The Machine Replaces the Apprentice: Artificial Intelligence, Robotics, and the Macroeconomic Crisis of Wage-Based Civilization

Full report

On this page

Introduction: The Paradox of Expanding Wealth and Contracting Labor

Modern political economy rests on a foundational premise: technological innovation enhances capital productivity, elevates total factor output, and, through competitive labor markets, distributes the resulting economic surplus via wages to the broader populace. This wage-based income distribution mechanism has historically transformed technological productivity gains into broad-based consumer demand, ensuring economic stability and socio-political cohesion. However, the rapid deployment of generative artificial intelligence, agentic cognitive systems, and advanced industrial robotics has introduced a severe structural rupture into this architecture. The core dilemma facing contemporary market economies is whether a wage-based civilization can remain stable when technology dramatically expands total wealth creation while simultaneously reducing the volume of human labor required to generate it. While aggregate output indicators exhibit continued expansion and productivity frontiers shift outward, administrative payroll records, occupational surveys, and capital deployment metrics reveal a structural decoupling. Capital is increasingly capturing the national income surplus, while entry-level employment opportunities in knowledge-intensive and routine physical sectors are undergoing contraction. This transformation does not present itself as immediate, catastrophic mass unemployment across the entire labor force1. Instead, it operates through nuanced, highly targeted structural mechanisms: hiring avoidance rather than headline mass layoffs, localized entry-level displacement, and the systematic erosion of the traditional organizational apprenticeship ladder1. Understanding the long-term viability of wage-centered macroeconomic models requires a comprehensive evaluation of early-career employment data, task-substitution frameworks, industrial physical automation rates, capital ownership concentration, and aggregate demand dynamics1.

The Fracturing Apprenticeship Ladder: Entry-Level Disruption and Hiring Avoidance

Empirical Evidence from Administrative Payroll Data

The empirical reality of AI-driven labor market disruption is clearest at the point of early-career entry. Comprehensive investigations conducted by the Stanford Digital Economy Lab, utilizing longitudinal administrative payroll records from ADP covering over 25,000 firms, 4.6 million individual workers, and 730 distinct occupations, provide rigorous descriptive metrics on labor market behavior since the late-2022 commercialization of generative AI1. While macro-level headline metrics show no widespread, economy-wide job displacement across the entire working population1, a profound structural divergence has emerged among young workers aged 22 to 251. As of mid-2026, employment among 22-to-25-year-olds in highly AI-exposed occupations sits approximately 19% below the level it would have attained had it tracked the trajectories of same-aged peers in low-exposure roles1. This relative shortfall has expanded monotonically over consecutive observation vintages, widening from a 15% deficit in mid-2025 to 19% by August 20261.

In absolute level terms, employment for workers aged 22 to 25 in the two most AI-exposed occupational quintiles fell by roughly 11% between November 2022 and June 20261. Conversely, employment for the same age cohort across the three least-exposed quintiles expanded by approximately 10% over the identical timeframe1. Experienced workers aged 26 to 70 in high-exposure roles exhibit no comparable employment gap, demonstrating that the structural shock of modern automation is heavily concentrated at the entry threshold of the labor market1.

Mechanics of Substitution: Codified Knowledge versus Tacit Experience

The microeconomic mechanism driving this selective labor divergence lies in the operational distinction between codified knowledge and tacit experience1. Generative artificial intelligence systems excel at processing, reproducing, and synthesizing formal, standardized, and procedurally documented information—the exact domain of codified knowledge1. Entry-level positions in knowledge-intensive fields such as software engineering, legal research, financial analysis, customer support, and administrative services predominantly comprise tasks built upon codified knowledge1. Conversely, experienced workers possess tacit knowledge—contextual judgment, intuition, relational navigation, complex political negotiation, and unstructured problem-solving acquired through years of physical practice and real-world exposure1. Generative model architectures remain structurally ill-equipped to replicate tacit knowledge without human oversight1. Consequently, enterprise deployment strategies leverage AI models to automate the routine, codified tasks historically assigned to entry-level professionals, while utilizing the technology as a productivity multiplier for senior personnel who supply the requisite tacit oversight1. This dynamic fundamentally alters corporate human capital acquisition. Enterprise adjustment manifests primarily through hiring avoidance rather than headline terminations1. Rather than incurring the financial, legal, and organizational friction of firing senior staff, organizations simply refrain from opening new junior requisitions2. Survey data from executive cohorts reinforce this finding: while 80% of Federal Reserve Bank of Atlanta executive respondents report that AI investments have not reduced total headcount2, and Census surveys show only 5% of firms reporting immediate aggregate workforce shifts2, human resource channels report widespread consolidation of junior roles and quiet reductions in university recruiting pipelines2.

