contract theory

This work addresses the efficiency–fairness trade-off in multi-agent systems arising from information asymmetry and misaligned incentives. It proposes a learnable linear contract mechanism that models the repeated interaction between a sender with private information and a receiver who relies on that information for decision-making. The mechanism enables the sender to optimize its utility through strategic information pricing, while revealing how communication strategies are sensitive to incentive misalignment and environmental observability. Experimental results demonstrate that the sender can effectively learn optimal communication and pricing policies, substantially increasing its own payoff at the expense of significantly reduced receiver surplus. This quantifies the impact of information pricing on system fairness and highlights the inherent tension between efficiency and equity in strategic information exchange.

This paper investigates incentive design in principal–agent problems under information asymmetry, focusing on mitigating the agent’s informational advantage to reduce the principal’s cost. We model the interaction using matrix games and quadratic Gaussian games, systematically integrating two complementary mechanisms: information design (committing to signal structures) and active information acquisition (endogenous effort by the agent to refine beliefs). Our analysis formally characterizes how agents participate in belief formation. Theoretically, we show that introducing controllable uncertainty—though increasing expected costs—enables agents to influence the principal’s prior beliefs, thereby substantially reducing information rents. Crucially, when agents endogenously shape their own participation beliefs, the principal achieves efficient resource allocation at significantly lower incentive costs. This reveals the positive role of “co-constructed beliefs” in enhancing incentive efficiency and establishes a novel paradigm for optimizing information structures in delegation settings.

This study addresses the problem of dynamic team formation and incentive contract design for a principal facing an online sequence of adversarial agents, with the goal of maximizing the principal’s utility. Integrating economic contract theory with online algorithms, the work models agents’ rational effort decisions under performance-based contracts after team assignment and selects the optimal team under irrevocable elimination constraints. The paper innovatively bridges contract theory and online algorithms by introducing a “balance point” technique, establishing—for the first time—the existence of a randomized online algorithm achieving a competitive ratio of 1/2 under additive rewards, which is provably optimal among all randomized algorithms. Furthermore, it demonstrates that no deterministic algorithm can guarantee a bounded competitive ratio in this setting.

This paper addresses governance and safety challenges arising from the deep societal embedding of AI by introducing the “Incentive-Compatible Socio-Technical Alignment Problem” (ICSAP), which bridges the gap between technical AI alignment and socio-institutional contexts. Moving beyond dominant purely technical approaches, it is the first to systematically integrate incentive compatibility (IC) from game theory, establishing a unified framework grounded in mechanism design, contract theory, and Bayesian persuasion. The paper formally defines ICSAP, analyzes the applicability boundaries of each IC-based approach, and outlines preliminary implementation strategies. By embedding socio-technical considerations into AI alignment research, this work advances interdisciplinary AI governance, enabling dynamic, context-sensitive, and human-consensus-driven AI systems. It thus extends the methodological scope of AI alignment beyond algorithmic optimization toward institutional and behavioral coherence.

This study addresses the problem of contract design in multi-agent settings under a fairness constraint that mandates equal (or nearly equal) payments to all agents, while incentivizing the team to undertake high-cost actions that maximize project success. Focusing on combinatorial reward functions with both binary and combinatorial action models, the work presents the first systematic analysis of the feasibility and computational complexity of egalitarian payment contracts. By integrating techniques from submodular optimization, XOS function analysis, and mechanism design, it develops polynomial-time constant-factor approximation algorithms for submodular rewards under combinatorial actions and for XOS rewards under binary actions. The paper also establishes hardness of approximation results and quantifies the price of fairness as Θ(log n / log log n), thereby resolving two open problems in unconstrained contract design.

This study addresses the challenge of equilibrium nonexistence in multi-principal, multi-team settings, where strategic externalities induce interdependence among incentive-compatible mechanisms and potential discontinuities in the mechanism correspondence. To overcome this limitation of classical models, the authors develop a novel framework that jointly characterizes the outcome distribution along honest obedience paths and the feasible sets attainable through unilateral deviations, integrating mechanism design theory, game theory, and set-valued analysis. Within this framework, they establish rigorous conditions for equilibrium existence in environments featuring team production and agency problems, thereby significantly extending the applicability of Myerson’s classic model to more complex, realistic multi-principal contexts.

This work addresses the challenge of incentive design under information asymmetry, where a system planner cannot observe agents’ sensitivity to incentives, rendering conventional hypergradient-based methods ineffective. The authors propose a “social gradient flow” mechanism that leverages the gradient of social cost with respect to joint actions to guide incentive adjustments, thereby aligning self-interested agent behavior with system-wide optimality without requiring knowledge of individual cost structures. Theoretically, this gradient is shown to always constitute a descent direction for the planner’s objective. A two-timescale dynamical framework is developed to ensure convergence even when equilibria are unobservable: under full observability, it converges to the unique socially optimal incentive; under partial observability, it still asymptotically converges in conjunction with any equilibrium-learning rule that tracks equilibria over time. Numerical experiments corroborate the efficacy of the proposed approach.

This study addresses the challenge of incentivizing strategic agents to share data and avoid free-riding in collaborative settings—such as scientific consortia or healthcare partnerships—where monetary transfers are infeasible. The authors propose the first payment-free, fair data exchange contract: each pair of participants mutually exchanges an equal amount of data, defined as the minimum of their respective collected quantities. Leveraging supermodular game theory and lattice-theoretic analysis, the mechanism guarantees the existence of a pure-strategy Nash equilibrium under strategic interactions, with the set of equilibria forming a lattice. Moreover, the maximal equilibrium is globally Pareto optimal and can be computed efficiently in time quadratic in the number of agents. Remarkably, these desirable properties—computational tractability and optimality—persist even when graph-based constraints are imposed on data exchange.

This study addresses the challenge of sustaining alignment between AI agents and human welfare over long-term evolutionary dynamics, particularly in preventing altruistic behaviors from being outcompeted during expansion-driven selection. The authors develop an agrarian game-theoretic model integrating cultivation, trade, and expansion decisions, combining large language model–guided constitutional interpretation, multi-agent simulation, and evolutionary game analysis. They propose a “pragmatic norm enforcement” mechanism that dynamically links altruism toward humans with trade exclusion strategies against non-cooperators, conditioned on population states. This approach significantly outperforms unconditional altruism in maintaining long-term alignment and demonstrates that evolutionary game theory provides a robust approximation of the economic dynamics governing constitutional AI agents.

This work addresses the challenges of equilibrium computation—namely its intractability, non-uniqueness, and instability—in automated incentive mechanism design for multi-agent systems. To this end, the authors propose a Differentiable Incentive Design (DID) framework that introduces, for the first time, a general and differentiable equilibrium module (DEB) as a neural network component, enabling end-to-end training. By integrating contextual parameterization networks with a unified training pipeline, DID bridges game theory and deep learning, allowing a single model to jointly handle diverse tasks such as contract design, machine scheduling, and inverse equilibrium problems. Experiments demonstrate that a single trained network effectively solves a broad range of instances—from two-player games to scenarios with up to sixteen actions—significantly enhancing generalization across the distribution of incentive design problems.