Mapping AI Economic Complexity

TEXT START: Green economic complexity provides a generalizable framework for examining countries' productive capabilities in a defined product set.

The Dissection

This paper builds a trade-based map of AI-enabling industrial capacity. Using 103 goods and export data from 2007–2023, it measures current specialization, nearby diversification opportunities, and the complexity of those opportunities. It answers a narrow question: which countries already export goods adjacent to AI-enabling production?

It does not measure control of models, compute, capital, software, intellectual property, energy, logistics, or the rents generated by automation. Its own disclaimer is accurate: this is a preliminary capability map, not a forecast of national AI performance.

The Core Fallacy

The structural danger is a category error: treating export adjacency as AI power.

A country can manufacture or export AI-enabling components while foreign firms control the designs, platforms, capital, standards, and revenue. Assembly capacity is not sovereignty. Export growth is not productive participation. Under the Discontinuity Thesis, even a strong position in hardware and related goods does not restore the mass employment–wage–consumption circuit once cognitive automation dominates.

The paper does not explicitly make that overclaim; policymakers and investors will be tempted to make it for the paper. Its metrics identify proximity to an industrial niche, not ownership of the AI economy or protection from labor displacement.

Hidden Assumptions

• Export specialization reflects domestic capability rather than re-exporting, foreign-owned assembly, or dependence on global value chains.
• Statistical proximity between products predicts successful diversification rather than merely correlated trade patterns.
• The selected 103 goods remain economically central as AI systems evolve.
• Physical production translates into strategic control and rent capture.
• Export-share growth translates into broad national prosperity or employment.
• Trade data can capture service, software, cloud, model, and infrastructure power.
• Historical export relationships from 2007–2023 remain useful guides through a discontinuous automation transition.

The abstract itself partially acknowledges these limits by reporting that raw potential predicts changes in export share but not statistically significant changes in complexity or specialization counts. That is weak evidence of industrial momentum, not proof of a durable AI advantage.

Social Function

Partial truth with transition-management utility and prestige-signaling risk.

The dashboard can help states locate temporary niches in hardware, components, logistics, maintenance, and adjacent manufacturing—the physical perimeter of the AI system. But when presented as a national AI-performance index, it becomes elite reassurance: a map of where to stand near the machine, mistaken for evidence of control over it.

The Verdict

This is a useful industrial adjacency map and an inadequate map of AI power. Japan’s AECI lead and China’s AECP lead may identify different positions in the emerging supply chain, but neither result disproves the Discontinuity Thesis. The paper locates potential carcass-management niches; it does not show that mass labor remains necessary, that wages remain the distribution mechanism, or that exporting AI-enabling goods grants control of AI capital.

Its value is tactical. Its use as evidence that the post-WWII economic order can survive is false.