The Intertemporal Skill Accumulation Deficit

While skipping early-career hiring appears microeconomically rational on a corporate balance sheet seeking to maximize short-term operating margins3, it creates a severe intertemporal externality for the macroeconomy3. The institutional elimination of entry-level positions destroys the historical mechanism through which tacit knowledge is transferred across generations3. The institutional role of junior staff has historically extended beyond immediate task delivery; entry-level roles served as a paid corporate apprenticeship3. In executing routine analysis, basic code drafting, and preliminary document preparation, junior staff internalized domain-specific paradigms, developed professional networks, and incrementally cultivated tacit judgment1. By automating the base of this human capital formation process, firms are dismantling the structural ladder that builds future service experts and senior architects3. An economy that ceases to train 23-year-old associates today guarantees an acute aggregate deficit of capable 33-year-old directors and principal engineers a decade later3. When current senior cohorts retire or leave the workforce, organizations will face a structural vacuum of senior talent that AI systems—lacking autonomous tacit reasoning—cannot fill1. This breakdown represents a systemic risk to the long-term institutional stability of knowledge-based industries3.

Dual Automation: The Convergence of Cognitive and Physical Task Substitution

Cognitive Task Displacement in Knowledge Sectors

Past industrial revolutions were characterized by localized task displacement: 19th-century mechanization replaced agricultural labor while expanding industrial manufacturing, and 20th-century computing automated clerical record-keeping while expanding white-collar professional services6. The contemporary paradigm differs fundamentally because software capabilities are expanding simultaneously across cognitive domains while robotic hardware advances in physical environments4. In white-collar cognitive sectors, task exposure is positively correlated with educational attainment and historical wage premiums1. Econometric exposure metrics demonstrate that occupations heavily reliant on structured digital input-output transformations face high rates of automated substitution1. Software development, legal preparation, tax compliance, data administration, and routine technical writing exhibit substantial task automation, driving down labor demand for early-career knowledge workers1.

Physical Automation Dynamics: Global Industrial Robotics

Simultaneously, physical task automation is accelerating rapidly across global supply chains, manufacturing centers, and logistics facilities4. According to the World Robotics 2026 Report published by the International Federation of Robotics (IFR), the global operational stock of industrial robots reached a record 5 million units in 2025, representing a 9% year-over-year increase4. This growth was fueled by an 11% annual surge in new deployments, with over 600,000 factory units installed worldwide in a single year4.

Geographically, industrial physical automation is heavily concentrated in Asia, with China maintaining a dominant position4. In 2025, Chinese factory installations grew 20% year-over-year to 354,000 units, accounting for 59% of all global deployments4. Furthermore, Chinese domestic robotics manufacturers increased their internal market share to 55%, selling 195,000 units4. The United States expanded installations by 12% to nearly 38,500 units, surpassing Japan to become the second-largest robot market globally4. Industry forecasts project annual global robot installations to expand by another 9% in 2026 to 655,000 units, reaching 806,000 annual installations by 202915.

The Structural Collapse of Labor Refuge Sectors

The concurrent expansion of software-based cognitive automation and hardware-based physical robotics fundamentally alters labor market adjustment pathways4. Historically, when one economic sector automated, displaced workers transitioned into alternative "refuge sectors"—for instance, workers leaving automated manufacturing facilities found absorption in expanding white-collar offices or routine service roles6. Under the dual-automation regime, both cognitive white-collar channels and physical assembly or logistics channels are compressing their labor inputs per unit of value added4. Displaced knowledge workers cannot easily transition into physical manufacturing or logistical warehousing, as those facilities are installing industrial robotics at record rates4. Consequently, labor reallocation is systematically channeled into personal service occupations characterized by low physical automation potential and low capital intensity—such as healthcare support, social assistance, basic food preparation, and residential construction14. Bureau of Labor Statistics data reflect this trend: employment growth is concentrated in private healthcare support (+13.3% projected through 2035\) and social services16, while high-paying information, finance, and professional service sectors exhibit subdued hiring or headcount consolidation14. This structural crowding of displaced labor into low-wage service roles exerts persistent downward pressure on real median wages across the broader workforce14.

