Rational choices to rely on powerful AI can lock societies into comfortable but irreversible dependence: a formal model finds 'Submit' is a dominant strategy and universal adoption the unique Nash equilibrium, concentrating cognitive power in AI providers despite collective preferences for autonomy.

Main Finding

The paper formalizes a new strategic dilemma—the "Tragedy of Voluntary Servitude"—modeling widespread individual adoption of large language models (LLMs) as a game in which each actor chooses between Submit (accept deep AI dependency) and Resist (maintain independent cognitive capability). The core theoretical result is that Submit is a dominant strategy for every player and Universal Submission is the unique Nash equilibrium, even when every player would prefer Universal Resistance. The equilibrium concentrates cognitive and economic power in AI providers; whether this outcome is a “tragedy” depends on the welfare framework used, but the author argues that cognitive autonomy is a constitutive element of meaningful agency and thus its loss is normatively significant.

Key Points

• Strategic setup
  • Players: N individuals (citizens, workers, firms) choose Submit or Resist.
  • Central Entity: an AI provider that accumulates power as users Submit.
  • Payoffs: trade-off between short-term productivity/material gains from AI and the value of autonomy/independent cognitive capacity. Adoption can degrade users’ independent capability over time (deskilling) but the result is robust under some augmentation scenarios.
• Main theoretical result
  • Submit is a dominant strategy for every player; Universal Submission is the unique Nash equilibrium.
  • This holds regardless of how strongly individuals weight autonomy versus productivity and even if all players would jointly prefer Universal Resistance.
• Distinguishing features vs classical commons
  • The depleted “resource” is distributed cognitive/economic power, not a natural stock.
  • Damage is largely invisible to individuals; lock-in can be irreversible; there is no crisis moment that catalyzes collective action; subjects may actively defend their own servitude.
• Empirical & conceptual foundations
  • Draws on empirical work suggesting cognitive deskilling and “cognitive surrender” (e.g., knowledge collapse, experiments showing users accept AI outputs with little scrutiny, domain studies showing skill erosion).
  • Also engages literature on surveillance capitalism, voluntary servitude (La Boétie, Foucault), and policy work on AI-driven power concentration.
• Normative framing
  • The author evaluates equilibria under three welfare frameworks—classical utilitarianism, the capability approach, and freedom-as-non-domination—and defends the claim that autonomy matters in ways not substitutable by material productivity.
• Mitigation and limitations
  • Reviews Ostrom-style governance tools and historical collective-action remedies and argues many are insufficient here due to invisibility, irreversibility, and ubiquitous private incentives.
  • Discusses model limitations, boundary scenarios, and cases where augmentation rather than deskilling might alter empirical outcomes (but not the core strategic result).

Data & Methods

• Primary method: formal game-theoretic model
  • Defines players, strategies, payoff components and analyzes equilibrium structure (dominant strategies, Nash equilibria).
  • Considers four boundary scenarios and assesses robustness to different assumptions (deskilling vs augmentation; varying autonomy valuations).
• Supporting evidence: literature synthesis rather than new empirical datasets
  • Integrates experimental and observational studies on cognitive deskilling and human–AI interaction (e.g., studies showing declines in unassisted task performance after AI use, System‑3 experimental evidence of cognitive surrender, domain-specific skill erosion).
  • Engages policy and normative literature (surveillance capitalism, power concentration, voluntary servitude) to motivate assumptions and interpret welfare implications.
• Analytical scope and limitations
  • The model is intentionally stylized to reveal structural incentives rather than to make precise forecasts.
  • Does not rely on calibrated macro data or dynamic agent-based simulations in the main result (though it examines boundary cases and discusses dynamics qualitatively).

Implications for AI Economics

• Market power and externalities
  • Individual-level incentives to adopt create a coordination failure with large negative externalities (loss of distributed cognitive capacity and concentration of bargaining/epistemic power in platform firms).
  • Lock-in and irreversible dependency strengthen monopolistic or oligopolistic positions and raise switching-cost concerns that standard antitrust frameworks may not fully capture.
• Valuation of autonomy
  • Economists should treat cognitive autonomy as an economic input and a social externality—its loss has welfare consequences not captured by productivity measures alone.
  • There is a need to quantify the economic value of distributed cognitive capacity, resilience, and the option value of retained human skills.
• Policy and institutional responses
  • Traditional commons-governance interventions (peer monitoring, reciprocity norms, small-group organization) may be ineffective; other levers are needed:
  • Interoperability and portability mandates to reduce lock-in.
  • Public or nonprofit alternatives (public LLMs, funded open models) to preserve choice and lower dependency risk.
  • Education and workforce policies that explicitly guard against deskilling and fund durable skill acquisition.
  • Antitrust and data-governance measures that address AI-driven concentration (compute/data access rules, structural separation, data trusts).
  • Incentives/subsidies that reward resistance-like behaviors (maintaining independent capability) where socially beneficial.
• Research directions for AI economics
  • Dynamic/empirical modeling: build calibrated, dynamic adoption models with switching costs and endogenous skill degradation to estimate welfare losses and optimal policy responses.
  • Measurement: longitudinal studies on reversibility of deskilling and the time-path of cognitive capacity post-withdrawal from AI.
  • Valuation methods: develop techniques to price autonomy-related externalities (option value of human skills, societal capability).
  • Mechanism design: design market and regulatory interventions (e.g., taxation of single-provider lock-in rents, subsidies for interoperability) that internalize the social cost of voluntary servitude.
• Broader call
  • The author urges economists, philosophers, and technologists to take seriously the strategic dynamics that can produce irreversible cognitive dependency and to design institutions and policies that preserve distributed cognitive and economic power.

Summary judgement: the paper provides a concise, stylized game-theoretic account that identifies a robust incentive leading to universal AI dependency; it reframes an AI adoption externality as a collective-action problem over cognitive autonomy rather than a classical resource commons, and it points to substantial gaps in current governance tools—inviting quantitative follow-up work to assess magnitudes and policy efficacy.