Some AI systems now operate as de facto centers of power; policymakers should treat them as a fourth societal actor and redesign institutions with federalized, polycentric checks and balances to preserve democratic legitimacy and accountability.

Main Finding

The paper argues contemporary advanced AI systems should sometimes be treated not as tools or infrastructure but as a distinct societal actor — the “Digital Gorilla” — alongside People, the State, and Enterprises. This Four Societal Actors framework, together with five power modalities (economic, epistemic, narrative, authoritative, physical), reveals how AI can concentrate power, reconfigure accountability, and undermine existing doctrinal frames. The authors propose a federalized, polycentric governance architecture that institutionalizes dynamic checks and balances among the four actors and shifts policy emphasis from controlling a risky technology to designing institutions that accommodate powerful, semi-autonomous digital systems while preserving democratic legitimacy and public goods.

Key Points

• The “analogy trap”: current AI policy borrows inconsistent analogies (product, platform, infrastructure, general-purpose technology), producing a fragmented, often contradictory regulatory patchwork that misses how some AI systems operate as de facto centers of power.
• Functional claim: calling AI a societal actor is not anthropomorphism but a recognition that certain AI systems exercise decision-making power, coordinate at scale, and reshape social and economic realities in ways that demand distinct governance approaches.
• Four Societal Actors model: People, the State, Enterprises, and the Digital Gorilla — used to map power flows and diagnose where AI disrupts traditional allocations of authority.
• Five power modalities used to analyze influence: economic (market power, rents), epistemic (knowledge production, inference), narrative (agenda-setting, framing), authoritative (delegated decision-making, rule enforcement), and physical (control over material systems, safety-critical actuation).
• Doctrinal strain: AI’s actor-like effects stress constitutional, administrative, and private-law doctrines (e.g., nondelegation, due process, fiduciary duty, product liability) that assume human-centered decision-making.
• Governance prescription: adopt polycentric, federalized institutions inspired by separation-of-powers and federalism to create upstream, lifecycle-aware, capability-sensitive governance; institutionalize checks and balances among all four actors rather than relying chiefly on ex post compliance and sectoral rules.
• Scope limitation: analysis assumes liberal-democratic, market-oriented states; implications for authoritarian or state-capitalist regimes require separate treatment.
• Practical doctrinal implications: reframe debates in administrative law (delegation, review), corporate law (responsibility and accountability for AI-driven decisions), competition policy (platform/control of essential AI components), and liability/regulatory design (upstream capability oversight, lifecycle obligations).

Data & Methods

• Primarily conceptual and doctrinal analysis rather than empirical estimation.
• Methodological elements:
  • Literature synthesis across AI governance, administrative and constitutional law, technology policy, and political theory (e.g., separated powers, federalism, general-purpose-technology literature).
  • Comparative policy review and case examples: EU AI Act, U.S. executive actions and NIST work, China’s AI measures and platform regulations, export controls, and national industrial policies (CHIPS etc.).
  • Analytic mapping: development of a Four Societal Actors model and a five-modalities taxonomy to trace power flows and failure modes.
  • Use of empirical repositories and mapping efforts (e.g., MIT AI Governance Mapping, AI incidents repositories) as background evidence of regulatory focus and gaps.
• No primary quantitative data collection or formal econometric analysis; empirical claims are supported by policy documents, legal cases, and existing incident/mapping datasets.
• Limitations acknowledged: conceptual focus, normative institutional design, and jurisdictional assumptions (liberal-democratic contexts).

Implications for AI Economics

• Market structure and concentration
  • Viewing powerful AI systems as actors highlights new sources of market power (control of foundational models, compute, proprietary data, and distribution channels).
  • Governance should scrutinize upstream concentration (compute, model training, foundation models) because downstream-focused sectoral rules leave entry barriers and rents intact.
• Rents, incentives, and allocation
  • The Digital Gorilla can capture rents via algorithmic coordination, platform effects, and epistemic dominance. Policy design (access rules, compelled interoperability, public compute) will shape where rents accrue — to private firms, the public sector, or open commons.
  • Licensing, taxation, or public-provision mechanisms can reallocate returns from concentrated AI capability to public goods or to offset externalities.
• Innovation and investment
  • Regulatory design affects incentives for R&D: capability-based upstream constraints (e.g., pre-deployment evaluations, certification) may raise costs and alter risk-taking; but clearer institutional rules and access conditions can lower uncertainty and encourage socially beneficial direction of innovation.
  • Cross-border regulatory divergence (export controls, differing liability regimes) will influence geographic allocation of investment and strategic trade-offs in innovation policy.
• Externalities, public goods, and commons
  • AI’s epistemic and narrative power produces systemic externalities (misinformation, coordination failures, market expectations) that private markets underprovide remedies for — legitimizing public interventions (public models, datasets, validation infrastructure).
  • There is an economic case for public goods provision: public-interest compute, shared evaluation datasets, independent auditing capacities, and public infrastructures for model stewardship to internalize social benefits.
• Competition policy and platform remedies
  • Traditional antitrust frameworks may need extension to capture non-price harms (epistemic/narrative dominance, control of downstream attention markets) and to intervene upstream (essential-input access, forced interoperability, structural remedies).
• Labor, matching, and distributional effects
  • As AI systems take on coordination and decision tasks, labor market impacts and reallocation dynamics will depend on governance choices that affect deployment incentives, access to capability, and safety constraints.
  • Redistribution and retraining policies interlock with governance choices: concentrated Digital Gorilla power can exacerbate inequality unless governance channels include redistributive or public employment/education responses.
• Policy design and economic modeling needs
  • New empirical metrics and models are needed to measure “Digital Gorilla” power (e.g., market shares in foundational models and compute, measures of epistemic reach, influence on attention markets).
  • Welfare analysis should incorporate nonmarket harms (information quality, democratic legitimacy, epistemic monopolies) and consider institutional design choices (polycentric vs. centralized regulation) as choice variables affecting dynamic outcomes.
• Practical policy levers relevant to economists
  • Capability-sensitive interventions (pre-deployment review, capability-contingent restrictions).
  • Access and interoperability mandates (reduce gatekeeping, lower entry barriers).
  • Compulsory stewardship, fiduciary duties for entities deploying actor-like AI.
  • Public investment in shared compute/datasets and independent verification infrastructure.
  • Competition and liability reforms to account for diffuse, systemic harms.

Overall, the paper reframes many AI policy questions as allocation-of-power problems with clear economic content: who captures returns, how markets and information structures change, which public goods are undersupplied, and how regulatory architecture shapes incentives for innovation and distribution. Economists should therefore extend models to include institutional actors (Digital Gorillas) and nonprice sources of market power, and evaluate governance interventions both for static efficiency and for dynamic, distributional, and informational welfare effects.