Concentration of Capital Ownership and the Shrinking Labor Share

The Hyperscaler Capital Buildout

The unprecedented capital intensity of artificial intelligence infrastructure is driving an extraordinary concentration of economic ownership among a small cadre of technology conglomerates5. Global AI investment is projected to exceed \$1.00 trillion in 2026 alone, with US-based capital commitments reaching \$581 billion5. The primary engine of this capital accumulation is the corporate hyperscaler group—comprising Amazon, Microsoft, Alphabet, Meta, and Oracle—which is committing over \$800 billion in annual infrastructure expenditure5. Cumulative global investment in AI hardware and data infrastructure is on track to total \$1.80 trillion by the close of 20265. The scale and pace of this capital deployment are historically unprecedented20. Analysis of corporate capital flows reveals that hyperscaler AI expenditure is building at approximately 0.85 percentage points of US Gross Domestic Product (GDP) per year20. This rate of capital accumulation is nearly double the speed of the mid-2000s housing boom at its absolute peak (0.50 percentage points of GDP per year) and more than five times the pace of the late-1990s telecommunications buildout (0.15 percentage points per year)20. Consensus projections expect hyperscaler capex to absorb approximately 3% of US GDP annually between 2027 and 2029, up from 0.3% historically20.

The entry barrier to frontier AI capability is defined by extreme fixed capital requirements22. Engineering cost models developed by Epoch AI demonstrate that constructing a single 1-gigawatt (GW) flagship AI data center demands \$38.0 billion in upfront capital expenditure and \$0.9 billion in annual operating expenses22. Converting capital assets over their operational lifespan yields an annualized Total Cost of Ownership (TCO) of \$8.5 billion per gigawatt22. IT hardware—specifically GPU clusters, high-bandwidth memory, and advanced networking fabrics—accounts for \$5.0 billion, or 60%, of the annualized TCO22. Facility and power infrastructure represent 16% ($1.4billion),energyconsumptionaccountsfor14%($1.2 billion), and operating maintenance, taxes, and direct labor constitute the remaining 10% (\$0.9 billion)22. Long-term infrastructure forecasts from PwC project cumulative global data center capital expenditure to reach between \$31.6 trillion and \$50.0 trillion through 205021. This colossal capital barrier concentrates ownership of the foundational compute substrate within a tightly integrated oligopoly of hyperscalers and state-backed entities14.

Task Economics: Displacement, Reinstatement, and Market Power

To formalize the macroeconomic consequences of this capital concentration on functional income distribution, economists employ task-based frameworks developed by Acemoglu and Restrepo6. Aggregate economic production is modeled as a continuum of tasks [Figure omitted from source export]6: [Figure omitted from source export] where [Figure omitted from source export] represents the elasticity of substitution across tasks6. Tasks [Figure omitted from source export] are automated and produced exclusively using capital, while tasks [Figure omitted from source export] are performed by human labor6. The change in total labor demand following technological progress is governed by two opposing economic forces: [Figure omitted from source export] The productivity effect arises because automation increases value added, expanding labor demand in non-automated tasks6. When automation expands the set of automated tasks (an increase in [Figure omitted from source export]), capital directly substitutes for human labor, generating a negative displacement effect that shifts the task content of production against labor and always reduces labor's share of value added6. Historically, this displacement was counterbalanced by the reinstatement effect—the technological creation of entirely new, complex tasks ([Figure omitted from source export]) in which human labor held a clear comparative advantage6. In the current AI deployment cycle, the displacement effect is accelerating across both cognitive and physical domains, while the reinstatement effect has proven comparatively weak6. The creation of new tasks (such as prompt engineering or basic data annotation) has failed to generate sufficient high-wage volume to compensate for displaced routine tasks6. Furthermore, market concentration among hyperscalers introduces extreme asymmetric information and pricing power14. As major platforms integrate generative tools into enterprise workflows, they gain granular operational visibility into user workflows14. This allows platform owners to identify high-value, highly repeatable cognitive tasks, develop targeted automated workflows, and capture corporate economic surplus through increased platform markups14. Capital owners capture supernormal economic rents, accelerating the long-term decline in labor's share of national income6.

