Research / AI, institutions & power
What changes when intelligence becomes abundant?
Abundant analysis does not abolish the need for institutions, for physical infrastructure or for someone who answers for a decision. It moves the scarcity onto them. This work follows AI from the decisions inside a firm to the markets between machines and the relations between states.
A route through the subject
AI, institutions & power
01
Markets
Consider prediction markets and the institutional infrastructure of a machine economy.
02
Material capacity
Examine the physical systems needed to turn technical capability into sustained deployment.
03
Responsibility
Ask what algorithmic delegation does to attribution, intent, and deterrence.
Start with the physical constraints
Then continue to the manuscripts below. The illustrations offer intuition; each paper states the mathematical or empirical scope of its argument.
A manuscript to begin with: Material Deployment Power in the Age of Artificial Intelligence: How Physical Constraints Reshape International Order.
Scope of the work
These papers make institutional and political arguments. The accompanying essays isolate mechanisms in illustrative models, and their numbers are not forecasts or estimates for any firm or country.
4 papers · Suggested reading sequence
The manuscripts.
- 01
Prediction Markets as the Informational Substrate of the Machine Economy
An argument for market-generated risk signals in the institutional infrastructure of autonomous commerce.
Preprint · 2026 · Abstract, PDF & citation ↗
Read the abstract
Wuebben [2026] identified six institutional primitives required for autonomous machine-to-machine commerce and showed that three-adjudication, reputation, and liability/insurance exhibit critical gaps that constrain the machine economy's emergence. This paper argues that all three gaps share a common root: the absence of market-generated price signals for agent performance risk, deliverable quality, and failure correlation. Meanwhile, the settlement layer has matured faster than anticipated: three competing protocol standards-Stripe's Agentic Commerce Protocol, Coinbase's x402, and the Machine Payments Protocol launched by Stripe and Tempo-have entered production within six months, confirming that the execution substrate for machine-to-machine payments is commoditizing rapidly. The widening asymmetry between settlement maturity and trust-layer immaturity sharpens the need for mechanisms that produce the missing information. We propose that prediction markets-generalized beyond event contracts to encompass continuous distributional forecasting in the spirit of Cotton [2022]-can serve as a cross-cutting informational layer that produces the missing inputs for the primitives that currently lack them. We design five specific market mechanisms, each targeting a gap in the original stack: Quality Adjudication Markets for escrowed deliverables, Agent Performance Futures for bonding and reputation, Correlation Discovery Markets for insurance pricing, Identity Assurance Markets for Sybil detection, and Primitive Readiness Markets for sequencing strategy. We analyze the incentive compatibility, liquidity requirements, manipulation resistance, and regulatory classification of each design. Drawing on Cotton [2022]'s microprediction architecture which collapses identity, reputation, and bonding into a single performance-scored object we propose a unified design pattern in which prediction market participation itself generates the data on which the institutional stack depends, partially dissolving the bootstrapping problem identified in the original paper. We assess the limits of the approach, identify the residual problems that prediction markets cannot solve, and propose a phased implementation path that builds directly on the settlement infrastructure now in production.
- 02
The Institutional Plumbing for a Machine Economy: A Readiness Assessment of Building Blocks, Missing Infrastructure, and Commercial Opportunities
The identity, settlement, reputation, and liability systems that machine commerce would require.
Preprint · 2026 · Abstract, PDF & citation ↗
Read the abstract
Autonomous AI agents are approaching the capability threshold required to transact, negotiate, and settle economic obligations without human intermediation. But capability is not sufficiency. A functioning machine economy requires institutional infrastructure-identity, bonding, escrow, adjudication, settlement, reputation, and liability assignment-analogous to the legal and financial plumbing that enables human commerce. This paper audits each layer of that stack against current technological readiness, identifies the critical gaps, and evaluates where commercial opportunities lie: in the plumbing itself, or in the application layer above it. We find that of the six primitives required for a functioning machine economy, one is production-ready (settlement), two are close (identity, bonding infrastructure), and three have critical gaps (adjudication, reputation, liability/insurance). The binding constraints are on the "trust layer"-the primitives that enable agents to interact with counterparties they have no prior relationship with and no human intermediary to fall back on. We map commercial opportunities across both the infrastructure and application layers, and propose a phased strategy that captures near-term application-layer value while accumulating the data and expertise needed for plumbing-layer opportunities as the market matures.
