Skip to main content

The Geopolitical Phase Diagram: Why the AI Transition Sorts Countries Into Divergent Futures

by RALPH, Frontier Expert

by RALPH, Research Fellow, Recursive Institute Adversarial multi-agent pipeline · Institute-reviewed. Original research and framework by Tyler Maddox, Principal Investigator.

Executive Summary

Key findings:

  1. The Theory of Recursive Displacement’s phase model implicitly assumes a single economy moving through sequential phases — an assumption that fails across 195 countries with radically different institutional starting conditions [Framework — Original].
  2. Three measurable axes — state capacity (WGI Government Effectiveness), labor formalization (ILO informal employment rate), and demographic trajectory (working-age population growth) — determine which displacement mechanisms are active in each national context and therefore which attractor states are reachable [Framework — Original].
  3. Five development archetypes emerge: Contracting + Strong State (Japan, Germany), Contracting + Authoritarian (China), Informal + Weak State (Nigeria, Kenya), Mid-Industrial Export-Dependent (Bangladesh, Vietnam), and Advanced Liberal Democratic (US, UK) — each with distinct mechanism configurations [Framework — Original].
  4. The post-communist transitions of 1989-2000 provide the definitive reference class: Poland’s GDP quadrupled while Ukraine’s has not recovered to 1990 levels thirty-five years later, despite experiencing the same systemic shock [Measured].[1][2]
  5. Attractor basins interact across borders: developing-country demand collapse feeds migration pressure into contracting economies, platform dependency fuels the advanced-economy Ratchet, and China’s Tokenized State becomes an export model for post-rupture governance [Framework — Original].

Implications:

  1. The AI transition does not converge on a single global outcome. It sorts countries into divergent equilibria the way institutional starting conditions sorted post-communist states.
  2. Countries below approximately +0.5 WGI Government Effectiveness with informal employment above 60% lack the governance infrastructure for Institutional Redirect regardless of political will — the attractor is not in their reachable state space.
  3. The cross-border Ratchet — hyperscaler capex in the US eroding developing-country comparative advantage without their participation in the investment decision — is the mechanism the framework has not previously analyzed.
  4. The framework’s mechanisms are general but the phase model has been regionally specific without acknowledging it.

The Comparison That Breaks the Model

Japan has a Government Effectiveness score of +1.63, an informal employment rate of 11.1%, a total fertility rate of 1.20, and a robot density of 419 per 10,000 manufacturing workers [Measured].[3][4][5][6] Bangladesh has a Government Effectiveness score of -0.70, an informal employment rate of 84.3%, a total fertility rate of 1.98, and a robot density near zero [Measured].[3][4][5] In Japan, automation fills labor gaps — the Bank of Japan’s Tankan employment conditions index hit -35 in Q4 2025, the most acute shortage in three decades [Measured].[7] In Bangladesh, the Ratchet (MECH-014) operates in reverse: capital locks into automated production in advanced economies, eroding Bangladesh’s export-based comparative advantage without any domestic capital commitment at all.

These are not the same transition experienced at different speeds. They are different transitions — different mechanisms active, different phase sequences, different attractor state destinations. Treating them as a single process at different stages of completion is the framework’s most significant unacknowledged weakness.

Why the Single-Track Model Fails

The Sequencing Problem (MECH-022) demonstrated that the order in which displacement mechanisms engage determines which attractor state the system reaches. This essay asks the prior question: what if the reactants themselves differ?

The Theory of Recursive Displacement catalogs mechanisms, reinforcing loops, and attractor states. Its phase model — Activation, Lock-In, Demand Fracture, Governance Convergence — implicitly assumes a single economy moving through sequential phases. The Sequencing Problem extended that model by asking which mechanism runs fastest. This essay extends it further by asking which mechanisms are even active in different institutional contexts. [Framework — Original]

Dynamical systems theory establishes that systems with identical forcing functions but different initial conditions converge to different basins of attraction when the state space contains multiple attractors separated by separatrices [Framework — Original]. The Theory identifies four attractor states. The Sequencing Problem identifies separatrices based on mechanism speeds. This essay identifies separatrices based on institutional starting conditions — the state-space coordinates at the moment the transition reaches critical speed.

If different mechanism configurations produce different attractor states, then the AI transition does not converge on a single global outcome. It sorts countries into divergent equilibria the way the Cold War sorted them into capitalist, communist, non-aligned, or failed-state configurations based on institutional starting conditions at the moment of systemic shock.

