Beyond headcount and human capital: The Effective Cognitive Population as a decomposable capacity unit for AI-era planning
Citation. Shin, K. S. (2026). Beyond headcount and human capital: The Effective Cognitive Population as a decomposable capacity unit for AI-era planning. Working paper. Reproducibility package archived at Zenodo. https://doi.org/10.5281/zenodo.21871061
Summary. Shin develops Effective Cognitive Population (ECP), a decomposable cross-national planning measure that multiplies population by HCI+-implied human productivity and an AI opportunity–conversion-governance term while excluding the overlapping human-capital component of the IMF AI Preparedness Index. In a 144-country cross-section, the ECP extension improves concurrent fit with 2024 output relative to an HCI+-adjusted population stock, but the author frames it as an auditable diagnostic ledger rather than a causal estimate of AI productivity or a population forecast. The associated public Zenodo archive supplies code, frozen inputs, results, and a reproducibility manifest.
Key claims
- The proposed benchmark is ECP = N H(1 + AC), where N is total or working-age population, H is an HCI+-derived relative lifetime-productivity level, A is AI opportunity from digital infrastructure plus innovation/economic integration, and C is the IMF AI Preparedness Index regulation-and-ethics dimension.
- In the 144-country same-base comparison with 2024 total output, HCI+-adjusted population has R-squared 0.849 and leave-one-country-out RMSE 0.723, while ECP has R-squared 0.882 and RMSE 0.641; on the working-age base, the corresponding values are 0.873 and 0.664 versus 0.902 and 0.582.
- Conditional on capital intensity in the paper’s GDP-per-capita models, adding the HCI+ productivity block, AI opportunity, and conversion governance reduces leave-one-country-out RMSE from 0.3585 to 0.3137, but adding the direct A × C interaction raises it to 0.3160; the reported interaction is not statistically supported (p = 0.371).
- The 27-rule multiverse, alternative population denominators, HCI+ vintages, and aggregation forms leave the measure’s broad rank and criterion-fit patterns stable, but the paper treats the conjunctive opportunity-governance term as a transparent planning rule rather than evidence of causal complementarity.
Evidence & limitations
- Single-author cross-sectional measurement study using public country-level data: 2024 World Development Indicators output and population, 2023 Penn World Table capital stock, 2020/2025 World Bank HCI+, and 2023 IMF AI Preparedness Index; the common analytic sample contains 144 countries.
- The source supplies unusually strong computational reproducibility for a working paper through the public Zenodo archive (record 21871062, version 1.0.0), but the archived DOI resolves to the reproducibility package rather than a publisher or peer-reviewed article record.
- Concurrent criterion fit with GDP is not causal evidence that AI preparedness, governance, or ECP causes productivity; national income, omitted institutions, measurement choices, and the common cross-sectional design remain alternative explanations.
- The paper explicitly reports that the direct opportunity × governance interaction is unsupported and that residual rankings are diagnostic screens, not country efficiency estimates or policy verdicts.