how automation is specified (capital vs. task replacement, complementarity vs. substitution).
  • Encourages modelers to carefully define BGPs and check asymptotic consistency when drawing policy or societal conclusions about AI-driven automation.

Limitations and next steps: - The critique is confined to the theoretical model as specified; empirical validation of the corrected implications is needed. - Extensions worth exploring: endogenous skill acquisition, task-based frameworks, non-neutral automation technical change, transitional dynamics, and heterogeneity in automation adoption across sectors.