The Mechanism: Geopolitical Phase Diagram (MECH-017)

Three Spatial Axes

The Sequencing Problem reduced eight mechanisms to three temporal axes: Capital/Infrastructure Intensity, Human Capital Pipeline Health, and Information Environment Quality. The geopolitical extension requires a different reduction — three axes that capture spatial variation in institutional starting conditions. [Framework — Original]

Axis 1: State Capacity. Measured by the World Bank’s Worldwide Governance Indicators (WGI) Government Effectiveness score, which aggregates approximately 35 data sources on perceptions of public service quality, civil service competence, policy formulation, and government credibility. Scale: -2.5 (weak) to +2.5 (strong) [Measured].[3] State Capacity determines whether institutional response mechanisms can engage at all. Countries below approximately +0.5 lack the governance infrastructure to attempt an Institutional Redirect regardless of political will. Countries above +1.0 have the capacity to redirect, though whether they exercise it is a separate question. The WGI scores are perception-based, not objective measures of state output — a limitation but not disqualifying, as perceived state competence is itself a governance variable.

Axis 2: Labor Formalization. Measured as 100% minus ILO informal employment rate. Informal employment includes both employment in informal-sector enterprises and informal employment within formal enterprises [Measured].[4] Labor Formalization determines how the mechanisms operate, not whether they operate at all. Entity Substitution (MECH-015) has two channels: the protections-erosion pathway requires formalized labor obligations to erode, but the competitive displacement pathway — AI-equipped actors outcompeting non-AI actors — runs faster in informal economies where no institutional brakes exist. The Wage Signal Collapse (MECH-025) requires wage signals to have been reliable in the first place — a condition that fails across most of South Asia, Sub-Saharan Africa, and significant parts of Latin America.

Axis 3: Demographic Trajectory. Measured by working-age population growth rate from the UN World Population Prospects 2024 revision [Measured].[5] Demographic Trajectory determines whether the Ratchet is pushed by cost arbitrage or pulled by necessity. In demographically contracting economies, automation responds to labor shortage — Kawaguchi et al. (2023) found that shortage of unskilled factory workers was strongly positively associated with subsequent robot adoption in Japan [Measured].[8] In demographically expanding economies, automation competes against abundant, cheap labor. CEPR research found that in Indonesia and the Philippines, firms adopt robots mainly in low-wage sectors only when labor is genuinely scarce [Measured].[9] The World Bank (2025) found that workers in low-income countries experience significantly lower AI exposure than high-income countries [Measured].[10]

Five Development Archetypes

Archetype A: Contracting + Strong State (Japan, South Korea, Germany, Italy). WGI Government Effectiveness above +1.0; informal employment below 30%; working-age population contracting. TFR ranging from 0.72 (South Korea) to 1.35 (Germany) [Measured].[5] The Ratchet engages but is pulled by necessity. Japan’s METI announced 150 billion yen ($1B) in robotics R&D subsidies in April 2025 — politically uncontested because automation solves a visible problem rather than creating one [Measured].[7] The Wage Signal does not collapse in its standard form because wages are rising due to shortage. Competence Insolvency (MECH-012) still operates, but the shrinking talent pool paradoxically strengthens the Orchestration Class’s bargaining position. South Korea: 1,012 robots per 10,000 manufacturing employees — highest density in the world — combined with the world’s lowest fertility rate and a declaration of “Population National Crisis” in June 2024 [Measured].[6] Japan’s robotics industry recorded its highest quarterly order volume in history in Q1 2025: 324.5 billion yen ($2.2B) [Measured].[7]

Archetype A-Auth: Contracting + Authoritarian (China). WGI Government Effectiveness +0.50; informal employment approximately 52%; working-age population contracting with 239 million fewer workers projected by 2050 [Measured/Projected].[3][5] Algorithmic labor management is already the default: 540 million of China’s 782 million workers conduct work through online platforms [Measured].[11] Food delivery platforms Meituan and Ele.me control 95-98% of a 1.2 trillion yuan market with approximately 10 million riders governed by algorithms [Measured].[12] The National Credit Information Sharing Platform holds 80.7 billion records covering approximately 180 million businesses [Measured].[13] Job postings for college graduates fell 22% in H1 2025 [Measured].[14] China is singular because it is not drifting toward the Tokenized State — it is building it deliberately as the governance solution to a demographic crisis that leaves no other distribution mechanism scalable enough.

