AI requires global-commons governance: policymakers should regulate data, energy and compute across social, planetary and safety domains to curb monopolistic power and algorithmic harms.

Main Finding

The paper argues that AI should be governed as a global commons. Effective governance must coordinate interventions across three interconnected domains—social, planetary, and safety—because risks and root drivers (e.g., monopolistic AI power and an AI-infused media environment) create cross-domain feedbacks. Data, energy, and compute are identified as core regulatory levers that cut across these domains. Policy priorities include limiting agentic AI, incentivising depolarising platform algorithms, and embedding AI governance within global equity frameworks.

Key Points

• Threefold contribution:
• A classification of AI risks across social, planetary, and safety domains.
• Policy argument: AI governance requires a global-commons approach that accounts for inter-domain feedbacks and root drivers.
• Identification of data, energy, and compute as high-leverage regulatory dimensions spanning all domains.
• Social-domain risks:
  • Misinformation, polarization, and degraded information ecosystems due to AI-driven content and recommendation systems.
  • Labour-market disruption, distributional inequality, and concentration of market power in platform/AI monopolies.
  • Erosion of democratic processes and social trust through targeted manipulation.
• Planetary-domain risks:
  • High energy and material footprints of training and running large models (compute- and data-intensive AI).
  • Supply-chain impacts (resource extraction, e‑waste) and embodied emissions associated with AI infrastructure.
• Safety-domain risks:
  • Alignment failures, misuse, dual-use technologies, and the potential for agentic/runaway AI to cause catastrophic outcomes.
  • Concentration of capabilities in a few actors amplifies systemic risk and reduces redundancy/resilience.
• Root drivers and feedbacks:
  • Monopolistic incentives (scale economies in compute and data) push consolidation and centralised control.
  • AI-infused media environments amplify social harms and create incentives for ever-larger models and more targeted data collection.
  • Energy- and compute-intensive model development accelerates planetary impacts, which in turn shape geopolitical and economic dynamics.
• Regulatory levers:
  • Data governance (access, portability, privacy, cross-border flows) affects market structure and social outcomes.
  • Energy and emissions regulation (carbon pricing, efficiency standards, reporting) internalise planetary externalities from compute.
  • Compute governance (caps, licensing, transparency, export controls) can limit risky capability build-up and concentration.
• Policy recommendations highlighted:
  • Limit development/deployment of agentic AI architectures or tightly condition their use.
  • Incentivise platform algorithms that reduce polarization (e.g., through reward structures, regulation).
  • Embed governance in global-equity frameworks to ensure fair distribution of benefits and burdens.

Data & Methods

• Methodological approach: conceptual and synthetic. The paper constructs a taxonomy of risks, maps inter-domain feedbacks, and advances a normative governance framework.
• Primary methods include literature synthesis, conceptual classification, policy analysis, and tracing causal feedback loops across domains.
• No new empirical micro- or macro-datasets or econometric estimates are reported; instead the paper integrates existing findings and theoretical arguments to motivate regulatory priorities.
• Where quantitative claims are implicit (e.g., scale economies in compute), the paper relies on documented industry patterns (rising model size, energy intensity) and established economic reasoning rather than original empirical estimation.

Implications for AI Economics

• Inputs and market structure:
  • Compute, data, and energy are strategic scarce inputs—regulation of these will alter firms’ cost structures, competitive dynamics, and returns to scale.
  • Policies that restrict or price compute/energy (caps, taxes, carbon pricing) shift comparative advantages across firms and countries and may slow rapid capability races.
  • Data governance (access rules, portability, public-data provision) reshapes barriers to entry and the distribution of economic rents.
• Externalities and public goods:
  • Many AI harms are global externalities (information ecosystem degradation, planetary emissions, systemic safety risks), motivating international coordination and public‑good provision.
  • Treating AI as a global commons implies institutional choices (international agreements, common standards, shared monitoring infrastructure) akin to climate governance or internet governance.
• Innovation and incentive effects:
  • Constraint-based governance (compute limits, licensing) trades off speed of capability growth against risk mitigation; economic models should quantify these trade-offs for welfare assessment.
  • Subsidies or prizes for depolarising/beneficial algorithms and investments in public‑interest models can redirect innovation toward socially valuable outcomes.
• Distributional and development considerations:
  • Governance must account for global equity: restrictions that raise costs or centralise capabilities risk exacerbating North–South divides unless paired with capacity-building and access mechanisms.
  • Redistribution mechanisms (technology transfer, finance for low-income countries, shared compute pools) may be required to align global welfare incentives.
• Research and policy agenda for economists:
  • Quantify the elasticities of AI output and harms to compute, data, and energy inputs.
  • Model strategic interactions among firms and states in capacity races, including enforcement of international norms.
  • Evaluate policy instruments (taxes, caps, licensing, antitrust, public provision) on innovation, welfare, and distributional outcomes.
  • Empirically measure cross-domain feedbacks (e.g., how information harms affect political economy outcomes that, in turn, change regulation and investment patterns).

Overall, the paper reframes AI governance as a problem of managing a global commons with three tightly coupled domains. For AI economists, that implies shifting analysis toward scarce-input regulation, global coordination mechanisms, and models that internalise cross-domain externalities and distributional consequences.