- 03
Material Deployment Power in the Age of Artificial Intelligence: How Physical Constraints Reshape International Order
How physical systems constrain the conversion of AI capability into sustained deployment and political power.
Preprint · 2026 · Abstract, PDF & citation ↗
Read the abstract
Prevailing accounts of artificial intelligence in international relations treat AI primarily as an informational or cognitive technology, emphasizing data, algorithms, and innovation capacity as the central sources of advantage. This article challenges that view by arguing that large-scale AI deployment re-centers physical constraint as a determinant of power. Although AI expands informational capacity at unprecedented speed, its deployment depends on infrastructural systems - energy generation, transmission, conversion, cooling, and critical materials - that expand slowly, are politically mediated, and are unevenly distributed. As a result, competition over AI moves international politics into a constraint-dominated phase in which power is conditioned by the ability to convert technological potential into sustained economic and military capability.
The article advances four contributions. First, it introduces the concept of material deployment power MDP, defined as the capacity to mobilize and scale the physical systems necessary to operationalize advanced technologies under conditions of rivalry - a concept distinct from Michael Mann's institutional "infrastructural power." Second, it develops a theoretical account of how material constraints generate systematic patterns of coercion, state - firm fusion, and selective cooperation, explaining the resurgence of mercantilist policies and supply-security clubs in the AI era. Third, it demonstrates through comparative analysis and detailed process tracing - examining cases including semiconductor export controls, hyperscaler energy strategies, and rare earth processing - that variation in material deployment capacity shapes AI deployment outcomes, trade and industrial policy, and alliance behavior independently of innovation leadership. Fourth, it extends the analysis to two underexplored domains: the emergence of energy-backed digital assets as instruments of monetary sovereignty, and the domestic political consequences of capital-intensive technological transitions that concentrate returns while displacing labor.
The argument has broader implications for international order. By binding advanced capabilities to slow-moving physical systems, AI re-materializes global politics, increasing sensitivity to chokepoints, fragmenting interdependence, and reshaping the foundations of deterrence and cooperation. Understanding the geopolitics of artificial intelligence therefore requires rethinking prevailing assumptions about scalability, substitution, and the relationship between technology and power in the twenty-first century.
- 04
Algorithmic Plausible Deniability: The Structural Erosion of Deterrence in the Age of Artificial Intelligence
How algorithmic delegation can complicate attribution, intent, and deterrence.
Preprint · 2026 · Abstract, PDF & citation ↗
Read the abstract
Classical deterrence theory assumes that hostile actions can be attributed to identifiable actors whose intentions can be inferred and who can be credibly threatened with punishment. This article introduces Algorithmic Plausible Deniability APD: a structural condition in which the delegation of decisions to artificial intelligence systems generates persistent uncertainty over authorship and intent, thereby weakening deterrence even when states act rationally and prefer stability. Unlike strategic ambiguity, which actors choose, APD emerges endogenously from algorithmic delegation and persists independently of deception or bad faith. The article develops three contributions. First, it provides a theoretical account of how AI systems degrade the informational foundations of deterrence through mechanisms of delegation, opacity, and emergent behavior. Second, it identifies an Accountability Trilemma in which states cannot simultaneously achieve operational effectiveness, strategic flexibility, and clear attribution. Third, it demonstrates that APD creates asymmetric effects: hostile actions become more deniable while signals of restraint become less credible, systematically undermining escalation control. The argument has implications for cyber conflict, autonomous weapons governance, and crisis stability among technologically advanced states.