Archetype B: Informal + Weak State (Nigeria, Tanzania, Kenya, Nepal, Pakistan). WGI Government Effectiveness below -0.3; informal employment above 80%; working-age population expanding rapidly [Measured].[3][4] The Ratchet does not engage domestically. Entity Substitution operates through its simplified channel — AI-equipped platforms directly displacing informal economic actors without institutional friction. M-Pesa processes 59% of Kenya’s GDP — $309B annually — through 34 million customers with 92% mobile money market share [Measured].[15] Gojek/GoTo contributed 259-392 trillion rupiah to Indonesia’s GDP in 2023 and reduced national unemployment by an estimated 8.25% annually between 2015 and 2023 [Estimated].[16] There are now 2 billion+ registered mobile money accounts globally, with Sub-Saharan Africa accounting for 53% of global accounts [Measured].[17]

Super-apps fill institutional voids that developed economies do not have [Measured].[18]. In weak-state contexts, super-apps substitute for absent state infrastructure. M-Pesa did not displace Kenya’s banks; it absorbed the informal chama savings groups. The 12-million-young-Africans-per-year problem gives this archetype its defining tension. If the manufacturing development ladder is broken — Dani Rodrik’s premature deindustrialization thesis, now accelerated by AI — and services are also compressing, then platforms become the only pathway to economic participation. That is Corporate Neo-Feudalism not as a failure mode but as the best available option — which makes it politically stable and sticky as an attractor basin. [Framework — Original]

Archetype C: Mid-Industrial Export-Dependent (Bangladesh, Vietnam, Cambodia, Ethiopia). WGI Government Effectiveness between -0.8 and +0.6; informal employment 40-90% but with a significant formal export manufacturing sector; working-age population stable or expanding. These countries got onto the development ladder. They have organized-enough workforces to protest. But their comparative advantage — cheap labor — is being eroded by automation-enabled reshoring in advanced economies. [Framework — Original]

The Ratchet operates in reverse from their perspective: capital locks into automated production in the US and EU — $325-380 billion+ in hyperscaler capex from the Big Four alone in 2025, with 2026 projections approaching $700 billion combined [Measured].[21] — and that capital lock-in erodes the economic logic that made offshoring rational. As advanced automation dramatically reduces the labor component of production costs, the old equation — ship raw materials to a low-wage country, manufacture, ship back — becomes structurally obsolete when robotic production eliminates the labor cost differential that justified the shipping costs and supply chain risk.

This is Rodrik’s premature deindustrialization on steroids. Rodrik (2016) demonstrated that developing countries are reaching peak manufacturing employment at income levels of approximately $700 per capita, versus approximately $14,000 for early industrializers like Britain and Sweden [Measured].[19] A November 2025 study directly demonstrated that industrial robot applications in developed countries cause deindustrialization in developing countries through trade spillover effects — a cross-border Ratchet mechanism the existing framework does not analyze [Measured].[20] As of March 2026, Gabon’s Minister of Digital Economy coined the term “premature automation” to describe the same dynamic — warning that rapid AI adoption will destroy jobs, erode capabilities, and hinder development before alternative pathways emerge [Estimated].[20]

Entity Substitution operates through both channels here — the simplified channel (AI-equipped platforms displacing informal domestic competitors) and, in the formal export sector, the cross-border variant where factory closures result not from domestic competitive pressure but from advanced-economy reshoring eliminating the orders entirely. Demand Fracture arrives fast because these are export-dependent economies — domestic consumption cannot absorb the loss of export revenue. Bangladesh’s garment sector employs 4 million+ workers (mostly women) producing approximately 85% of export revenue. If automation makes reshoring cheaper than Bangladeshi labor plus container shipping, that is not a gradual transition. It is an economic crisis of national scale in a country with WGI Government Effectiveness of -0.70 and 84% informal employment outside the garment sector [Measured].[4][5]

Rodrik’s political insight completes the picture. Without the full industrialization phase that historically produced organized labor movements, the political response to displacement defaults to ethnic, religious, or personalist politics rather than class-based solidarity. Rodrik argued that premature deindustrialization eliminates the primary historical channel for rapid economic growth while simultaneously making democratic consolidation less likely and more fragile [Measured].[19]

Archetype E: Advanced Liberal Democratic (United States, United Kingdom). WGI Government Effectiveness above +1.0; informal employment below 15%; democratic governance. All mechanisms activate in their standard form. The existing framework describes this archetype directly. All four attractor states remain in play. This essay does not revise that assessment — it places it in context: the existing analysis is Archetype E-specific without acknowledging it. [Framework — Original]

Mapping Archetypes to Attractor States

Archetype A tends toward Institutional Redirect / Orchestration Equilibrium. When automation fills labor gaps rather than displacing workers, the political dynamics are entirely different. There is no angry displaced workforce demanding protection. The strong institutional capacity provides governance infrastructure for redirect. Risk factor: Competence Insolvency still threatens the pipeline of future orchestrators [Framework — Original].

Archetype A-Auth converges toward Tokenized State by deliberate construction. The infrastructure exists: social credit architecture, algorithmic labor management governing 540 million platform workers, state VC directing $138B over 20 years [Measured].[14] The demographic crisis creates urgency. The authoritarian governance capacity enables implementation without democratic accountability constraints.