Sectoral Reallocation and Median Wage Stagnation

As high-wage cognitive tasks are automated, the structural reallocation of labor alters worker earning capacity14. Economic impact models indicate that between 4% and 7% of the total US labor force faces high direct displacement risks, with a substantially larger share facing task modification14. Displacement is heavily weighted toward high-relative-wage sectors, including information technology, legal services, management consulting, and financial intermediation14. When workers are displaced from these high-productivity industries, they do not remain permanently unemployed; rather, they are absorbed by low-relative-wage, non-automatable service sectors, such as healthcare assistance, hospitality, food service, and personal care14. This structural shift produces a subtle macroeconomic outcome: overall national output and labor productivity measures continue to grow—driven by massive capital productivity in the automated sectors—while median real wages stall or decline due to the oversupply of labor in low-wage service refuges14. Consequently, the gap between economy-wide output and household wage earnings widens, further depressing labor's functional share of gross national income6.

Macroeconomic Instability: Underconsumption, Demand Deficits, and Equilibrium Limits

Keynesian Technological Unemployment in the Digital Era

The macroeconomic consequence of a declining labor share is the emergence of a structural aggregate demand deficit14. In a closed economic system, aggregate demand relies on household consumption, which is predominantly funded through labor compensation7. Capital income, by contrast, accrues to high-net-worth individual shareholders and institutional entities with a significantly higher marginal propensity to save7. When technology shifts national income away from wage-earning households and toward capital owners, total consumption potential contracts relative to productive capacity7. This dynamic realizes the phenomenon originally articulated by John Maynard Keynes: technological unemployment—defined not merely as structural joblessness, but as an economy's inability to generate sufficient wage demand to clear the market of goods produced by capital24. As automated platforms and industrial robotics scale, aggregate supply expands efficiently4. However, because automated systems, algorithms, and industrial robots do not purchase consumer goods, residential housing, or consumer services, the economic loop breaks down7. Without sufficient wage income distributed across the working populace, final product markets face chronic underconsumption and persistent demand shortfalls7.

Inflation Dynamics, Precautionary Savings, and the Neutral Rate ([Figure omitted from source export])

The macroeconomic transition from early-stage AI investment to late-stage deployment exhibits a distinct, two-phase inflationary profile14. In the near term, the massive capital buildout generates supply-side price pressures14. High corporate demand for specialized semiconductors, high-bandwidth memory, grid infrastructure, land, and energy drives up localized price indices, creating upward pressure on Personal Consumption Expenditures (PCE) inflation14. However, in the medium-to-long term, as infrastructure matures and AI systems assume widespread operational workloads, macroeconomic dynamics transition from inflationary to strongly disinflationary14. By significantly reducing labor cost per unit of output and eliminating human bottlenecks in production, capital-intensive AI generates structural cost savings14. Simultaneously, persistent labor market uncertainty and entry-level employment fragility compel households to increase precautionary savings14. As workers anticipate higher career volatility, reduced income mobility, and potential displacement, household consumption spending is curtailed in favor of liquid reserve accumulation14. This surge in precautionary savings, combined with concentrated capital wealth, elevates the global supply of loanable funds relative to productive domestic investment opportunities14. Consequently, the macroeconomic neutral rate of interest ([Figure omitted from source export])—the real interest rate that maintains full employment and target inflation—faces downward structural pressure14. Central banks attempting to stabilize output face monetary policy constraints, as the equilibrium neutral rate drops toward the zero lower bound14.

The Structural Infeasibility of Pure Wage-Based Distribution

Evaluating the central research question reveals that a purely wage-based civilization cannot maintain long-term equilibrium when capital autonomously generates the vast majority of economic value7. When capital assets achieve operational substitution for human cognitive and physical labor across a broad task spectrum, the marginal market-clearing wage for routine human labor declines relative to total output6. Attempting to preserve a social system where household survival, healthcare access, and wealth generation are strictly tied to labor hours worked creates severe socio-economic instability7. Under traditional market distribution, as technology approaches high labor-replacement capabilities, national income concentrates among capital holders and hyper-specialized tacit experts1. The broader working populace experiences real wage stagnation, underemployment, or labor force detachment1. This concentration suppresses the aggregate consumer purchasing power necessary to monetize capital investments7. Without structural policy interventions to decouple household purchasing power from wage labor alone, the economic engine enters a self-reinforcing cycle of capital overcapacity, consumer underconsumption, escalating income inequality, and heightened political polarization7.

