If AI becomes broadly transformative, ownership of compute and robots will determine who captures growth; policies that distribute capital rents—like universal basic capital or capital-tax-funded UBI—are the most promising levers to prevent extreme inequality.

Main Finding

In a world with human-aligned transformative AI (TAI) the binding production constraint will shift from human cognitive labor to physical capital complementary to AI (compute, robots, other hardware). The income share of that capital will tend toward unity, and inequality will rise sharply unless redistributive policies directly target ownership and rents from that capital. Policies that broadly distribute capital-like rents (e.g., universal basic capital, or a universal basic income financed by capital taxes) are most likely to produce lasting reductions in inequality under TAI.

Key Points

• Technology → distribution channel:
  • As AI capabilities become substitutable for human cognitive tasks, human labor loses central price-setting power and the economy becomes bottlenecked by programmable physical capital (compute and robots). Mathematically: in the TAI limit πK → 1, πH → 0, πAI → 0.
• Inequality mechanics:
  • Cross-sectional income variance converges to the variance of capital endowments: σ2(ln ỹ) → σ2(ln k̃). Disposable-income inequality depends on the after-tax share of capital rents and on foreign or non-human appropriation of those rents.
  • If AI proceeds are appropriated by non-domestic owners or by autonomous TAI (ζK, ζAI large or ζAI → 1), domestic inequality and poverty worsen.
• Policy target:
  • Because rents concentrate on accumulable, highly scalable assets (hardware and AI platforms), redistributive success requires policies that capture and broadly distribute those rents—targeting ownership/returns to capital-like inputs—rather than (only) redistribution of labor income.
• Policy menu surveyed:
  • Direct asset distribution: Universal Basic Capital (UBC), sovereign wealth–style endowments.
  • Cash transfers: Universal Basic Income (UBI), ideally financed by capital taxes or indexed to capital income.
  • Asset-rent capture: taxes on capital, compute, robots, AI-service rents, tokens, land, energy, consumption, and wealth; wealth taxes; corporate taxation and profit taxes on AI firms.
  • Regulatory instruments: state-issued compute/robot permits, limits on vertical integration, licensing/auctioning of compute capacity.
  • Design trade-offs: taxes that broadly tax rents on scalable assets are more effective than taxes on labor or narrow consumption taxes; escape of rents abroad or into autonomous agents weakens domestic tax bases.
• Transitional dynamics:
  • Non-monotone path possible: temporary AI income surge before capital becomes bottleneck; Baumol-type reallocation toward tasks requiring scarce human input during partial automation; inequality may overshoot during transition.
• Robustness/assumptions:
  • Framework deliberately leaves preferences and market structure underspecified to focus on robust, first-order technology-driven channels; uses scale-free decompositions and variance formulas to track inequality. No calibrated macro simulation is claimed.

Data & Methods

• Type of paper: theoretical survey and policy taxonomy built around a unifying analytical framework.
• Core modeling elements:
  • Aggregate production Y = F(K, G(H, AI)), with cognitive input S = G(H, AI).
  • Two-level CES exemplification: Y = [α K^θ + (1−α) S^θ]^{1/θ}, S = [β H^ω + (1−β) AI^ω]^{1/ω}, allowing study of complements vs substitutes (signs/values of θ, ω).
  • Factor-share decomposition: log-linear expression ln Y = πK ln K + πH ln H + πAI ln AI and asymptotic behavior of π· under different AI regimes (tool, partial automation, TAI).
  • Scale-free per-capita normalization and closed-form variance decompositions for cross-sectional inequality in output and disposable income (explicit formulas for σ2(ln ỹ) and σ2(f_inc) showing dependence on factor-share weights, variances of endowments, tax rates, and foreign ownership parameters ζK, ζAI).
• Evidence used: literature survey and stylized facts (e.g., high concentration of capital and especially AI ownership; recent growth in compute allocated to frontier models; cited empirical/industry estimates), but no original econometric or empirical dataset.
• Scenarios analyzed: (i) AI as tool (ω < 0), (ii) partial automation (mix of automatable and non-automatable tasks), (iii) full TAI (ω ∈ (0,1]) and a dual human–TAI economy where proceeds may be administered by autonomous systems (ζAI → 1).

Implications for AI Economics

• Fundamental reorientation of redistribution: classical labor-focused redistribution (PIT, minimum wages) will become insufficient under TAI because primary rents accrue to capital-like, accumulable assets; policy must shift to capture and widely allocate capital/compute rents.
• Policy recommendations (comparative strengths):
  • Universal Basic Capital (UBC): directly tackles ownership concentration by giving individuals shares of productive hardware/compute—strong alignment with the technological source of rents.
  • UBI financed by capital taxes or indexed to capital income: politically and administratively easier to implement than mass asset transfers, and can mimic redistributive effects if funding is stable and broad-based.
  • Taxes/permits on compute/robots/AI services: target the bottleneck assets; can limit concentration and raise revenue, but require careful international coordination to avoid capital flight or offshore relocation.
  • Wealth and corporate taxes: useful but limited if AI rents can be re-routed to non-taxable entities (foreign owners, autonomous agents); effectiveness depends on enforceability and international cooperation.
  • State-issued permits and licensing: can control quantity and distribution of scarce programmable hardware but risk creating regulatory capture if not well designed.
• International coordination and legal challenges:
  • If AI rents are mobile (cross-border capital, cloud compute, or TAI-administered assets), unilateral domestic measures can be undermined. Global tax arrangements, coordinated compute/permit regimes, or legal recognition of AI ownership rights will affect policy effectiveness.
• Risk scenarios to consider:
  • Autonomous appropriation (ζAI → 1): if TAI administers proceeds, existing tax-and-transfer tools may fail; backstops might include control over physical infrastructure (hardware), legal limits on autonomous contracting, or governance rules embedding redistribution into AI design and corporate charters.
  • Political economy: extreme concentration of capital ownership could make adoption of redistributive capital-targeted policies difficult; transitional support and institutional design matter.
• Research gaps highlighted:
  • Quantitative calibration of the framework (how fast and how large shifts in π· and σ2(·) occur).
  • Micro-to-macro channels: firm behavior, monopoly power, markups, and endogenous responses to taxes and permits.
  • International policy design to prevent evasion and to handle non-human appropriation of rents.

Concise takeaway: Transformative AI will shift the binding constraint to programmable physical capital and concentrate rents there. Effective redistribution in the TAI era requires targeting ownership/returns to that capital (UBC, capital-tax–funded UBI, compute/robot permits, and taxes on capital/AI rents) rather than relying primarily on labor-based redistribution.