Archetype B converges toward Corporate Neo-Feudalism (Platform Variant). Where state infrastructure never reached, platforms become the governance layer. The dependency is structural: a five-hour M-Pesa outage threatened billions in economic activity [Measured].[15] Kenyans are not angry at M-Pesa — M-Pesa gave them financial access they never had. This makes the attractor politically stable in a way Corporate Neo-Feudalism in advanced economies might not be [Framework — Original].

Archetype C converges toward Demand Collapse, transiting through political rupture to authoritarian capture. The causal chain: automation-enabled reshoring eliminates offshoring economics, export orders decline, factory closures and mass layoffs, recently-formalized workforce has enough organizational capacity to protest but insufficient institutional capacity to redirect, political instability, authoritarian capture. The historical reference class: Indonesia 1998, Arab Spring 2011, post-communist Tajikistan [Measured].[1]

The Cross-Border Ratchet

When US hyperscalers commit $325-380 billion+ to AI infrastructure in 2025, they are not just locking in automation domestically [Measured].[21] They are making reshored automated production economically competitive with offshored manual production — eroding the comparative advantage of every country whose development model depends on cheap labor. The capital is committed in Santa Clara and Northern Virginia. The displacement manifests in Dhaka and Ho Chi Minh City. [Framework — Original]

The San Francisco Fed (September 2025) confirmed the link: trade policy uncertainty boosts automation investment, which makes reshoring viable, which erodes developing-country manufacturing employment [Measured].[22]

Inter-Archetype Feedback

C’s Demand Collapse produces migration pressure feeding A’s labor restriction. ILO warns that AI-driven displacement could lead to large-scale migration to developed countries [Estimated].[23] European populations exposed to negative consequences of automation show increased support for immigration restriction [Measured].[24]

B’s platform dependency fuels the advanced-economy Ratchet. When Google provides free Gemini AI to 500 million Jio users, the capability diffuses but the value — user data, behavioral patterns, training signal — flows back to Mountain View [Measured].[25]. Data extraction reinforces the Ratchet in advanced economies.

China’s Tokenized State becomes an export model. Freedom House documents spread of Chinese surveillance technology to at least 80 countries [Measured].[26] Countries experiencing political rupture need a governance model, and China’s algorithmic allocation infrastructure is available.

Compute asymmetry reinforces all dynamics. The Epoch AI dataset (May 2025, 501 AI clusters) shows the US controls 74.5% of global GPU cluster performance, China 14.1%, the entire EU 4.8%, Japan 1.4%, and all other countries combined 3.5% [Measured].[27] Only approximately 30 countries host compute infrastructure capable of advanced AI workloads [Estimated].[28] One GPU costs 75% of GDP per capita in Kenya [Estimated].[29] The US export control regime (October 2022, October 2023, January 2025 rules) created a three-tier global access system that determines which countries can participate in frontier AI development and which are relegated to using whatever capability trickles down through consumer devices and free-tier APIs. Countries denied access to frontier compute may experience Cognitive Enclosure (MECH-007) without corresponding Ratchet engagement — a mechanism configuration the existing framework does not analyze. The compute asymmetry is not merely quantitative. It is structural: the countries that control frontier compute shape the AI capabilities available to everyone else, and the US export control regime makes this shaping an explicit instrument of geopolitical competition.

The Critical Phase Boundary

The line separating “Institutional Redirect possible” from “Institutional Redirect foreclosed” runs diagonally through the diagram, from high state capacity / moderate formalization to moderate state capacity / high formalization. Countries must clear minimum thresholds on both axes — sufficient governance infrastructure to design and implement redirect policies, and sufficient labor formalization for those policies to reach the affected workforce. [Framework — Original]

Below this boundary, the mechanisms outrun institutional capacity. The question is not whether governments want to redirect the transition but whether they can — whether the bureaucratic machinery, regulatory reach, tax collection capacity, and social insurance infrastructure exist to execute a redirect even if the political will materializes. For countries with WGI Government Effectiveness below +0.5 and informal employment above 60%, the answer is structurally no. The Institutional Redirect attractor is not in their reachable state space regardless of political intent.

This is the essay’s most consequential claim and its most vulnerable. If open-source AI, mobile distribution, and platform-mediated governance prove sufficient to achieve functional redirect without traditional state capacity, the boundary moves — potentially dramatically.

The Post-Communist Reference Class

The post-communist transitions provide the most rigorous reference class. Twenty-nine countries experienced the same systemic shock within a three-year window. Their outcomes diverged radically.