Institutional and Policy Pathways for a Post-Apprenticeship Economy

Capital Redistribution and Steering Technological Progress

To resolve the systemic impasse of a wage-based economy undergoing dual-automation shocks, theoretical models developed by Korinek, Stiglitz, and Acemoglu emphasize two imperative policy mechanisms: redistributing capital rents and actively steering technological progress7. Steering technological progress requires realigning fiscal incentives7. Current tax structures in many advanced economies exhibit a structural bias toward capital adoption: corporate investments in software, machinery, and automated infrastructure enjoy accelerated depreciation, tax deductions, and preferential capital gains rates, whereas human labor is subject to payroll taxes, health insurance mandates, and income taxation7. Harmonizing tax burdens across capital and labor factors removes artificial fiscal incentives for labor replacement, encouraging enterprises to direct research toward human-augmenting technologies rather than pure task-replacing automation6. Concurrently, capturing supernormal rents generated by oligopolistic hyperscalers can be executed through targeted corporate taxation, platform surcharges, or sovereign capital funds7. Establishing national wealth funds that purchase equity stakes in frontier capital assets provides a structural mechanism to socialize capital returns7. By distributing a portion of total factor capital returns directly to citizens as a national capital dividend, consumer purchasing power can be sustained independently of total labor hours worked, directly counteracting aggregate underconsumption7.

Labor Market Transition Subsidies and Retraining Architecture

Because labor market adjustments in high-exposure roles manifest primarily through reduced entry hiring rather than immediate firing, transitional policy must focus heavily on early-career job placement and skills acquisition1. Search-and-matching models incorporating generative AI demonstrate that targeted wage subsidies for early-career hires substantially reduce transitional labor market losses26. By offsetting the initial cost of training junior personnel, government-sponsored apprenticeship subsidies incentivize firms to maintain early-career hiring pipelines, preserving the human capital development chain3.

Simultaneously, public retraining infrastructure must be restructured8. Educational programs historically focused on teaching standardized, rules-based technical competencies—such as basic coding, routine data processing, or standardized report drafting—are preparing students for tasks that AI executes at near-zero marginal cost1. Curricula must pivot toward non-codified, tacit competencies: unstructured problem synthesis, ethical governance, cross-domain strategic reasoning, physical dexterity, and high-empathy interpersonal management1.

Restructuring Human Capital Formation and Public Equity Endowments

Long-term socio-economic stability requires decoupling human welfare from sole reliance on wage earnings7. As the capital share of national income expands, institutions must grant citizens direct access to capital ownership7. This structural re-anchoring can be achieved through universal capital endowments, employee equity participation mandates, and expanded public wealth holdings7. By transitioning from a pure wage-income standard to a hybrid wage-plus-capital-equity standard, households participate directly in the wealth expansion generated by corporate AI deployment7. When technological productivity gains expand capital earnings, household equity portfolios appreciate, offsetting labor earnings stagnation and maintaining total aggregate demand7.

Conclusion: Re-Anchoring Civilization Beyond the Wage Standard

The empirical evidence derived from administrative payroll records, global industrial robotics deployment metrics, and macroeconomic capital flows points to a profound structural shift: artificial intelligence and robotics are fundamentally altering the historic balance between capital and labor. The widening 19% employment gap for young workers in AI-exposed fields serves as an early indicator of institutional disruption1. By automating codified tasks while preserving senior tacit oversight, corporate enterprises are systematically bypassing entry-level hiring1. This behavior secures short-term operating efficiencies while breaking the historical human capital pipeline and threatening future skilled talent reserves3. Concurrently, the dual-automation shock—combining cognitive software deployment with record physical industrial robotics installations—is squeezing traditional labor refuge sectors and concentrating national income among capital hyperscalers4. This concentration produces a severe macroeconomic vulnerability: aggregate output and asset values expand driven by capital productivity, but aggregate consumer demand faces stagnation due to eroded worker purchasing power and declining labor income shares7. A pure wage-based civilization cannot remain stable if technological progress permanently decouples output growth from human labor demand. Preserving long-term economic stability requires a structural modernization of institutional political economy. By harmonizing capital-labor fiscal policies, establishing public capital equity dividends, subsidizing early-career human capital formation, and transitioning toward a hybrid capital-labor income standard, society can successfully navigate the transition beyond the traditional wage standard—ensuring that the extraordinary wealth generated by autonomous technology yields shared, sustainable prosperity7.

