Runaway self-improving AI coupled with financial amplification can produce an economic blow-up and runaway inequality unless taxes on AI and financial returns are implemented early and sufficiently; delay past a critical time makes redistribution politically impossible.

Main Finding

A stylized dynamical-systems model with quadratic self-amplification in both AI capability (term λ A^2) and financial capital (term γ_F K_f^2), coupled by investment flows, generically produces a finite-time singularity in which AI capability, AI capital, and financial capital diverge. This divergence drives super-exponential growth in the wealth ratio between capital owners and workers; financialization amplifies that exponent by a factor γ_F/η. Introducing taxes on AI returns (τ_ai) and financial gains (τ_f) generates three qualitatively distinct long-run regimes (low-, moderate-, and high-tax), but there exists a finite “policy window”: beyond a critical time before the singularity, feasible redistribution becomes politically infeasible because concentrated wealth forces feasible tax rates toward zero.

Key Points

• Model structure

  • Two nonlinear self-amplifying processes: AI capability grows with λ A^2 (recursive self-improvement) and financial capital grows with γ_F K_f^2 (financialization/leverage).
  • Coupling occurs through investment allocations between AI capital, financial capital, and wages/consumption.
  • Agents are aggregated into capital owners and workers; taxation is modeled as rates on AI returns (τ_ai) and financial gains (τ_f).

• Finite-time singularity

  • Under mild parameter conditions, the coupled nonlinear dynamics admit solutions that blow up in finite time (A, K_ai, K_f → ∞).
  • Near the singularity the dominant growth is governed by the quadratic terms; asymptotic analysis yields closed-form leading-order behavior.

• Inequality dynamics

  • The wealth (capital-owner to worker) ratio diverges super-exponentially as the singularity is approached.
  • Financialization amplifies the divergence rate: the exponent is multiplied by γ_F/η (η is a model parameter governing worker-income dynamics/response), so stronger financial self-amplification accelerates inequality.

• Taxation and long-run regimes

  • Two tax instruments (τ_ai, τ_f) produce three regimes:
  • Low-tax: taxes below a first critical threshold → extreme inequality and singularity-driven concentration.
  • Moderate-tax: taxes between thresholds → a stable mixed economy (AI and finance large but bounded; wages maintained).
  • High-tax: taxes above a second threshold → de facto post-scarcity, where redistributed returns can finance universal basic income (UBI) or similar.
  • Thresholds are derived analytically from the workers’ budget constraint and characterize the minimum taxation needed to prevent the singularity or to achieve high redistribution.

• Policy irreversibility

  • There exists a latest feasible intervention time (a critical time before the singularity) beyond which political-economy constraints (extreme concentration) make required tax rates infeasible; feasible tax space collapses to zero.
  • This creates urgency: outcomes depend crucially on when redistribution efforts begin, not just on their eventual scale.

Data & Methods

• Methodology

  • Analytical dynamical-systems approach: specification of ODEs for AI capability A(t), AI capital K_ai(t), financial capital K_f(t), and worker wealth/consumption.
  • Nonlinear terms: λ A^2 and γ_F K_f^2 produce superlinear growth; coupling terms model investment flows and income shares.
  • Asymptotic analysis: identification of finite-time blow-up solutions and leading-order scaling laws near singularity.
  • Bifurcation / regime analysis: derive critical tax thresholds analytically from steady-state / budget-constraint relations; classify long-run regimes.
  • Numerical simulations: parameter sweeps and time-path plots illustrate trajectories, validate asymptotics, and show policy-window effects.

• Data

  • The paper is theoretical / model-based and does not rest on micro empirical estimation of λ, γ_F, η, or tax enforcement capacity. Simulations use illustrative parameter choices to demonstrate qualitative behavior and robustness.

• Assumptions & limitations

  • Stylized aggregation: two agent classes (capital owners, workers) with simplified behavioral rules.
  • Deterministic dynamics (no stochastic shocks).
  • Quadratic self-amplification is a modelling choice capturing runaway feedback; results depend on nonlinearity type/degree.
  • Political feasibility and enforcement of taxes are captured abstractly (feasible tax set shrinks with concentration) rather than modeled micro-politically.
  • International capital mobility, regulatory responses, institutional evolution, and micro-strategic behavior of firms are not modeled in detail.

Implications for AI Economics

• Policy urgency

  • Early intervention substantially expands feasible policy space. Waiting risks hitting the policy irreversibility threshold where redistribution is no longer implementable.
  • Policymakers should prioritize measures that limit runaway self-reinforcing dynamics in AI and finance before extreme concentration occurs.

• Targets for intervention

  • Instruments that directly reduce self-amplification (e.g., limiting certain forms of recursive self-improvement, throttling leverage in financial vehicles linked to AI) can lower effective λ and γ_F and help avoid singular paths.
  • Fiscal tools: taxes on AI returns and financial gains can, if set above analytically identifiable thresholds, prevent singularity or enable high redistribution (UBI). But taxes must be politically enforceable early on.

• Financialization as an amplifier

  • Financial-sector dynamics magnify inequality trajectories; regulating financial leverage, shadow banking, and novel AI-linked financial products may be as important as tech-specific controls.

• Institutional design

  • To maintain feasible redistribution later in the transition, build political and institutional capacity early (e.g., progressive taxation mechanisms, public ownership stakes in key AI assets, international coordination).
  • Design transfer systems (UBI, wage floors, public investment) that can scale with rapidly rising returns if needed.

• Research directions

  • Calibrate model parameters (λ, γ_F, η) using empirical proxies for AI capability growth and financial amplification to estimate realistic policy windows.
  • Extend the model to include heterogeneous agents, stochastic shocks, endogenous political economy (voter and lobbying dynamics), international capital flows, and institutional evolution.
  • Explore alternative nonlinearities (other than quadratic) and interventions (capital controls, ownership rules, antitrust) to assess robustness of the singularity and irreversibility results.

Summary takeaway: quadratic self-reinforcing dynamics in AI and finance can generate a rapid, policy-sensitive transition to extreme concentration; taxes and other interventions can avert worst outcomes but must be implemented early because a finite-time policy window exists beyond which redistribution becomes infeasible.