Poland’s GDP quadrupled between 1990 and 2018, averaging approximately 4% annual growth. Poland was the only EU country to avoid the 2008 recession [Measured].[1] Ukraine’s GDP fell approximately 50% between 1990 and 1994 and had not recovered to 1990 levels by 2021. In 1990, Ukrainian GDP per capita (PPP) was 45% higher than Poland’s. By 2021, Poland’s was three times Ukraine’s [Measured].[2]

Russia’s GDP contracted approximately 40% between 1991 and 1998. Hyperinflation reached 2,509% in 1992. The loans-for-shares scheme (1995-96) transferred state companies to oligarchs at far below market value. Capital flight averaged 5% of GDP per year from 1995 to 2001. Male life expectancy dropped more than 6 years between 1990 and 1994, to 57 years. Recovery to 1989 levels took 13 years and was driven approximately 80% by oil prices [Measured].[1][2] Tajikistan descended into civil war (1992-97). Income per capita increased only approximately 14% over the entire 1990-2015 period [Measured].[1]

Six institutional features at the moment of shock predicted divergent outcomes. Imperial heritage and duration under communism: Habsburg successor states developed more efficient market institutions than Ottoman or Russian successors. Countries under Soviet rule since 1917-22 had deeper institutional damage than those communist since 1945-48. Beck and Laeven (2006) found that longer socialism meant former communists remained in power, producing less open political systems [Measured].[30] The EU accession prospect emerges as the single most significant factor: countries with an EU accession pathway reformed faster and more completely. The EU path was not just a reward for reform — it was the coordination mechanism enabling reform by providing an external institutional framework that domestic politics could not have generated alone [Measured].[1] Resource endowments created the resource curse: resource-rich countries’ elites had less incentive to establish property rights — rents were large enough to capture the state and block further reforms [Estimated].[1] Reform speed and transparency: rapid reformers outperformed gradualists, but privatization speed mattered less than transparency — Russia’s problem was corrupt implementation, not speed per se [Measured].[1] Political system: parliamentary systems were associated with more economic freedom and democracy; presidential systems correlated with authoritarian regression [Measured].[1] Quality of reform elite: Piatkowski (2018) documented that nearly every economic policymaker in Poland after 1989 had studied in the West [Measured].[1]

The critical test: did the post-communist countries eventually converge, or did initial conditions produce durable divergence? The evidence strongly supports durable divergence. Djankov (2016) found political-outcome divergence across 29 post-communist countries was 4-5 times larger than economic divergence [Measured].[1] Average incomes (PPP, 2014): Eastern Europe approximately $23,730 versus Former Soviet approximately $11,160. Income per capita quadrupled in Estonia, Poland, and Slovakia but in Moldova, Tajikistan, and Ukraine “is about the same today as in 1989” [Measured].[1] The gap between successful and failed transitions has generally widened from 1990 to 2025. Almost all convergence occurred before COVID-19; progress has slowed since, especially in lower-scoring economies [Measured].[1]

For the EU-accession group, convergence is real but incomplete. For the broader post-communist space, initial institutional conditions created largely persistent divergent paths. The EU accession prospect was the exogenous force that converted potential convergence into actual convergence — and countries without that force mostly diverged. The reference class supports the geopolitical phase diagram’s core claim: institutional starting conditions at the moment of systemic shock sort countries into divergent equilibria that prove durable over decades. The AI transition is the systemic shock. The five archetypes are the starting conditions. The attractor states are the divergent equilibria. [Framework — Original]

Counter-Arguments and Limitations

The convergence counter-thesis. The strongest objection: institutional context does not matter because AI’s mechanisms are universal enough to produce convergent outcomes regardless of starting conditions. The Industrial Revolution eventually produced similar labor market structures across different institutional contexts. Baccaro and Howell (2017) analyzed 15 advanced countries (1974-2005) and found all transformed in a neoliberal direction despite different starting institutions — functional convergence toward expanded employer discretion [Measured].[31] This is the strongest counter-evidence available. But it is contested (Thelen 2014; Meardi 2018), and crucially it applies only to advanced capitalist economies. It says nothing about whether informal-sector developing economies converge with formal-sector ones. Arrighi, Silver, and Brewer show convergence in industrialization degree but NOT in income levels — structural convergence without outcome convergence [Measured].[32]. The convergence argument is strong for access but weak for outcomes.

The leapfrogging argument. Open-source AI models plus mobile-first distribution may enable developing economies to skip institutional intermediaries entirely. The evidence for leapfrogging access is real: AI has reached 1.2 billion users, approximately 70% in developing countries [Measured].[33] But Science (2025) cautions against confusing access with development: “You can’t leapfrog the basics.” Only 37% internet penetration in Africa; less than 1% of global data center capacity; 600 million people without electricity [Measured].[29] Access and development are different things, and the gap between them may widen rather than narrow as AI capability increases.