Works cited

1. No Widespread Displacement, but the AI Employment Gap for, https\://digitaleconomy.stanford.edu/news/canariesaug26/

2. https\://siepr.stanford.edu/publications/policy-brief/what-really-happening-jobs-separating-ai-hype-reality

3. Stanford Payroll Data Shows AI Has Cut Young Workers' Job Gap to, https\://startupfortune.com/stanford-payroll-data-shows-ai-has-cut-young-workers-job-gap-to-19/

4. Five Million Robots now Operate in Factories Globally, https\://ifr.org/ifr-press-releases/news/five-million-robots-now-operate-in-factories-globally

5. Global AI Investment Is Forecast to Exceed \$1 Trillion in 2026, https\://www\.goldmansachs.com/insights/articles/global-investment-is-forecast-to-exceed-1-trillion-in-2026

6. acemoglu-restrepo-2019-automation-and-new-tasks-how, https\://shapingwork.mit.edu/wp-content/uploads/2023/10/acemoglu-restrepo-2019-automation-and-new-tasks-how-technology-displaces-and-reinstates-labor.pdf

7. Artificial Intelligence and Its Implications for Income Distribution and, https\://www\.nber.org/system/files/working\_papers/w24174/w24174.pdf

8. AI and Labor Markets: What We Know and Don't Know, https\://digitaleconomy.stanford.edu/news/ai-and-labor-markets-what-we-know-and-dont-know/

9. Canaries in the Coal Mine? Six Facts about the Recent Employment, https\://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/

10. The AI Economic Indicators \- Stanford Digital Economy Lab, https\://digitaleconomy.stanford.edu/project/indicators/

11. Who Will Fix the Machines When Your Veteran Engineers Retire?, https\://www\.cmswire.com/customer-experience/are-you-automating-away-the-apprenticeship-that-builds-service-experts/

12. Automation and New Tasks: How Technology Displaces and, https\://www\.aeaweb.org/articles?id=10.1257/jep.33.2.3

13. Artificial Intelligence and Its Implications for Income Distribution and, https\://onwork.edu.au/bibitem/2019-Korinek,Anton-Stiglitz,Joseph+E-Artificial+Intelligence+and+Its+Implications+for+Income+Distribution+and+Unemployment-excerpt+p.350/

14. Macro Signposts | AI, Market Power, and Diminishing Labor Share, https\://www\.pimco.com/us/en/insights/ai-market-power-and-diminishing-labor-share

15. IFR: global operational stock of industrial robots rose 9% to a record, https\://www\.thescxchange.com/move/produce/ifr-global-operational-stock-of-industrial-robots-rose-9-to-a-record-5-million-units-in-2025

16. Employment Projections Home Page \- Bureau of Labor Statistics, https\://www\.bls.gov/emp/

17. The January 2026 Jobs Report & December 2025 JOLTS \- iHire, https\://www\.ihire.com/resourcecenter/employer/pages/the-january-2026-jobs-report-december-2025-jolts

18. Job Openings and Labor Turnover Survey News Release, https\://www\.bls.gov/news.release/archives/jolts\_09012026.htm

19. I. Progress and peril \- the BIS, https\://www\.bis.org/publications/aer-2026/progress-peril

20. The AI Capex Boom Is Building Twice as Fast as the Housing Boom, https\://www\.apollo.com/wealth/insights-news/insights/daily-spark/the-ai-capex-boom-is-building-twice-as-fast-as-the-housing-boom

21. Where \$31.6 trillion of capex flows in the era-defining AI build-out, https\://www\.pwc.com/gx/en/1/services/consulting/technology/data-centre-outlook.html

22. Total cost of ownership of a one-gigawatt AI data center \- Epoch AI, https\://epoch.ai/data-insights/ai-datacenter-cost-breakdown

23. The AI expectations gap \- Anthropic, https\://www\.anthropic.com/research/ai-expectations-gap

24. Joseph Stiglitz and Anton Korinek's INET-funded research is cited in, https\://www\.ineteconomics.org/about/news/2021/joseph-stiglitz-and-anton-korineks-inet-funded-research-is-cited-in-the-ny-times

25. Technological Unemployment \- ResearchGate, https\://www\.researchgate.net/publication/365321772\_Technological\_Unemployment

26. Artificial Intelligence and Technological Unemployment \- NBER, https\://www\.nber.org/system/files/working\_papers/w33867/w33867.pdf

27. Homepage of Anton Korinek \- Research, https\://www\.korinek.com/research

28. AI datacenter capex will far exceed forecast revenue through 2030, https\://www\.computerworld.com/article/4236641/ai-datacenter-capex-will-far-exceed-forecast-revenue-through-2030-3.html

29. Artificial Intelligence and Its Implications for Income Distribution and, https\://econpapers.repec.org/bookchap/nbrnberch/14018.htm