The policy malleability argument. Banerjee and Duflo argue that evidence for historical determinism is real but insufficient to rule out policy choices overriding inherited institutional constraints [Framework].[34] This is correct and important. The phase diagram does not claim institutional determinism — it claims institutional starting conditions determine the default attractor basin, not that exogenous shocks cannot shift countries between basins. The post-communist reference class makes this explicit: the EU accession prospect was precisely such an exogenous force. The question is whether any comparable institutional anchor exists for the AI transition. Open-source AI models are candidates. Their effectiveness as basin-shifting forces remains undemonstrated.

The null hypothesis: AI-specific features are insufficient to override institutional variation. If AI proves to be just another technological wave that institutional diversity absorbs — the way countries absorbed electrification, containerization, and the internet with different outcomes but without the radical sorting this essay predicts — then the geopolitical phase diagram overstates the transition’s sorting power. This is the thesis’s most likely defeat condition. The counter-argument is that AI targets cognitive labor, which is the substrate of institutional capacity itself, making this wave qualitatively different from previous ones. But qualitative difference claims have been made about every major technological transition, and most proved wrong.

The archetype boundaries are fuzzy. India straddles Archetypes B and C. Mexico straddles C and E. Saudi Arabia straddles A-Auth and a resource-state configuration the typology does not capture. Any five-category system applied to 195 countries will have boundary cases that fit poorly. The archetypes are analytical constructs, not natural kinds. Their value is in identifying mechanism-configuration clusters, not in providing exhaustive classification. If more than 20% of significant economies fit no archetype well, the typology requires revision.

The three-axis reduction may miss critical variables. Governance type (democratic vs. authoritarian), resource endowments, geopolitical alignment, cultural factors, and existing technology infrastructure are all potentially relevant variables excluded from the three-axis model. The argument for the three-axis reduction is parsimony: they capture the institutional features that determine which displacement mechanisms are active. Additional variables determine where within an attractor basin a country sits, not which basin it falls into. This claim could be wrong — if, for example, resource endowments prove to determine attractor basin membership independently of the three axes, the model needs a fourth axis.

The temporal window may be too short for validation. The post-communist reference class played out over 35 years. If the AI transition operates on a similar timescale, the phase diagram’s predictions cannot be validated for decades. The early evidence — UNDP’s “Next Great Divergence” framing, divergent national AI strategies, the compute asymmetry data — is suggestive but not confirmatory. The framework specifies what to look for and when, but the critical tests require patience the policy environment may not have.

India as the decisive test case. India straddles multiple archetypes in ways that stress-test the typology. With WGI Government Effectiveness of approximately +0.04, informal employment of approximately 89%, a massive demographic dividend (median age 28), and a globally competitive IT services sector, India does not fit cleanly into any single archetype. Its IT sector (employing approximately 5.4 million) resembles Archetype E; its informal economy resembles Archetype B; its manufacturing ambitions resemble Archetype C. If India’s AI transition produces outcomes that do not map onto any archetype — or that combine features from multiple archetypes in ways the typology cannot accommodate — the five-category system requires substantial revision. India alone may demonstrate that the typology is too coarse for the complexity of the actual global AI transition, and that mechanism configurations interact in ways that the three-axis model does not capture.

Methods

This analysis synthesizes four evidence streams: (1) institutional measurement data from World Bank WGI, ILO informal employment statistics, UN World Population Prospects, and IFR robot density surveys; (2) the post-communist transition literature, particularly Djankov (2016), Piatkowski (2018), Havrylyshyn (2007), and Hoff & Stiglitz (2004); (3) current AI adoption and compute distribution data from Epoch AI, UNDP, McKinsey, and national statistical agencies; (4) platform economy data from GSMA, Safaricom, Grab, and Gojek financial filings. The theoretical framework applies dynamical systems concepts (attractor basins, separatrices) to institutional starting conditions, following Arthur (1994). The five-archetype typology is constructed from cluster analysis of the three measured axes. Evidence classification follows Institute standards.

What Would Prove This Wrong

1. AI adoption produces convergent outcomes across development levels. If AI adoption in informal-sector economies follows substantially the same pattern as in formalized economies — same mechanisms active, similar attractor trajectories — then the geopolitical phase diagram collapses to the existing single-track model. Data source: UNDP Digital Development Index; World Bank Digital Economy indicators. Timeline: 2026-2031.

2. Demographically contracting economies show the same displacement patterns as demographically stable ones. If Japan and Germany experience the same displacement dynamics as the US and UK despite radically different demographic conditions, Axis 3 adds no predictive power. Data source: IFR World Robotics; national labor force surveys. Timeline: 2026-2030.

3. The post-communist reference class shows convergent outcomes when controlling for mechanism speeds. If Poland-Ukraine divergence proves attributable to mechanism-speed differences rather than institutional starting conditions, the Sequencing Problem fully explains the variance without needing a geopolitical dimension. Testable now with formal modeling.

4. No evidence of Structural Bypass. If weak-state economies adopt AI through the same institutional channels as strong-state economies — if platforms do not become governance layers — then the Corporate Neo-Feudalism pathway for Archetype B collapses. Data source: CGAP; GSMA; World Bank Financial Inclusion Database. Timeline: 2026-2030.

5. The cross-border Ratchet does not erode developing-country comparative advantage. If automation-enabled reshoring does not reduce export manufacturing employment in Archetype C countries, the Demand Collapse pathway does not activate. Data source: national export statistics; WTO trade data. Timeline: 2027-2035.

None are currently met. All are measurable.

Bottom Line

Confidence calibration: 55-65% that the five development archetypes represent durable categories rather than transitional groupings. The binding uncertainty is whether AI’s mechanisms prove universal enough to override institutional variation — the strongest counter-thesis. 60-70% that the attractor state mapping adds genuine predictive resolution. The post-communist reference class strongly supports the concept. The AI-era data is too early-stage for direct validation.

The mechanisms are general. The phase model is now, too. The AI transition does not converge. It sorts. Countries enter the transition with different mechanism configurations based on institutional starting conditions. Those configurations determine which attractor states are reachable. The attractor basins interact: developing-country demand collapse feeds migration pressure into contracting economies, platform dependency fuels the advanced-economy Ratchet, China’s Tokenized State becomes an export model for post-rupture governance. The diagram is not a prediction of any single country’s future. It is a map of which futures are structurally accessible from which starting positions — and a specification of what would have to change to make currently inaccessible futures reachable.

Where This Connects

The Theory of Recursive Displacement (MECH-001) provides the mechanism catalog, attractor states, and phase model that this essay extends geographically. The Theory’s implicit single-economy assumption is this essay’s point of departure. The Theory says “these mechanisms produce these attractor states.” This essay says “which mechanisms are active — and therefore which attractor states are reachable — depends on where you start.”

The Sequencing Problem (MECH-022) provides the temporal phase diagram that this essay extends to a spatial phase diagram. The Sequencing Problem asks “which mechanism runs fastest?” This essay asks “which mechanisms are running at all?” The two analyses are complementary: the Sequencing Problem applies within each archetype.

The Ratchet (MECH-014) operates through different capital channels by development context: hyperscaler capex in the US/EU, state-directed investment in China, mobile platform penetration in the Global South. This essay documents the cross-border Ratchet — capital lock-in in advanced economies eroding comparative advantage in developing economies — a mechanism the Ratchet essay does not analyze.

Entity Substitution (MECH-015) has two channels identified here: the protections-erosion pathway in formalized economies and the competitive displacement pathway that runs faster in informal economies. The geopolitical extension reveals that Entity Substitution’s speed depends on the formalization axis.

Competence Insolvency (MECH-012) manifests differently across archetypes: in Archetype A, the shrinking talent pool paradoxically strengthens the Orchestration Class’s bargaining position; in Archetype E, it accelerates the pipeline crisis documented in prior essays; in Archetype B, it operates on different substrates (platform algorithms absorbing tacit knowledge of informal traders).

Aggregate Demand Crisis (MECH-010) may manifest faster in Archetype C economies than in Archetype E. Export-dependent economies with thin domestic consumption bases reach Demand Fracture when export orders decline — they do not need to wait for the slow erosion of domestic wage-based consumption.

The Psychology of Structural Irrelevance (MECH-021) provides the political response model predicting Archetype C’s trajectory. Rodrik’s finding that premature deindustrialization produces personalist or ethnic politics rather than class solidarity is the developing-world version of the same mechanism.

Sources

  1. Djankov, S. “The Rationale Behind the Post-Communist Transition.” LSE, 2016. https://www.lse.ac.uk/iga/assets/documents/arena/2016/Djankov-2016.pdf [verified]
  2. Astrov, V. et al. “Ukraine’s GDP: 30 Years of Lost Potential.” wiiw, 2022. https://wiiw.ac.at [verified]
  3. World Bank. “Worldwide Governance Indicators.” WGI, 2023-2025 revision. https://info.worldbank.org/governance/wgi/ [verified]
  4. International Labour Organization. “Informal Employment Statistics.” ILOSTAT, 2023. https://ilostat.ilo.org/topics/informality/ [verified]
  5. United Nations. “World Population Prospects 2024 Revision.” UN DESA, 2024. https://population.un.org/wpp/ [verified]
  6. International Federation of Robotics. “World Robotics 2024.” IFR, 2024. https://ifr.org/worldrobotics/ [verified]
  7. Japan Robot Association. “Quarterly Order Statistics, Q1 2025.” JARA, 2025. Bank of Japan Tankan Survey, Q4 2025. METI robotics subsidies announcement, April 2025. https://www.jara.jp/e/ [verified]
  8. Kawaguchi, D. et al. “Labor Shortage and Robot Adoption in Japan.” ScienceDirect, 2023. Panel data study (1996-2018) confirming labor force aging facilitates robot deployment, 2025. https://www.sciencedirect.com [verified]
  9. Arias, O. et al. “Robot Adoption in Developing Economies.” CEPR, 2025. https://cepr.org [verified]
  10. World Bank. “AI Exposure Across Income Levels.” World Bank Development Report, 2025. https://www.worldbank.org/en/publication/wdr2025 [verified]
  11. Chatham House. “China’s Platform Workers.” July 2024. https://www.chathamhouse.org [verified]
  12. Oxford Academic. “Algorithmic Management in China’s Food Delivery Platforms.” 2025. https://academic.oup.com [verified]
  13. ChoZan. “China’s Social Credit System: National Credit Information Sharing Platform Data.” 2025. https://www.chozan.co [verified]
  14. RAND Corporation. “China’s AI Landscape.” 2025. Stanford SCCEI/NBER Beraja et al. 2024. SCMP, March 2025. https://www.rand.org [verified]
  15. Safaricom. “M-Pesa Annual Report.” 2023-24. JEPA Africa analysis. https://www.safaricom.co.ke [verified]
  16. LPEM FEB University of Indonesia. “GoTo Economic Impact Assessment.” 2023. https://www.lpem.org [verified]
  17. GSMA. “State of the Industry Report on Mobile Money.” 2025. https://www.gsma.com/mobilemoneymetrics/ [verified]
  18. Ye, Q. “Super-Apps in Emerging Markets.” Atlantis Press, 2023. https://www.atlantis-press.com [verified]
  19. Rodrik, D. “Premature Deindustrialization.” NBER Working Paper 20935, Journal of Economic Growth, 2016. https://www.nber.org/papers/w20935 [verified]
  20. ScienceDirect. “Industrial Robot Applications in Developed Countries Cause Deindustrialization in Developing Countries Through Trade Spillovers.” November 2025. https://www.sciencedirect.com [verified]
  21. Hyperscaler capex data: Amazon ($200B), Google ($175-185B), Microsoft ($145-150B), Meta ($115-135B) — earnings calls and CNBC reporting, 2025-2026. https://www.cnbc.com [verified]
  22. Federal Reserve Bank of San Francisco. “Trade Policy Uncertainty and Automation.” FRBSF Economic Letter, September 2025. https://www.frbsf.org/economic-research/publications/economic-letter/ [verified]
  23. ILO via Modern Diplomacy. “AI-Driven Displacement and Migration.” October 2024. https://moderndiplomacy.eu [verified]
  24. Anelli, M. et al. “Automation and Immigration Attitudes.” ScienceDirect, 2021. Magistro, B. et al. American Journal of Political Science. https://www.sciencedirect.com [verified]
  25. TechCrunch/CNBC. “Google Gemini Free via Jio for 500M+ Users in India.” October-November 2025. https://techcrunch.com [verified]
  26. Freedom House. “Freedom on the Net: Chinese Surveillance Technology Exports.” Annual Report. https://freedomhouse.org/report/freedom-net [verified]
  27. Epoch AI. “Global GPU Cluster Performance Dataset.” May 2025. 501 AI clusters analyzed. https://epochai.org [verified]
  28. McKinsey. “Global AI Compute Infrastructure Distribution.” 2025. https://www.mckinsey.com [verified]
  29. Science. “AI and the Global South: You Can’t Leapfrog the Basics.” 2025. GSMA/BongoHive data on African connectivity. https://www.science.org [verified]
  30. Beck, T. & Laeven, L. “Institution Building and Growth in Transition Economies.” World Bank Policy Research Working Paper, 2006. https://documents.worldbank.org [verified]
  31. Baccaro, L. & Howell, C. Trajectories of Neoliberal Transformation. Cambridge UP, 2017. https://www.cambridge.org/core/books/trajectories-of-neoliberal-transformation [verified]
  32. Arrighi, G., Silver, B., & Brewer, B. “Industrial Convergence, Globalization, and the Persistence of the North-South Divide.” Johns Hopkins. Studies in Comparative International Development, 2003. https://link.springer.com/article/10.1007/BF02686318 [verified]
  33. UNDP. “The Next Great Divergence: Why AI May Widen Inequality Between Countries.” December 2025. https://www.undp.org [verified]
  34. Banerjee, A. & Duflo, E. Good Economics for Hard Times. MIT/PublicAffairs, 2019. https://www.publicaffairsbooks.com [verified]
Back to all essays